2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana
2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana
2 Theoretical Foundations 2.1 Disaster System Theory Disaster system theory, grounded in systems science, conceptualizes disasters as complex, dynamic entities characterized by integrality and systematicity (Shi et al., 2020). This study adopts the Environment (E) - Hazard (H) - Subject (S) model (Fig. 2a), which identifies disaster events as the synergistic outcome of interactions between natural environmental conditions, physical hazards, and the affected subjects (Wang et al., 2024b). Historically, the evaluation of disaster risk has relied upon the HEV (Hazard-Exposure-Vulnerablility) model (Fig. 2b). In this context, hazard characterizes the intrinsic properties of the climate event, such as frequency and intensity. Exposure reflects the degree to which populations and assets are situated in hazard-prone regions. Vulnerability denotes the susceptibility to damage shaped by socioeconomic fragilities (D'Ambrosio et al., 2023; Rising et al., 2022). A pivotal advancement in recent risk management scholarship is the extension of this model into the HEVR (Four-element) framework, which introduces response capacity as a critical fourth dimension (Fig. 2c) (Ayanlade et al., 2023; Simpson et al., 2021). Fig.2 Disaster system theory: (a) disaster formation model, (b) three elements of risk and (c) four elements of risk. 2.2 Disaster Risk Management Building upon the HEVR theory, this study proposes a comprehensive governance framework that facilitates a paradigm shift from static risk description to mechanism-oriented explanation (Ayanlade et al., 2023). This framework operates across two synergistic dimensions: spatial and temporal. For the spatial dimension, this study introduces a typological zoning approach to move beyond conventional risk mapping (Fig.3 a). By transitioning from basic risk assessment to the classification of regions into distinct risk-dominant types, the framework pinpoints the specific structural weaknesses of different areas, thereby enabling the formulation of targeted, precision-based interventions. Temporally, the governance of climate risk is conceptualized through a continuous disaster full-cycle management framework (Fig. 3b). This dynamic, closed-loop system spans the entire disaster timeline through four interdependent phases: Prevent and Prepare (pre-disaster), Response (mid-disaster), and Recover (post-disaster). Post-disaster recovery efforts continuously inform and enhance proactive prevention and preparedness, systematically reshaping the future risk landscape. By integrating this dual-dimensional framework with interpretable machine learning, this study contributes a scalable theoretical foundation for systematic risk governance. It transforms abstract climate variables into actionable parameters, empowering policymakers to optimize resource allocation and strengthen long-term resilience against the escalating threats of extreme heat. Fig.3 Disaster risk management framework: (a) risk assessment and typological zoning and (b) disaster full-cycle framework.
07
Hero
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
031
Nano Banana
