The reviewed record of science sign in
Pith

arxiv: 2604.26283 · v4 · pith:IUBQOMLS · submitted 2026-04-29 · cs.CV · cs.AI

MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution

Chunzheng Zhu , Jiaqi Zeng , Junyu Jiang , Jianxin Lin , Yijun Wang This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-04 01:32 UTCgrok-4.3pith:IUBQOMLSrecord.jsonopen to challenge →

classification cs.CV cs.AI
keywords medical VLMslatent memory evolutiondiagnostic accuracycausal counterfactual refinementintrinsic memory transitionchain-of-thoughtclinical intuition
0
0 comments X

The pith

MedSynapse-V evolves latent diagnostic memories inside medical vision-language models to internalize clinical intuition and outperform chain-of-thought methods in diagnostic accuracy.

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

The paper identifies a misalignment in medical VLMs stemming from discrete tokenization, which causes loss of long-range information and case-specific expertise. It proposes MedSynapse-V to address this by dynamically synthesizing implicit diagnostic memories in the model's hidden states. The framework uses a Meta Query to capture anatomical priors, Causal Counterfactual Refinement to enforce causal alignment via reinforcement learning, and Intrinsic Memory Transition to embed teacher patterns into the student model. If successful, this would allow models to invoke expert-like memory without external prompting, improving reliability in high-stakes diagnosis. The empirical results show gains over state-of-the-art approaches.

Core claim

The central claim is that by transferring external expertise into endogenous parameters through latent memory evolution, MedSynapse-V significantly outperforms existing state-of-the-art methods, particularly chain-of-thought paradigms, in diagnostic accuracy across multiple datasets.

What carries the argument

Latent diagnostic memory evolution, implemented via Meta Query for Prior Memorization, Causal Counterfactual Refinement using counterfactual rewards from feature masking, and Intrinsic Memory Transition through full-vocabulary divergence alignment.

If this is right

  • Models can prune redundant memories and align representations more closely with diagnostic logic.
  • External clinical expertise becomes internalized, reducing dependence on explicit chain-of-thought reasoning.
  • Performance improves on various medical imaging datasets for diagnostic tasks.
  • The dual-branch paradigm allows privileged training that transfers patterns autonomously.

Where Pith is reading between the lines

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

  • If the mechanisms work as described, similar memory evolution could be applied to non-medical VLMs for other expert domains like legal or scientific reasoning.
  • The approach might enable more efficient models by embedding knowledge in parameters rather than relying on long contexts.
  • Testing on diverse patient populations could reveal whether the internalized memories generalize beyond the training distributions.
  • Integration with real-time clinical workflows might be facilitated if the memory states prove interpretable.

Load-bearing premise

That the Meta Query, CCR, and IMT mechanisms can directly simulate clinical diagnostic logic without introducing new artifacts or failing to capture the full range of expert intuition.

What would settle it

Compare diagnostic accuracy of the base VLM against the full MedSynapse-V on a new unseen medical dataset; if the performance gap disappears or reverses, the claim of effective memory evolution would be falsified.

Figures

Figures reproduced from arXiv: 2604.26283 by Chunzheng Zhu, Jianxin Lin, Jiaqi Zeng, Junyu Jiang, Yijun Wang.

