ECG-WM: A Physiology-Informed ECG World Model for Clinical Intervention Simulation

Changshui Zhang; Sen Cui; Tianling Ren; Tingting Zhu; Yue Wang; Yu Zhang; Zhikang Chen

arxiv: 2605.17580 · v1 · pith:MYTZ25CRnew · submitted 2026-05-17 · 💻 cs.AI

ECG-WM: A Physiology-Informed ECG World Model for Clinical Intervention Simulation

Zhikang Chen , Yue Wang , Sen Cui , Yu Zhang , Changshui Zhang , Tianling Ren , Tingting Zhu This is my paper

Pith reviewed 2026-05-20 12:17 UTC · model grok-4.3

classification 💻 cs.AI

keywords ECG simulationworld modelphysiological ODE priorslatent diffusionintervention simulationclinical risk assessmentuncertainty quantification

0 comments

The pith

Embedding heart-physiology equations into a diffusion model produces realistic simulated ECG traces after drug interventions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to create a simulation system that forecasts how electrocardiogram waveforms will change when a patient receives a pharmacological or other clinical intervention. It achieves this by folding known physiological rules expressed as ordinary differential equations directly into the latent space of a diffusion model through an energy-regularization term. A sympathetic reader would care because such a tool could let clinicians preview likely cardiac responses to candidate treatments on a virtual patient before any real administration occurs. The work further adds a way to gauge both average risk and its spread by exploiting the random sampling inherent in diffusion generation.

Core claim

A framework integrates physiological ordinary differential equation priors into latent diffusion dynamics via energy regularization; the resulting structural constraint produces physiologically plausible post-intervention ECG trajectories, reduces generative hallucinations, and supports an uncertainty-aware evaluation that uses diffusion stochasticity to quantify both expected clinical risk and its variability.

What carries the argument

Energy regularization that injects physiological ODE priors into latent diffusion dynamics to enforce realistic cardiac evolution.

If this is right

Post-intervention ECG trajectories can be synthesized while respecting known cardiac physiology.
Generative hallucinations that deviate from real dynamics are suppressed by the ODE constraint.
Stochastic diffusion sampling yields both an expected clinical risk value and a measure of its variability.
Risk calibration improves and generated scenarios align more closely with expert treatment preferences.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same regularization technique could be applied to simulate other time-series physiological signals such as arterial pressure or EEG under intervention.
Large ensembles of simulated patients could support in-silico screening of new compounds before first-in-human trials.
Variability estimates might help clinicians identify which patients are likely to show unusually large or small responses to a given intervention.

Load-bearing premise

Energy regularization derived from physiological ODEs will keep the generated ECG trajectories close to actual cardiac responses instead of producing only superficially believable but dynamically inaccurate outputs.

What would settle it

Compare model-generated ECG changes after a known drug dose against real pre- and post-administration recordings collected from the same patients in a prospective clinical study.

Figures

Figures reproduced from arXiv: 2605.17580 by Changshui Zhang, Sen Cui, Tianling Ren, Tingting Zhu, Yue Wang, Yu Zhang, Zhikang Chen.

**Figure 1.** Figure 1: Comparison of methods. Left: Unlike prior open-loop or static models, our world model simulates post-intervention ECGs to recommend optimal treatments. Right: Our approach significantly outperforms GPT baselines in rollout predictions. as arrhythmia classification and disease prediction, these approaches remain largely focused on discriminative or predictive objectives. They often lack the ability to chara… view at source ↗

**Figure 2.** Figure 2: The overview of our inverse-dynamics pipeline for treatment evaluation and analysis. Given [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: The overview of rollout settings used in this study. One-step rollout evaluates single-step [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗

**Figure 4.** Figure 4: Ablation study on the Drug-response ECG dataset. We compare our full model with [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

**Figure 5.** Figure 5: Sensitivity analysis of the risk-aversion coefficient [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗

**Figure 6.** Figure 6: Counterfactual drug-response simulation with the ECG World Model. Given a normal pre [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

**Figure 7.** Figure 7: Robustness of the predictive world model under varying degrees of data corruption. The [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗

**Figure 8.** Figure 8: Reconstruction under single-lead missing in an abnormal exemplar patient. Given pre-dose [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗

**Figure 9.** Figure 9: Empirical stability of the EPK mechanism under increasing prior mismatch. We com [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗

read the original abstract

Electrocardiogram (ECG)-based models have achieved strong performance in diagnostic tasks, yet they remain limited in modeling how cardiac dynamics evolve under external interventions. In particular, existing approaches focus primarily on static prediction and lack mechanisms to capture ECG variations under different pharmacological conditions. In this work, we propose an ECG World Model for action-conditioned predictive simulation of cardiac electrophysiology. Moving beyond disjoint pipelines, our framework features a principled integration of physiological ordinary differential equation (ODE) priors into latent diffusion dynamics via energy regularization. This structural constraint enables the synthesis of physiologically plausible post-intervention ECG trajectories while effectively mitigating generative hallucinations. Building on this simulation process, we introduce an uncertainty-aware evaluation strategy that leverages the stochasticity of diffusion sampling to characterize both the expected clinical risk and its variability, allowing a more reliable comparative assessment of candidate interventions. We evaluate our method across diverse settings, including controlled drug-response scenarios and real-world clinical records. Beyond standard waveform metrics, experimental results demonstrate improved risk calibration and strong alignment with expert-informed treatment preferences. These results establish our approach as a robust foundation for safe and intervention-aware clinical decision support.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The ODE-prior energy regularization in a latent diffusion setup for action-conditioned ECG simulation is a reasonable integration but the abstract supplies no equations or ablations to show it delivers real fidelity rather than just plausible waveforms.

read the letter

The main takeaway is that this work tries to build a simulation model that generates post-intervention ECG trajectories by folding physiological ODEs into a diffusion process via energy regularization, then uses the sampling noise to estimate both average risk and its spread. That combination is new enough in the ECG modeling space and directly targets the gap between static diagnostic models and dynamic intervention planning. The uncertainty-aware evaluation that treats diffusion stochasticity as a source of risk variability is a practical addition for comparing candidate treatments. They report results on drug-response cases and real clinical records with gains in calibration and expert preference alignment, which shows they are thinking about downstream use rather than just waveform quality. The approach avoids obvious circularity by bringing in external ODE priors instead of fitting everything to itself. The soft spot is the lack of visible math or controls: no derivation of the energy term, no ablation on how strongly it constrains the diffusion objective, and no numbers showing that outputs match actual post-drug ECG changes rather than just avoiding gross artifacts. The stress-test point about plausibility versus quantitative fidelity lands because the abstract positions the method as a clinical simulation tool, yet without those checks it is unclear whether the regularization dominates enough to prevent systematic deviations from real dynamics. This is for people working on physics-informed generative models or AI tools for cardiac decision support. A reader already familiar with diffusion and ODEs in time series would get the most out of it and could assess whether the integration holds up. I would send it to peer review; the framing and evaluation idea are solid enough to deserve referee time even if the current evidence is thin.

Referee Report

2 major / 2 minor

Summary. The paper proposes ECG-WM, a physiology-informed world model for action-conditioned simulation of post-intervention ECG trajectories. It integrates physiological ODE priors into latent diffusion dynamics through energy regularization to enforce plausibility and reduce hallucinations, then introduces an uncertainty-aware evaluation that exploits diffusion stochasticity to estimate expected clinical risk and variability. Experiments on controlled drug-response scenarios and real clinical records report improved risk calibration and alignment with expert treatment preferences.

Significance. If the energy-regularized integration demonstrably enforces quantitative fidelity to real post-intervention cardiac dynamics (rather than merely ODE-consistent smoothness), the framework could offer a valuable simulation tool for safe intervention assessment and uncertainty-quantified risk comparison in clinical decision support. The uncertainty-aware strategy is a constructive addition if the underlying trajectories are verifiably accurate.

major comments (2)

[§3.2] §3.2 (Energy-Regularized Latent Diffusion): The manuscript describes the addition of an ODE-based energy term but provides no derivation, weighting schedule, or ablation demonstrating that this term dominates the diffusion objective sufficiently to enforce quantitative match to empirical post-intervention trajectories rather than only penalizing gross inconsistencies. This is load-bearing for the claim of mitigated hallucinations and accurate simulation.
[§4.3] §4.3 (Risk Calibration Results): The reported improvements in risk calibration and expert alignment are presented without direct comparison to baselines that also incorporate physiological priors or without statistical tests showing that gains arise specifically from the ODE energy regularization; waveform metrics alone do not establish fidelity to real drug-response dynamics.