Create a BioRender-style vector infographic. Place the panel header “研究目标” as a small caption in the upper-left corner (not a large title). Set its font size to ~80% of the main box titles, use regular weight (not bold), and keep it visually subtle. Use a clean professional layout, flat colors, thick outlines, minimal shadows, and consistent sans-serif font (Microsoft YaHei). Canvas: 16:9 or 16:10 landscape. The figure should be non-technical (no equations), focusing on goals hierarchy. All text inside boxes must be Chinese exactly as provided. Layout Split the figure into two parts: Left Part (Overall Goals Pyramid): Draw a 3-layer stacked pyramid (or three stacked rounded rectangles) labeled from top to bottom: “理论目标”, “方法目标”, “应用目标”. Use subtle distinct colors for each layer. Add a small left-side label “总体目标” above the pyramid. Right Part (Three Specific Objectives Cards): Place three numbered rounded-rectangle cards vertically aligned (G1, G2, G3). Each card contains the specific objective text. Draw thin arrows from each card to the related pyramid layer(s): G1 arrows to “理论目标” (primary) and slightly to “方法目标” (secondary) G2 arrows to “方法目标” (primary) and slightly to “理论目标” (secondary) G3 arrows to “应用目标” (primary) and slightly to “方法目标” (secondary) Bottom Bar (One-line Summary): Add a wide rounded rectangle at the bottom spanning the width, labeled “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系”. Icons (small, minimal, optional) Next to “理论目标”: a Lyapunov/analysis icon (V(x)) Next to “方法目标”: a neural network + event-trigger clock icon Next to “应用目标”: drone + robot icons Next to bottom bar: a balance scale icon labeled “性能—通信—能耗” Chinese text to place (exact) Overall Goals (left pyramid): Top layer title: “理论目标” Text: “建立一致性误差动力学描述与统一的收敛/有界性分析框架；给出误差判据、误差上界、无Zeno与IET下界等可验证结论。” Middle layer title: “方法目标” Text: “提出以DRNN为核心的学习辨识器、动态事件触发机制与分布式控制协议协同设计方法；揭示触发参数、拓扑、辨识误差与一致性性能的定性—定量关系；建立‘一致性性能—通信资源与能耗开销’权衡机制。” Bottom layer title: “应用目标” Text: “依托多无人机与多机器人平台开展仿真与实验验证；形成可推广的低通信、高可靠、可验证协同控制方法；服务无人系统集群、网络化制造单元及分布式能源等场景。” Right cards (specific objectives): Card 1 header: “具体目标1（G1）” Body: “构造显式利用触发区间信息的looped-functional型Lyapunov分析工具，降低收敛性与有界性结论的保守性；推导误差上界、无Zeno条件与IET下界；刻画IET与拓扑结构、触发参数及系统状态之间的定性—定量关系。” Card 2 header: “具体目标2（G2）” Body: “建立基于DRNN学习辨识的分布式事件触发一致性控制框架；研究学习误差、触发误差与拓扑耦合误差对闭环性能的影响；设计权值更新律与触发机制，实现对未知非线性/不确定项的在线补偿并提升鲁棒性。” Card 3 header: “具体目标3（G3）” Body: “面向模型信息不完备与参数不确定等情形，研究模型弱依赖的学习辨识–事件触发一致性控制方法；给出一致性误差收敛性与有界性条件，以及DRNN权值与内部递归状态的有界性结论；提升方法的工程可实施性与适用范围。” Bottom bar text: “预期形成：低通信、高可靠、可验证的一致性控制理论与方法体系” Negative prompt Avoid photorealism, avoid dense paragraphs, avoid tiny illegible text, avoid complex mathematical derivations, avoid cluttered decorative elements.
023
Nano Banana
Create a BioRender-style, publication-ready vector infographic titled “研究内容框架图” for a grant proposal. Use clean flat BioRender vectors, thick outlines, minimal shadows, consistent spacing, and a readable sans-serif font (Microsoft YaHei). Use a 16:10 landscape canvas (taller than 16:9). All text inside boxes must be Chinese exactly as specified. Do not include any mathematical letters, symbols, or formulas. Layout The figure has two main sections: Section A (left/center): Research Content Framework (main flowchart) A large framed panel with a top-down or left-to-right flow of four major blocks (Step 1 → Step 2 → Step 3 → Step 4). Each block is a rounded rectangle with a short title plus 2–4 bullet