Figure 1
Figure 1. Figure 1: Existing medical VLMs suffer from coarse symbolic granularity and long-range information dissipation in discrete reasoning. MedSynapse-V addresses this by evolving diagnostic implicit memory in latent space via anatomical prior condensation, causal counterfactual refinement, and autonomous latent memory internalization. that enables near-instantaneous pattern recognition against accumulated case knowledge … view at source ↗
Figure 2
Figure 2. Figure 2: Stages I and II of MedSynapse-V. The hook features from an encoder are con￾densed into diagnostic implicit memory via learnable meta-query probes and injected into the VLM hidden stream. The memory is then refined through RL with composite rewards, ensuring causal alignment between memory and clinical decision logic. chain of diagnostic memory: Fana Meta Query −−−−−−−−−→ M CCR −−−−→ M⋆ IMT −−−−→ Mauto, whe… view at source ↗
Figure 3
Figure 3. Figure 3: Intrinsic Memory Transition (IMT) is achieved via Jensen–Shannon divergence alignment between the teacher (π +, conditioned on encoder-derived Mpri) and student (π −, conditioned on Mauto) branches. Gradients propagate solely to Aψ, enabling complete removal of the anatomical encoder at inference with negligible overhead. Privileged Branch and Autonomous Branch. The teacher branch (priv￾ileged) retains the… view at source ↗
Figure 4
Figure 4. Figure 4: Effect of diagnostic probe count N. Performance peaks around N=16 across benchmarks; further increasing N dilutes diagnostically relevant signals view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison across CT, MRI, and Ultrasound cases. MedSynapse￾V produces concise, correct diagnoses, while Med-R1 and MMedExpert-R1 generate verbose CoT with hallucinated findings (red) leading to misdiagnoses. full pipeline (Avg 67.7) confirms non-redundant contributions: MQPM grounds semantics, CCR refines via exploration, IMT compresses into an autonomous pathway. (ii) Reward design. rcausal i… view at source ↗
Figure 6
Figure 6. Figure 6: Accuracy–latency trade-off across compared VLM categories. 0 250 500 750 1000 1250 1500 1750 2000 Training Steps 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Average Reward exploration dip Full (w/ rcausal) w/o rcausal view at source ↗
Figure 7
Figure 7. Figure 7: The RL training reward dy￾namics with and without rcausal. Performance–efficiency trade-off. As shown in view at source ↗
Figure 8
Figure 8. Figure 8: t-SNE visualization of implicit memory Mauto after CCR. (a) Eight imaging modalities form well-separated clusters with clinically coherent proximity. (b, c) Within CT and Pathology, disease subtypes further segregate into distinct regions. tinguish memory-dependent from shortcut trajectories; without causal pressure the model bypasses M entirely, treating injected memory as inert padding. Latent space structure view at source ↗
Figure 9
Figure 9. Figure 9: Detailed architecture of the Diagnos￾tic Memory Sampler Pϕ. The frozen anatomi￾cal encoder Eana extracts spatial features F ∈ R Hf ×Wf ×df , which are flattened into a token se￾quence and used as key–value pairs for the learnable meta-query probes Q0. Through L layers of self￾attention, feed-forward processing, cross-attention, and a final linear projection (df → dh), the module produces N compact implicit… view at source ↗
Figure 10
Figure 10. Figure 10: Training dynamics across three stages: (a-c) Stage II reward optimization and gradient stabilization via causal refinement; (d) Stage I NTP loss convergence; (e) Stage II policy-KL evolution; (f) Stage III distillation fidelity and output agreement. 6 Training Dynamics Analysis view at source ↗
Figure 11
Figure 11. Figure 11: Causal intervention visualization on fundus (left group) and dermoscopy (right group). Each group: original image, MedSAM3 region mask B, and post-CCR memory attention map. After refinement, memory attention concentrates on diagnos￾tically critical structures while suppressing background. 8.2 Visualization of Causal Counterfactual Intervention view at source ↗
Figure 12
Figure 12. Figure 12: Memory evolution across training stages. view at source ↗
Figure 13
Figure 13. Figure 13: Qualitative comparison across Chest X-ray, Pathology, and Head CT cases. MedSynapse-V produces concise, correct diagnoses (∼38–43 tokens), while other meth￾ods generate verbose CoT (∼195–215 tokens) with hallucinated findings (red). 9.2 Failure Case Analysis CT MRI X-ray Dermoscopy Fundus OCT Pathology Utrasound 0 20 40 60 80 100 Training sample (%) 70 60 40 20 0 78% ACC 52% ACC Single Lesion Multi-lesion… view at source ↗
Figure 14
Figure 14. Figure 14: Three representative challenging modes view at source ↗
Figure 15
Figure 15. Figure 15: Prompt template for closed-ended multi-choice VQA (VQA-RAD, SLAKE, PathVQA, PMC-VQA, MMMU*, MedXpertQA-MM, GMAI-MMBench). The number of options varies by dataset (2–5); the template adapts accordingly. System: You are a helpful medical assistant. Provide a concise answer to the question. User: <image> {question} Answer the question using a single word or phrase. Assistant view at source ↗
Figure 16
Figure 16. Figure 16: Prompt template. Notably, Mauto is autonomously generated and injected in the hidden stream without altering the text prompt. the y-axis is binary diagnostic correctness (1=correct, 0=incorrect; vertical jit￾ter applied for visibility). While high-confidence predictions are predominantly correct, a notable cluster at conf < 0.3 with correctness= 0 reveals that border￾line cases (e.g., benign vs. dysplasti… view at source ↗
read the original abstract

High-precision medical diagnosis relies not only on static imaging features but also on the implicit diagnostic memory experts instantly invoke during image interpretation. We pinpoint a fundamental cognitive misalignment in medical VLMs caused by discrete tokenization, leading to quantization loss, long-range information dissipation, and missing case-adaptive expertise. To bridge this gap, we propose ours, a framework for latent diagnostic memory evolution that simulates the experiential invocation of clinicians by dynamically synthesizing implicit diagnostic memories within the model's hidden stream. Specifically, it begins with a Meta Query for Prior Memorization mechanism, where learnable probes retrieve structured priors from an anatomical prior encoder to generate condensed implicit memories. To ensure clinical fidelity, we introduce Causal Counterfactual Refinement (CCR), which leverages reinforcement learning and counterfactual rewards derived from region-level feature masking to quantify the causal contribution of each memory, thereby pruning redundancies and aligning latent representations with diagnostic logic. This evolutionary process culminates in Intrinsic Memory Transition (IMT), a privileged-autonomous dual-branch paradigm that internalizes teacher-branch diagnostic patterns into the student-branch via full-vocabulary divergence alignment. Comprehensive empirical evaluations across multiple datasets demonstrate that ours, by transferring external expertise into endogenous parameters, significantly outperforms existing state-of-the-art methods, particularly chain-of-thought paradigms, in diagnostic accuracy. The code is available at https://github.com/zhcz328/MedSynapse-V.