minor comments (2)

[Abstract and §3] The abstract and method sections would benefit from an explicit equation for the combined loss (diffusion + energy) and a table listing all hyperparameters including the regularization coefficient.
[Figures 4-6] Generated ECG figures should include side-by-side ground-truth post-intervention recordings with quantitative error bands to allow visual and numerical assessment of trajectory fidelity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We address each major comment below and will revise the manuscript accordingly to provide the requested derivations, ablations, comparisons, and statistical analyses.

read point-by-point responses

Referee: [§3.2] §3.2 (Energy-Regularized Latent Diffusion): The manuscript describes the addition of an ODE-based energy term but provides no derivation, weighting schedule, or ablation demonstrating that this term dominates the diffusion objective sufficiently to enforce quantitative match to empirical post-intervention trajectories rather than only penalizing gross inconsistencies. This is load-bearing for the claim of mitigated hallucinations and accurate simulation.

Authors: We agree that the current presentation of the energy regularization requires expansion to fully support the claims. In the revised manuscript we will add a new subsection in §3.2 that derives the ODE-based energy term directly from the physiological model (including the explicit form of the energy function and its gradient), specifies the time-dependent weighting schedule λ(t) used to balance it against the diffusion loss, and reports a controlled ablation that measures quantitative fidelity (MSE and physiological inconsistency scores) on held-out post-intervention ECG segments with and without the energy term. These additions will demonstrate that the regularization improves alignment with empirical trajectories beyond smoothness penalties. revision: yes
Referee: [§4.3] §4.3 (Risk Calibration Results): The reported improvements in risk calibration and expert alignment are presented without direct comparison to baselines that also incorporate physiological priors or without statistical tests showing that gains arise specifically from the ODE energy regularization; waveform metrics alone do not establish fidelity to real drug-response dynamics.

Authors: We acknowledge the need for more targeted evidence. The revised §4.3 will include new experiments comparing ECG-WM against baselines that also embed physiological priors (e.g., ODE-constrained latent diffusion variants and physics-informed generative models). We will add statistical significance tests (paired t-tests and Wilcoxon signed-rank tests with reported p-values) on the risk-calibration and expert-alignment metrics. In addition, we will report direct fidelity measures to drug-response dynamics, such as predicted versus observed changes in QTc interval and ST-segment deviation under specific interventions, using the available clinical subsets. These revisions will isolate the contribution of the ODE energy term. revision: yes

Circularity Check

0 steps flagged

No circularity detected; external physiological priors added to diffusion model.

full rationale

The paper describes a framework that integrates physiological ODE priors into latent diffusion dynamics through energy regularization as an external structural constraint. This is not derived from or equivalent to the model's own fitted outputs or self-citations; it is presented as importing independent physiological knowledge to constrain generation. No equations, predictions, or load-bearing steps in the abstract or described claims reduce by construction to the inputs themselves. The method is evaluated on external drug-response scenarios and clinical records, keeping the derivation self-contained against benchmarks rather than self-referential.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that standard cardiac electrophysiology ODEs remain valid under pharmacological interventions and can be effectively imposed as soft constraints on a diffusion model. No free parameters or invented entities are explicitly named in the abstract.

axioms (1)

domain assumption Physiological ODEs accurately describe cardiac electrophysiology changes under external interventions
Invoked when the paper states that ODE priors are integrated to enforce physiological plausibility.

pith-pipeline@v0.9.0 · 5738 in / 1331 out tokens · 54432 ms · 2026-05-20T12:17:43.260510+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

principled integration of physiological ordinary differential equation (ODE) priors into latent diffusion dynamics via energy regularization... E_ϕ(ẑ₀) = ∥ẑ₀ − Π_ϕ(eEPK)∥²₂
IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

McSharry dynamical system... dy(θ)/dθ = −∑ α_i Δθ_i exp(−Δθ_i² / 2b_i²) − (y(θ) − y₀)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · 8 internal anchors