points. Add clear arrows between steps. Add a small triangle badge near Step 3 showing the trade-off. Section B (right side): Three embedded mini-schematics aligned vertically, each framed, with titles: “闭环控制框架（流程图）” “耦合误差示意（维恩图）” “深度递归神经网络示意（时间展开）” Use thin dashed connectors from the main Step 1–3 blocks to the corresponding mini-schematics to show correspondence. Icons (flat, minimal) Multi-agent network graph (nodes + edges), drones and mobile robots, wireless signal, clock/bell for event-triggering, sample-and-hold icon, neural network/RNN icon, Lyapunov/stability icon, and a balance scale icon (performance vs communication vs energy). Keep icons minimal and consistent. Chinese text to place in boxes (exact) Title (top center) “学习辨识—事件触发耦合下非线性多智能体系统分布式一致性控制与收敛性/有界性分析：研究内容框架图” Section A: Main research content framework (4 steps) Step 1 (Block 1) Title: “一致性误差机理刻画” Bullets: “建立统一闭环误差建模框架” “刻画学习误差、触发保持误差与拓扑耦合误差的交叉作用” “解释收敛退化、触发频繁与性能下降的成因” “覆盖无领导一致、领导跟随一致与协同跟踪场景” Step 2 (Block 2) Title: “低保守收敛性与有界性分析” Bullets: “显式利用触发区间信息构造分析工具” “建立收敛性与有界性判据并降低保守性” “推导误差上界、无有限时间无限触发条件与触发间隔下界” “刻画触发间隔与拓扑、触发参数、辨识精度的定量关系” Step 3 (Block 3) Title: “协同设计与权衡机制” Bullets: “协同设计学习辨识器、动态事件触发与分布式控制协议” “保证学习参数与内部递归状态有界” “揭示学习率、触发参数、拓扑特征与一致性性能的定量关系” “建立一致性性能—通信次数—能耗开销的可计算权衡” Add-on icon near Step 3: A small triangular trade-off badge with vertex labels (Chinese): “一致性性能 / 通信次数 / 能耗开销” Caption next to triangle: “可计算权衡” Step 4 (Block 4) Title: “仿真分析与实验验证” Bullets: “搭建含未知非线性、扰动与通信约束的仿真平台” “对比不同触发规则、拓扑与学习精度下的性能与通信开销” “在多无人机与多机器人平台开展验证” “形成可推广的低通信、高可靠、可验证方法” Section B: Three mini-schematics (right side) Mini-panel 1: “闭环控制框架（流程图）” Draw a left-to-right flowchart with rounded blocks and arrows: Blocks (in order, Chinese text exact): “非线性多智能体系统” → “局部/邻域信息获取” → “一致性误差计算” → “学习辨识器（深度递归神经网络）” → “分布式控制器” → “动态事件触发器” → “网络传输与采样保持” → Back arrow to “非线性多智能体系统” Add two dashed feedback arrows from “一致性误差计算” to: “学习辨识器（深度递归神经网络）” (label: “误差驱动更新”) “动态事件触发器” (label: “误差驱动更新”) Add small notes: Under “动态事件触发器”: “按需通信/按需更新” Near “网络传输与采样保持”: “触发保持误差” Add a small timeline icon with ticks labeled in Chinese: “触发时刻…下一次触发时刻” and label “触发间隔”. Mini-panel 2: “耦合误差示意（维恩图）” Draw a three-circle Venn diagram with semi-transparent circles: Circle labels (Chinese): “学习辨识误差” (with RNN icon) “触发保持误差” (with clock + sample-and-hold icon) “拓扑耦合误差” (with network graph icon) Pairwise overlap labels: “学习更新×非均匀更新” “异步通信×拓扑传播” “分布式辨识×邻域耦合” Center overlap (bold): “耦合项集合” Under it: “影响一致性误差演化” Arrow from center to a right-side box titled “结果表征” with bullets: “收敛退化（渐近→最终有界）” “触发频繁/触发间隔变小” “稳态误差界增大/性能下降” Mini-panel 3: “深度递归神经网络示意（时间展开）” Draw a time-unrolled recurrent network schematic along a horizontal timeline labeled in Chinese: “上一时刻 → 当前时刻 → 下一时刻” At each time slice, show stacked recurrent blocks: Input label: “本体状态与邻域信息” → Middle label: “递归记忆状态” → Output label: “未知非线性与不确定项的在线辨识输出” Connect time slices with arrows labeled: “共享参数” Add a side arrow from “一致性误差” into a small box: “参数更新（投影/正则化/学习率调度）” Then arrow into: “学习参数更新” Style constraints BioRender clean scientific infographic, no photorealism, no clutter, high readability. Strict rule: do not include any math symbols, letters, equations, or subscripts. Negative prompt: Avoid photorealistic style, avoid dense paragraphs, avoid handwritten fonts, avoid low resolution, avoid formulas, avoid math letters.