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.

Referee Report

0 major / 1 minor

Summary. The manuscript introduces MedSynapse-V, a framework for medical vision-language models that addresses misalignment from discrete tokenization via three mechanisms: Meta Query for Prior Memorization (learnable probes retrieving priors from an anatomical encoder), Causal Counterfactual Refinement (CCR) using RL with counterfactual rewards from region-level feature masking to quantify causal contributions, and Intrinsic Memory Transition (IMT) via a dual-branch paradigm aligning teacher and student branches through full-vocabulary divergence. It claims that transferring external expertise into endogenous parameters yields significant outperformance over SOTA methods, especially chain-of-thought, in diagnostic accuracy across multiple datasets, with code released.

Significance. If the empirical claims and ablations hold under independent validation, the work could advance medical VLMs by internalizing case-adaptive diagnostic priors without external prompting. The code release is a strength for reproducibility.

minor comments (1)
  1. The abstract states comprehensive evaluations but provides no dataset names, metrics, or statistical details; the full manuscript should include these in §4 or Table 1 to support the outperformance claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their careful summary of MedSynapse-V and for recognizing the potential impact if the empirical claims hold, as well as the value of the code release. No specific major comments were listed in the report, so we have no point-by-point rebuttals to provide at this time. We remain available to supply additional experimental details, ablations, or clarifications should the editor or referee request them.

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained

full rationale

The abstract and framework description introduce Meta Query for prior memorization, CCR via RL with counterfactual rewards from feature masking, and IMT via divergence alignment as distinct mechanisms to internalize diagnostic patterns. No equations, self-citations, or derivations are shown that reduce a claimed prediction or result to its own fitted inputs or prior self-work by construction. The outperformance claim is presented as arising from external evaluations across datasets rather than internal redefinition. This matches the default expectation of no circularity when the central argument remains independent of the listed patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no specific free parameters, axioms, or invented entities can be extracted or audited from the provided text.

pith-pipeline@v0.9.1-grok · 5784 in / 1094 out tokens · 24281 ms · 2026-07-04T01:32:57.908033+00:00 · methodology

Review history (4 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. CoLT: Teaching Multi-Modal Models to Think with Chain of Latent Thoughts

    cs.CV 2026-06 unverdicted novelty 6.0

    CoLT replaces text-based chain-of-thought in MLLMs with 3-step latent thought chains supervised by a removable external decoder in forward and backward modes, yielding 10.1x faster inference on eight benchmarks.

  2. TopoAgent: An Agentic Framework for Automated Topology Learning in Medical Imaging

    cs.CV 2026-06 unverdicted novelty 6.0

    Proposes TopoAgent, an LLM agent framework that automatically selects and configures topological descriptors from persistent homology for medical image analysis without task-specific training.

  3. Deformba: Vision State Space Model with Adaptive State Fusion

    cs.CV 2026-05 unverdicted novelty 6.0

    Deformba introduces context-adaptive state fusion to vision SSMs for better spatial augmentation and cross-stream interactions, showing strong results on 2D classification/detection/segmentation and 3D BEV perception ...

Reference graph

Works this paper leans on

161 extracted references · 161 canonical work pages · cited by 3 Pith papers · 36 internal anchors

  1. [1]

    arXiv preprint arXiv.2407.15621

    Arasteh, S.T., Lotfinia, M., Bressem, K., Siepmann, R., Adams, L., Ferber, D., Kuhl, C., Kather, J.N., Nebelung, S., Truhn, D.: Radiorag: factual large language models for enhanced diagnostics in radiology using online retrieval augmented generation 2024. arXiv preprint arXiv.2407.15621

    work page arXiv 2024

  2. [2]

    Bai, S., Cai, Y., Chen, R., Chen, K., Chen, X., Cheng, Z., Deng, L., Ding, W., Gao,C.,Ge,C.,etal.:Qwen3-vltechnicalreport.arXivpreprintarXiv:2511.21631 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Cognitive research: principles and implications4(1), 7 (2019)

    Brunyé, T.T., Drew, T., Weaver, D.L., Elmore, J.G.: A review of eye tracking for understanding and improving diagnostic interpretation. Cognitive research: principles and implications4(1), 7 (2019)

    work page 2019

  4. [4]

    arXiv preprint arXiv:2510.12603 , year=

    Chen, C., Ma, Z., Li, Y., Hu, Y., Wei, Y., Li, W., Nie, L.: Reasoning in the dark: Interleaved vision-text reasoning in latent space. arXiv preprint arXiv:2510.12603 (2025)

    work page arXiv 2025

  5. [5]