[1]

GPT-4 Technical Report

Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774,

work page internal anchor Pith review Pith/arXiv arXiv
[2]

From generative engines to actionable simulators: The imperative of physical grounding in world models.arXiv preprint arXiv:2601.15533,

Zhikang Chen and Tingting Zhu. From generative engines to actionable simulators: The imperative of physical grounding in world models.arXiv preprint arXiv:2601.15533,

work page arXiv
[3]

ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools

Team Glm, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. Chatglm: A family of large language models from glm-130b to glm-4 all tools.arXiv preprint arXiv:2406.12793,

work page internal anchor Pith review Pith/arXiv arXiv
[4]

Version 1.0

doi: 10.13026/4nqg-sb35. URLhttps://doi.org/10.13026/4nqg-sb35. Version 1.0. Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. InFirst conference on language modeling,

work page doi:10.13026/4nqg-sb35
[5]

Ctrl-World: A Controllable Generative World Model for Robot Manipulation

10 Yanjiang Guo, Lucy Xiaoyang Shi, Jianyu Chen, and Chelsea Finn. Ctrl-world: A controllable generative world model for robot manipulation.arXiv preprint arXiv:2510.10125,

work page internal anchor Pith review Pith/arXiv arXiv
[6]

World Models

David Ha and J¨urgen Schmidhuber. World models.arXiv preprint arXiv:1803.10122, 2(3),

work page internal anchor Pith review Pith/arXiv arXiv
[7]

GPT-4o System Card

Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Os- trow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276,

work page internal anchor Pith review Pith/arXiv arXiv
[8]

Pin-wm: Learning physics-informed world models for non-prehensile manipulation.arXiv preprint arXiv:2504.16693, 2025

Chunyuan Li, Cliff Wong, Sheng Zhang, Naoto Usuyama, Haotian Liu, Jianwei Yang, Tristan Naumann, Hoifung Poon, and Jianfeng Gao. Llava-med: Training a large language-and-vision assistant for biomedicine in one day.Advances in Neural Information Processing Systems, 36: 28541–28564, 2023a. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Boots...

work page arXiv
[9]

Tts-cgan: A transformer time-series conditional gan for biosignal data augmentation.arXiv preprint arXiv:2206.13676,

Xiaomin Li, Anne Hee Hiong Ngu, and Vangelis Metsis. Tts-cgan: A transformer time-series conditional gan for biosignal data augmentation.arXiv preprint arXiv:2206.13676,

work page arXiv
[10]

MedGemma Technical Report

Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, C´ıan Hughes, Charles Lau, et al. Medgemma technical report. arXiv preprint arXiv:2507.05201,

work page internal anchor Pith review Pith/arXiv arXiv
[11]

Generation of drug-induced cardiac reactions towards virtual clinical trials.arXiv preprint arXiv:2502.07297,

Qian Shao, Bang Du, Zepeng Li, Qiyuan Chen, Hongxia Xu, Jimeng Sun, Jian Wu, and Jintai Chen. Generation of drug-induced cardiac reactions towards virtual clinical trials.arXiv preprint arXiv:2502.07297,

work page arXiv
[12]

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, and Daya Guo. DeepSeekMath: Pushing the limits of mathemat- ical reasoning in open language models.arXiv preprint arXiv:2402.03300,

work page internal anchor Pith review Pith/arXiv arXiv
[13]

Embodiedreamer: Advancing real2sim2real transfer for policy training via embodied world modeling.arXiv preprint arXiv:2507.05198,

Boyuan Wang, Xinpan Meng, Xiaofeng Wang, Zheng Zhu, Angen Ye, Yang Wang, Zhiqin Yang, Chaojun Ni, Guan Huang, and Xingang Wang. Embodiedreamer: Advancing real2sim2real transfer for policy training via embodied world modeling.arXiv preprint arXiv:2507.05198,

work page arXiv
[14]

Vcworld: A biological world model for virtual cell simulation.arXiv preprint arXiv:2512.00306,

Zhijian Wei, Runze Ma, Zichen Wang, Zhongmin Li, Shuotong Song, and Shuangjia Zheng. Vcworld: A biological world model for virtual cell simulation.arXiv preprint arXiv:2512.00306,

work page arXiv
[15]