137
Nano Banana
FIGURE 1. Study overview（全宽 180mm × ~120mm，5 panels: a-e） 排布：上面一行 panel a + panel b（a 占 ~60mm，b 占 ~115mm） 下面一行 panel c + panel d + panel e（各 ~58mm） --- Panel a: "The leaderboard conflation problem" --- 画一个简化的排行榜表格，6 行，3 列：Rank | Model | F1 选这 6 个模型，代表不同 data regime 和 architecture： #1 PET-OAM-XL F1 = 0.924 [深蓝色块] OMat24+sAlex+MPtrj #2 eSEN-30M-OAM F1 = 0.888 [深蓝色块] OMat24+sAlex+MPtrj #5 MatRIS-10M-OAM F1 = 0.877 [深蓝色块] OMat24+sAlex+MPtrj #11 ORB v3 F1 = 0.860 [天蓝色块] MPtrj+Alex+OMat24 #17 eSEN-30M-MP F1 = 0.797 [teal色块] MPtrj #22 Eqnorm MPtrj F1 = 0.756 [teal色块] MPtrj 每行右侧画一个小色块表示 training data regime（用上面定义的配色）。 关键标注： 在表格右侧画两个大括号： 上面的大括号框住 #1-#5（三个深蓝色模型），旁边写： "Same data regime, 3 different architectures → F1 range: 0.877-0.924" 下面的大括号框住 #17 和 #2（eSEN-30M-MP vs eSEN-30M-OAM），旁边写： "Same architecture, different data → ΔF1 = 0.091" 在表格上方加标题："Current leaderboard: architecture or data?" 核心信息：一眼看出 (1) 同数据不同架构的差异很小，(2) 同架构不同数据的差异很大。 --- Panel b: "Our decomposition framework" --- 从左到右的信息流图，四个阶段： 阶段 1（左）："Prediction matrix" 画一个小矩形网格，标注行 = "45 models"，列 = "256,963 materials" 矩阵内填淡色渐变表示预测值 阶段 2（中左）："Factor extraction" 从矩阵引出三条线，分别指向三个小标签： "Training data" (11 exact combinations) [teal 色] "Architecture" (5 groups) [橙色] "Parameters" (continuous) [灰色] 阶段 3（中右）："Five analyses" 五个小方框竖排，用细箭头从阶段 2 连过来： ① Variance decomposition → η² comparison ② Error clustering → ARI comparison ③ Scaling laws → slope comparison ④ Collective failures → structural modes ⑤ Resource allocation → Pareto frontier 阶段 4（右）："Core finding" 一个框，里面写： "Training data η² = 0.84 Architecture η² = 0.32 Data > Architecture across all metrics and robustness checks" Training data 那行用 teal 加粗，architecture 那行用橙色。 阶段之间用水平箭头连接。 --- Panel c: "Data scaling > parameter scaling" --- 一个概念性坐标图（不是真实数据，只是示意）： x 轴："log₁₀ investment" y 轴："F1" 画两条上升虚线/趋势线： teal 线，较陡，标注 "Scale data: +0.069/decade" 橙线，较缓，标注 "Scale parameters: +0.063/decade" 右侧一个小 inset 或并排小图： ensemble 饱和曲线，x = k, y = F1 标注 "Best F1 = 0.911 @ k=6"，k>6 之后曲线平坦 --- Panel d: "Failures in familiar chemistry" --- 画一个嵌套的两层椭圆（不用太复杂）： 外层大椭圆：标注 "256,963 WBM materials"，浅灰填充 内层小椭圆（偏右上方）：标注 "66,260 collective successes"，白色填充 外层但不在内层的区域中，画一个高亮的小区块（红色或深色）： 标注 "1,882 collective failures" 加一条引出线指向旁边的文字： "NOT chemistry-OOD → familiar formulas → sparse structural support → singleton failure rate 0.173" 核心视觉信息：红色区块在大椭圆内部（familiar chemistry）， 不在大椭圆边缘（不是 OOD）。 --- Panel e: "Budget-tier recommendations" --- 一条简化的阶梯曲线，x 轴分三段："Low", "Mid", "High" y 轴：F1，大约从 0.75 到 0.90 三个点标在阶梯上： Low: Eqnorm MPtrj, F1 = 0.779 [teal 圆点] Mid: MatterSim v1 5M, F1 = 0.838 [紫色 圆点] High: eSEN-30M-OAM, F1 = 0.902 [深蓝 圆点] 每个点旁边用小字标注模型名和 training data。 点之间用向上的箭头连接，箭头旁标注 "data regime upgrade"。
011
Nano Banana