    Chen, J., Jiang, Y., Yang, D., Li, M., Wei, J., Qian, Z., Zhang, L.: Can llms’ tuningmethodsworkinmedicalmultimodaldomain?In:InternationalConference on Medical Image Computing and Computer-Assisted Intervention. pp. 112–122. Springer (2024)

    work page 2024

  6. [6]

    Detecting and Evaluating Medical Hallucinations in Large Vision Language Models

    Chen, J., Yang, D., Wu, T., Jiang, Y., Hou, X., Li, M., Wang, S., Xiao, D., Li, K., Zhang, L.: Detecting and evaluating medical hallucinations in large vision language models. arXiv preprint arXiv:2406.10185 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  7. [7]

    In: Proceedings of the 2024 conference on empirical methods in natural language processing

    Chen, J., Gui, C., Ouyang, R., Gao, A., Chen, S., Chen, G.H., Wang, X., Cai, Z., Ji, K., Wan, X., et al.: Towards injecting medical visual knowledge into multi- modal llms at scale. In: Proceedings of the 2024 conference on empirical methods in natural language processing. pp. 7346–7370 (2024)

    work page 2024

  8. [8]

    arXiv preprint arXiv:2510.10052 (2025)

    Chen, K., Rui, S., Jiang, Y., Wu, J., Zheng, Q., Song, C., Wang, X., Zhou, M., Liu, M.: Think twice to see more: Iterative visual reasoning in medical vlms. arXiv preprint arXiv:2510.10052 (2025)

    work page arXiv 2025

  9. [9]

    Chen, W., Wu, L., Hu, Y., Li, Z., Cheng, Z., Qian, Y., Zhu, L., Hu, Z., Liang, L., Tang, Q., Liu, Z., Yang, H.: Autoneural: Co-designing vision-language models for npu inference (2025),https://arxiv.org/abs/2512.02924 Medical Latent Memory Evolution 29

    work page arXiv 2025

  10. [10]

    IEEE Transactions on Artificial Intelligence5(4), 1443– 1457 (2022)

    Chen, Z., Silvestri, F., Tolomei, G., Wang, J., Zhu, H., Ahn, H.: Explain the ex- plainer: Interpreting model-agnostic counterfactual explanations of a deep rein- forcement learning agent. IEEE Transactions on Artificial Intelligence5(4), 1443– 1457 (2022)

    work page 2022

  11. [11]

    SAM-Med2D

    Cheng, J., Ye, J., Deng, Z., Chen, J., Li, T., Wang, H., Su, Y., Huang, Z., Chen, J., Jiang, L., et al.: Sam-med2d. arXiv preprint arXiv:2308.16184 (2023)

    work page Pith review arXiv 2023

  12. [12]

    In: Proceedings of the 2026 3rd In- ternational Conference on Applied Economics, Management Science and Social Development (AEMSS 2026)

    Cheng, K., Qi, X., Cheng, Z., Lai, L., Liu, X.: Regime-dependent volatility dy- namics: Evidence from time-series analysis. In: Proceedings of the 2026 3rd In- ternational Conference on Applied Economics, Management Science and Social Development (AEMSS 2026). pp. 179–189. Atlantis Press (2026)

    work page 2026

  13. [13]

    From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step

    Deng, Y., Choi, Y., Shieber, S.: From explicit cot to implicit cot: Learning to internalize cot step by step. arXiv preprint arXiv:2405.14838 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    arXiv preprint arXiv:2602.06965 (2026)

    Deria, A., Kumar, K., Dukre, A.M., Segal, E., Khan, S., Razzak, I.: Medmo: Grounding and understanding multimodal large language model for medical im- ages. arXiv preprint arXiv:2602.06965 (2026)

    work page arXiv 2026

  15. [15]

    arXiv preprint arXiv:2601.10949 (2026)

    Ding, M., Zhang, J., Wang, W., Zhong, H., Luo, X., Chen, W., Shen, L.: Mmedexpert-r1: Strengthening multimodal medical reasoning via domain-specific adaptation and clinical guideline reinforcement. arXiv preprint arXiv:2601.10949 (2026)

    work page arXiv 2026

  16. [16]

    In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (2025)

    Du, G., Fang, Z., Li, J., Li, J., Jiang, R., Yu, S., Guo, Y., Chen, Y., Goh, S.K., Tang, H.K., He, D., Liu, H., Zhang, M.: Neural parameter search for slimmer fine- tuned models and better transfer. In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (2025). https://doi.org/10.18653/v1/2025.a...

    work page doi:10.18653/v1/2025.acl-long.1570 2025

  17. [17]

    In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS) (2024)