Ecg-moe: Mixture-of-expert electrocardiogram foundation model.arXiv preprint arXiv:2603.04589,

Yuhao Xu, Xiaoda Wang, Yi Wu, Wei Jin, Xiao Hu, and Carl Yang. Ecg-moe: Mixture-of-expert electrocardiogram foundation model.arXiv preprint arXiv:2603.04589,

work page arXiv
[16]

Medical world model: Generative simulation of tumor evolution for treatment planning.arXiv preprint arXiv:2506.02327,

Yijun Yang, Zhao-Yang Wang, Qiuping Liu, Shuwen Sun, Kang Wang, Rama Chellappa, Zongwei Zhou, Alan Yuille, Lei Zhu, Yu-Dong Zhang, et al. Medical world model: Generative simulation of tumor evolution for treatment planning.arXiv preprint arXiv:2506.02327,

work page arXiv
[17]

3D-VLA: A 3D Vision-Language-Action Generative World Model

Haoyu Zhen, Xiaowen Qiu, Peihao Chen, Jincheng Yang, Xin Yan, Yilun Du, Yining Hong, and Chuang Gan. 3d-vla: A 3d vision-language-action generative world model.arXiv preprint arXiv:2403.09631,

work page internal anchor Pith review Pith/arXiv arXiv
[18]

Hermes: A unified self-driving world model for simultaneous 3d scene understanding and generation

Pan Zhou, Zhao Yang, Yiming Hao, Fangfang Fan, Wenlang Zhao, Ziyu Wang, Qiuju Deng, Yongchen Hao, Na Yang, Lizhen Han, et al. A hybrid algorithm-based ecg risk prediction model for cardiovascular disease.European Heart Journal-Digital Health, 6(3):466–475, 2025a. Xin Zhou, Dingkang Liang, Sifan Tu, Xiwu Chen, Yikang Ding, Dingyuan Zhang, Feiyang Tan, Heng...

work page arXiv 2023

[1] [1]

GPT-4 Technical Report

Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774,

work page internal anchor Pith review Pith/arXiv arXiv

[2] [2]

From generative engines to actionable simulators: The imperative of physical grounding in world models.arXiv preprint arXiv:2601.15533,

Zhikang Chen and Tingting Zhu. From generative engines to actionable simulators: The imperative of physical grounding in world models.arXiv preprint arXiv:2601.15533,

work page arXiv

[3] [3]

ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools

Team Glm, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. Chatglm: A family of large language models from glm-130b to glm-4 all tools.arXiv preprint arXiv:2406.12793,

work page internal anchor Pith review Pith/arXiv arXiv

[4] [4]

Version 1.0

doi: 10.13026/4nqg-sb35. URLhttps://doi.org/10.13026/4nqg-sb35. Version 1.0. Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. InFirst conference on language modeling,

work page doi:10.13026/4nqg-sb35

[5] [5]

Ctrl-World: A Controllable Generative World Model for Robot Manipulation

10 Yanjiang Guo, Lucy Xiaoyang Shi, Jianyu Chen, and Chelsea Finn. Ctrl-world: A controllable generative world model for robot manipulation.arXiv preprint arXiv:2510.10125,

work page internal anchor Pith review Pith/arXiv arXiv

[6] [6]

World Models

David Ha and J¨urgen Schmidhuber. World models.arXiv preprint arXiv:1803.10122, 2(3),

work page internal anchor Pith review Pith/arXiv arXiv

[7] [7]

GPT-4o System Card

Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Os- trow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276,

work page internal anchor Pith review Pith/arXiv arXiv

[8] [8]

Pin-wm: Learning physics-informed world models for non-prehensile manipulation.arXiv preprint arXiv:2504.16693, 2025

Chunyuan Li, Cliff Wong, Sheng Zhang, Naoto Usuyama, Haotian Liu, Jianwei Yang, Tristan Naumann, Hoifung Poon, and Jianfeng Gao. Llava-med: Training a large language-and-vision assistant for biomedicine in one day.Advances in Neural Information Processing Systems, 36: 28541–28564, 2023a. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Boots...

work page arXiv

[9] [9]

Tts-cgan: A transformer time-series conditional gan for biosignal data augmentation.arXiv preprint arXiv:2206.13676,

Xiaomin Li, Anne Hee Hiong Ngu, and Vangelis Metsis. Tts-cgan: A transformer time-series conditional gan for biosignal data augmentation.arXiv preprint arXiv:2206.13676,

work page arXiv

[10] [10]

MedGemma Technical Report

Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, C´ıan Hughes, Charles Lau, et al. Medgemma technical report. arXiv preprint arXiv:2507.05201,

work page internal anchor Pith review Pith/arXiv arXiv

[11] [11]

Generation of drug-induced cardiac reactions towards virtual clinical trials.arXiv preprint arXiv:2502.07297,

Qian Shao, Bang Du, Zepeng Li, Qiyuan Chen, Hongxia Xu, Jimeng Sun, Jian Wu, and Jintai Chen. Generation of drug-induced cardiac reactions towards virtual clinical trials.arXiv preprint arXiv:2502.07297,

work page arXiv

[12] [12]

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, and Daya Guo. DeepSeekMath: Pushing the limits of mathemat- ical reasoning in open language models.arXiv preprint arXiv:2402.03300,

work page internal anchor Pith review Pith/arXiv arXiv

[13] [13]

Embodiedreamer: Advancing real2sim2real transfer for policy training via embodied world modeling.arXiv preprint arXiv:2507.05198,

Boyuan Wang, Xinpan Meng, Xiaofeng Wang, Zheng Zhu, Angen Ye, Yang Wang, Zhiqin Yang, Chaojun Ni, Guan Huang, and Xingang Wang. Embodiedreamer: Advancing real2sim2real transfer for policy training via embodied world modeling.arXiv preprint arXiv:2507.05198,

work page arXiv

[14] [14]

Vcworld: A biological world model for virtual cell simulation.arXiv preprint arXiv:2512.00306,

Zhijian Wei, Runze Ma, Zichen Wang, Zhongmin Li, Shuotong Song, and Shuangjia Zheng. Vcworld: A biological world model for virtual cell simulation.arXiv preprint arXiv:2512.00306,

work page arXiv

[15] [15]

Ecg-moe: Mixture-of-expert electrocardiogram foundation model.arXiv preprint arXiv:2603.04589,

Yuhao Xu, Xiaoda Wang, Yi Wu, Wei Jin, Xiao Hu, and Carl Yang. Ecg-moe: Mixture-of-expert electrocardiogram foundation model.arXiv preprint arXiv:2603.04589,

work page arXiv

[16] [16]

Medical world model: Generative simulation of tumor evolution for treatment planning.arXiv preprint arXiv:2506.02327,

Yijun Yang, Zhao-Yang Wang, Qiuping Liu, Shuwen Sun, Kang Wang, Rama Chellappa, Zongwei Zhou, Alan Yuille, Lei Zhu, Yu-Dong Zhang, et al. Medical world model: Generative simulation of tumor evolution for treatment planning.arXiv preprint arXiv:2506.02327,

work page arXiv

[17] [17]

3D-VLA: A 3D Vision-Language-Action Generative World Model

Haoyu Zhen, Xiaowen Qiu, Peihao Chen, Jincheng Yang, Xin Yan, Yilun Du, Yining Hong, and Chuang Gan. 3d-vla: A 3d vision-language-action generative world model.arXiv preprint arXiv:2403.09631,

work page internal anchor Pith review Pith/arXiv arXiv

[18] [18]

Hermes: A unified self-driving world model for simultaneous 3d scene understanding and generation

Pan Zhou, Zhao Yang, Yiming Hao, Fangfang Fan, Wenlang Zhao, Ziyu Wang, Qiuju Deng, Yongchen Hao, Na Yang, Lizhen Han, et al. A hybrid algorithm-based ecg risk prediction model for cardiovascular disease.European Heart Journal-Digital Health, 6(3):466–475, 2025a. Xin Zhou, Dingkang Liang, Sifan Tu, Xiwu Chen, Yikang Ding, Dingyuan Zhang, Feiyang Tan, Heng...

work page arXiv 2023