    Du, G., Lee, J., Li, J., Jiang, R., Guo, Y., Yu, S., Liu, H., Goh, S.K., Tang, H.K., He, D., Zhang, M.: Parameter competition balancing for model merging. In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS) (2024)

    work page 2024

  18. [18]

    In: Proceedings of the International Conference on Learning Representations (ICLR) (2026)

    Du, G., Li, Z., Zhou, X., Li, J., Shi, Z., Lin, W., Tang, H.K., Li, X., Liu, F., Wang, W., Zhang, M., Li, J.: Knowledge fusion of large language models via modular skillpacks. In: Proceedings of the International Conference on Learning Representations (ICLR) (2026)

    work page 2026

  19. [19]

    Dynamic Model Merging Made Slim

    Du, G., Lin, W.: Dynamic model merging made slim. arXiv preprint arXiv:2605.18904 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    CRAB: Codebook Rebalancing for Bias Mitigation in Generative Recommendation

    Fan, Z., Chen, Z., Ma, L., Huang, J., Morishetti, L., Nag, K., Kumar, S., Achan, K.: Crab: Codebook rebalancing for bias mitigation in generative recommenda- tion. arXiv preprint arXiv:2604.05113 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    arXiv preprint arXiv:2404.12372 (2024)

    Gai, X., Zhou, C., Liu, J., Feng, Y., Wu, J., Liu, Z.: Medthink: Explaining med- ical visual question answering via multimodal decision-making rationale. arXiv preprint arXiv:2404.12372 (2024)

    work page arXiv 2024

  22. [22]

    IEEE Transactions on Medical Imaging (2025)

    Ge, J., Zhang, D., Zhan, Y., Liu, J., Gong, T., Wu, J., Crispin, M., Li, C., Gao, Z.: Progis: Prototype-guided interactive segmentation for pathological images. IEEE Transactions on Medical Imaging (2025)

    work page 2025

  23. [23]

    Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

    Geiping, J., McLeish, S., Jain, N., Kirchenbauer, J., Singh, S., Bartoldson, B.R., Kailkhura, B., Bhatele, A., Goldstein, T.: Scaling up test-time compute with latent reasoning: A recurrent depth approach. arXiv preprint arXiv:2502.05171 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  24. [24]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Gu, T., Yang, K., Liu, D., Cai, W.: Lapa: Latent prompt assist model for medi- cal visual question answering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4971–4980 (2024) 30 C. Zhu et al

    work page 2024

  25. [25]

    In: Proceedings of the 34th ACM International Conference on Information and Knowledge Management

    Guo, H., Xie, Z., Cao, S., Wang, B., Liu, W., Ye, Z., Li, Z., Liu, Z., Lu, W.: Pet- bench: Benchmarking the abilities of large language models as e-pets in social network services. In: Proceedings of the 34th ACM International Conference on Information and Knowledge Management. pp. 6402–6407 (2025)

    work page 2025

  26. [26]

    Quantized-tinyllava: a new multimodal foundation model enables efficient split learning.arXiv preprint arXiv:2511.23402, 2025

    Guo, J., Luo, X., Zheng, J., Wang, Y., Chang, K.W., Wang, W., Liu, J.: Quantized-tinyllava: A new multimodal foundation model enables efficient split learning. In: arXiv preprint arXiv:2511.23402 (2025)

    work page arXiv 2025

  27. [27]

    In: Findings of the Association for Computational Linguistics: EMNLP

    Han,X.,Hu,P.,Lu,C.,Ding,J.E.,Liu,F.,Ning,Y.:Noblackboxes:Interpretable and interactable predictive healthcare with knowledge-enhanced agentic causal discovery. In: Findings of the Association for Computational Linguistics: EMNLP

    work page

  28. [28]

    23415–23427 (2025)

    pp. 23415–23427 (2025)

    work page 2025

  29. [29]

    Training Large Language Models to Reason in a Continuous Latent Space

    Hao, S., Sukhbaatar, S., Su, D., Li, X., Hu, Z., Weston, J., Tian, Y.: Training large language models to reason in a continuous latent space. arXiv preprint arXiv:2412.06769 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  30. [30]

    PathVQA: 30000+ Questions for Medical Visual Question Answering

    He, X., Zhang, Y., Mou, L., Xing, E., Xie, P.: Pathvqa: 30000+ questions for medical visual question answering. arXiv preprint arXiv:2003.10286 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  31. [31]

    Proceedings of the VLDB Endowment 15(12) (2022)

    He, Y., Zhang, Y., Gurukar, S., Parthasarathy, S.: Webmile: democratizing net- work representation learning at scale. Proceedings of the VLDB Endowment 15(12) (2022)

    work page 2022

  32. [32]

    Iclr1(2), 3 (2022)

    Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W., et al.: Lora: Low-rank adaptation of large language models. Iclr1(2), 3 (2022)

    work page 2022

  33. [33]

    Hu, P., Han, X., Wang, F., Ning, Y.: Udoncare: Hierarchy pruning for unseen domaindiscoveryinpredictivehealthcare.arXivpreprintarXiv:2506.06977(2025)

    work page arXiv 2025

  34. [34]

    Exploring Accurate and Transparent Domain Adaptation in Predictive Healthcare via Concept-Grounded Orthogonal Inference

    Hu, P., Lu, C., Liu, F., Ning, Y.: Exploring accurate and transparent domain adaptation in predictive healthcare via concept-grounded orthogonal inference. arXiv preprint arXiv:2602.12542 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  35. [35]

    IEEE Journal of Biomedical and Health Informatics (2026)

    Hu,P.,Lu,C.,Wang,F.,Ning,Y.:Bridgingstepwiselab-informedpretrainingand knowledge-guided learning for diagnostic reasoning. IEEE Journal of Biomedical and Health Informatics (2026)

    work page 2026

  36. [36]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Hu, Y., Li, T., Lu, Q., Shao, W., He, J., Qiao, Y., Luo, P.: Omnimedvqa: A new large-scale comprehensive evaluation benchmark for medical lvlm. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 22170–22183 (2024)

    work page 2024

  37. [37]

    arXiv preprint arXiv:2506.17252 (2025)

    Huang, Z., Ban, Y., Fu, L., Li, X., Dai, Z., Li, J., Wang, D.: Adaptive sample scheduling for direct preference optimization. arXiv preprint arXiv:2506.17252 (2025)

    work page arXiv 2025

  38. [38]

    Huang, Z., Xia, X., Ren, Y., Zheng, J., Wang, X., Zhang, Z., Xie, H., Liang, S., Chen, Z., Xiao, X., et al.: Does your reasoning model implicitly know when to stop thinking? arXiv preprint arXiv:2602.08354 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  39. [39]

    Huang, Z., Xia, X., Ren, Y., Zheng, J., Xiao, X., Xie, H., Li, H., Liang, S., Dai, Z., Zhuang, F., Li, J., Ban, Y., Wang, D.: Real-time aligned reward model beyond semantics (2026),https://api.semanticscholar.org/CorpusID:285240754

    work page 2026

  40. [40]

    arXiv preprint arXiv:2510.08668 (2025)

    Jiang, S., Wang, Y., Song, S., Hu, T., Zhou, C., Pu, B., Zhang, Y., Yang, Z., Feng, Y., Zhou, J.T., et al.: Hulu-med: A transparent generalist model towards holistic medical vision-language understanding. arXiv preprint arXiv:2510.08668 (2025)

    work page arXiv 2025

  41. [41]

    In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Sig- nal Processing (ICASSP)

    Jiang, Y., Chen, J., Yang, D., Li, M., Wang, S., Wu, T., Li, K., Zhang, L.: Comt: Chain-of-medical-thought reduces hallucination in medical report generation. In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Sig- nal Processing (ICASSP). pp. 1–5. IEEE (2025) Medical Latent Memory Evolution 31

    work page 2025

  42. [42]

    In: ICASSP 2026- 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    Jiang, Y., Han, M., Li, M., Hou, X., Zhang, H., Zhu, W., Li, H., He, Y., Wu, G., Yang, D., et al.: Multi-agent diagnostic collaboration and segmentation- aware residual decoding for hallucination-resistant medical vqa. In: ICASSP 2026- 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 11122–11126. IEEE (2026)

    work page 2026

  43. [43]

    In: ICML (2026)

    Jiang, Y., Jiang, X., Zhang, L., Wang, Z., Lu, Y., Wang, P., Han, B., Zheng, F., Yang, D.: Mm-snowball: Evaluating and mitigating hallucination snowballing in multimodal multi-turn dialogue. In: ICML (2026)

    work page 2026

  44. [44]

    Advances in Neural Information Processing Systems38(2026)

    Jiang, Y., Li, J., Liu, Y., Yang, D., Zhou, F., Kong, Q.: Danmakutppbench: A multi-modal benchmark for temporal point process modeling and understanding. Advances in Neural Information Processing Systems38(2026)

    work page 2026

  45. [45]

    In: AAAI (2026)

    Jiang, Y., Xue, H., Han, M., Li, M., Hou, X., Yang, D., Zhang, L., Zheng, X.: Satiredecoder: Visual cascaded decoupling for enhancing satirical image compre- hension. In: AAAI (2026)

    work page 2026

  46. [46]

    Ke, H., Morris, J., Liu, Y., Kitai, S., Oguchi, K., Ding, Y., Wang, H.: Deformba: Vision state space model with adaptive state fusion (2026),https://arxiv.org/ abs/2605.21308

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  47. [47]

    In: The Thirteenth International Conference on Learning Representations (2025)

    Ke, H., Morris, J., Oguchi, K., Cao, X., Liu, Y., Wang, H., Ding, Y.: Mam- bev: Enabling state space models to learn birds-eye-view representations. In: The Thirteenth International Conference on Learning Representations (2025)

    work page 2025

  48. [48]

    IEEE Transactions on Medical Imaging (2026)

    Lai, Y., Zhong, J., Li, M., Zhao, S., Li, Y., Psounis, K., Yang, X.: Med-r1: Rein- forcement learning for generalizable medical reasoning in vision-language models. IEEE Transactions on Medical Imaging (2026)

    work page 2026

  49. [49]

    Scientific data 5(1), 1–10 (2018)

    Lau,J.J.,Gayen,S.,BenAbacha,A.,Demner-Fushman,D.:Adatasetofclinically generated visual questions and answers about radiology images. Scientific data 5(1), 1–10 (2018)

    work page 2018

  50. [50]

    arXiv preprint arXiv:2510.22728 (2025)

    Le-Duc, K., Nguyen, D.M., Trinh, P.T., Nguyen, T.P., Diep, N.T., Ngo, A., Vu, T., Vuong, T., Nguyen, A.T., Nguyen, M., et al.: S-chain: Structured visual chain- of-thought for medicine. arXiv preprint arXiv:2510.22728 (2025)

    work page arXiv 2025

  51. [51]

    Journal of general internal medicine40(10), 2240–2247 (2025)

    Lew, D., Baratta, L.R., Xia, L., Eiden, E., Sinsky, C.A., Kannampallil, T., Lou, S.S.: Association of ehr-integrated secure messaging use with clinician workload and attention switching. Journal of general internal medicine40(10), 2240–2247 (2025)

    work page 2025

  52. [52]

    Advances in neural information processing systems 33, 9459–9474 (2020)

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.t., Rocktäschel, T., et al.: Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems 33, 9459–9474 (2020)

    work page 2020

  53. [53]

    arXiv preprint arXiv:2602.01541 , year=

    Li, B., Shen, Y., Liu, Y., Xu, Y., Liu, J., Li, X., Li, Z., Zhu, J., Zhong, Y., Lan, F., et al.: Toward cognitive supersensing in multimodal large language model. arXiv preprint arXiv:2602.01541 (2026)

    work page arXiv 2026

  54. [54]

    Advances in Neural Information Processing Systems36, 28541–28564 (2023)

    Li, C., Wong, C., Zhang, S., Usuyama, N., Liu, H., Yang, J., Naumann, T., Poon, H., Gao, J.: Llava-med: Training a large language-and-vision assistant for biomedicine in one day. Advances in Neural Information Processing Systems36, 28541–28564 (2023)

    work page 2023

  55. [55]

    Seek in the dark: Reasoning via test-time instance-level policy gradient in latent space.arXiv preprint arXiv:2505.13308, 2025

    Li, H., Li, C., Wu, T., Zhu, X., Wang, Y., Yu, Z., Jiang, E.H., Zhu, S.C., Jia, Z., Wu, Y.N., et al.: Seek in the dark: Reasoning via test-time instance-level policy gradient in latent space. arXiv preprint arXiv:2505.13308 (2025)

    work page arXiv 2025

  56. [56]

    In: International Conference on Medical Image Comput- ing and Computer-Assisted Intervention

    Li, J., Dong, D., Zheng, M., Zhang, J., Hang, Y., Zhang, L., Zhao, L.: High- precision mixed feature fusion network using hypergraph computation for cervical abnormal cell detection. In: International Conference on Medical Image Comput- ing and Computer-Assisted Intervention. pp. 250–259. Springer (2025) 32 C. Zhu et al

    work page 2025

  57. [57]

    Li, S., Guo, X., Liu, T., Yi, B., Gong, Z., Liu, Z., Chen, H., Zhang, W.: What’s missing in screen-to-action? towards a ui-in-the-loop paradigm for multimodal gui reasoning (2026),https://arxiv.org/abs/2604.06995

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  58. [58]

    In: Proceedings of the 2025 Con- ference on Empirical Methods in Natural Language Processing

    Li, S., Liu, Z., Gui, Z., Chen, H., Zhang, W.: Enrich-on-graph: Query-graph align- ment for complex reasoning with llm enriching. In: Proceedings of the 2025 Con- ference on Empirical Methods in Natural Language Processing. pp. 7683–7703. Association for Computational Linguistics (2025).https://doi.org/10.18653/ v1/2025.emnlp- main.390,http://dx.doi.org/1...

    work page doi:10.18653/v1/2025.emnlp- 2025

  59. [59]

    Li, S., Liu, Z., Zhu, Y., Chen, H., Zhang, W.: Last layer logits to logic: Empow- ering llms with logic-consistent structured knowledge reasoning (2025),https: //arxiv.org/abs/2511.07910

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  60. [60]

    arXiv preprint arXiv:2411.14522 (2024)

    Li, T., Su, Y., Li, W., Fu, B., Chen, Z., Huang, Z., Wang, G., Ma, C., Chen, Y., Hu, M., et al.: Gmai-vl & gmai-vl-5.5 m: A large vision-language model and a comprehensive multimodal dataset towards general medical ai. arXiv preprint arXiv:2411.14522 (2024)

    work page arXiv 2024

  61. [61]

    In: 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

    Li, Y., Ge, J., Cao, S., Zhan, Y., Gong, Z., Niu, J., Liu, J., Zhang, D., Li, C.: Similarity-aware dual-perspective learning for medical event prediction. In: 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). pp. 6263–6270. IEEE (2025)

    work page 2025

  62. [62]

    In: Proceedings of the 33rd ACM International Conference on Multimedia

    Liang, G., Hu, J., Xing, X., Zhang, J., Yu, Q.: Multi-object sketch animation with grouping and motion trajectory priors. In: Proceedings of the 33rd ACM International Conference on Multimedia. pp. 9237–9246 (2025)

    work page 2025

  63. [63]

    Render-in-the-Loop: Vector Graphics Generation via Visual Self-Feedback

    Liang, G., Wang, Z., Hu, J., Zhou, H., Xue, Z., Zhang, J., Xu, D., Yu, Q.: Render- in-the-loop: Vector graphics generation via visual self-feedback. arXiv preprint arXiv:2604.20730 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  64. [64]

    VAnim: Rendering-Aware Sparse State Modeling for Structure-Preserving Vector Animation

    Liang, G., Wang, Z., Wang, C., Hu, J., Zhou, H., Liu, J., Zhang, J., Xu, D., Yu, Q.: Vanim: Rendering-aware sparse state modeling for structure-preserving vector animation. arXiv preprint arXiv:2605.01517 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  65. [65]

    arXiv preprint arXiv:2602.18745 (2026)

    Lin, H., Bai, T., Chen, C., Zhang, J., Zeng, B., Zhang, W., Yuan, B.: Synthesizing multimodal geometry datasets from scratch and enabling visual alignment via plotting code. arXiv preprint arXiv:2602.18745 (2026)

    work page arXiv 2026

  66. [66]

    arXiv preprint arXiv:2602.18763 (2026)

    Lin, H., Bai, T., Zhang, J., Chang, X., Lu, S., Gu, F., Hu, Z., Zhang, W.: Tag: Thinking with action unit grounding for facial expression recognition. arXiv preprint arXiv:2602.18763 (2026)

    work page arXiv 2026

  67. [67]

    MMFineReason: Closing the multimodal reasoning gap via open data-centric methods

    Lin, H., Liu, Z., Zhu, Y., Qin, C., Lin, J., Shang, X., He, C., Zhang, W., Wu, L.: Mmfinereason: Closing the multimodal reasoning gap via open data-centric methods. arXiv preprint arXiv:2601.21821 (2026)

    work page arXiv 2026

  68. [68]

    IEEE Transactions on Information theory37(1), 145–151 (2002)

    Lin, J.: Divergence measures based on the shannon entropy. IEEE Transactions on Information theory37(1), 145–151 (2002)

    work page 2002

  69. [69]

    When Models Learn to Ask Why: Adaptive Causal Reasoning for Trustworthy Medical Vision-Language Models

    Lin, J., Zhu, C., Kneuertz, P.J., Bai, Y., Xue, Y.: Medcausalx: Adaptive causal reasoning with self-reflection for trustworthy medical vision-language models. arXiv preprint arXiv:2603.23085 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  70. [70]

    Lin, L., You, J., Li, Y., Lin, L., Wang, Y., Zhang, Z., Zheng, M.: Reflect-guard: Enhancing llm safeguards against adversarial prompts via logical self-reflection (2026),https://arxiv.org/abs/2605.24834

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  71. [71]

    arXiv preprint arXiv:2511.19046 (2025)

    Liu, A., Xue, R., Cao, X.R., Shen, Y., Lu, Y., Li, X., Chen, Q., Chen, J.: Medsam3: Delving into segment anything with medical concepts. arXiv preprint arXiv:2511.19046 (2025) Medical Latent Memory Evolution 33

    work page arXiv 2025

  72. [72]

    In: 2021IEEE18thinternationalsymposiumonbiomedicalimaging(ISBI).pp.1650–

    Liu, B., Zhan, L.M., Xu, L., Ma, L., Yang, Y., Wu, X.M.: Slake: A semantically- labeled knowledge-enhanced dataset for medical visual question answering. In: 2021IEEE18thinternationalsymposiumonbiomedicalimaging(ISBI).pp.1650–

    work page

  73. [73]

    arXiv preprint arXiv:2412.17747 (2024)

    Liu, L., Pfeiffer, J., Wu, J., Xie, J., Szlam, A.: Deliberation in latent space via differentiable cache augmentation. arXiv preprint arXiv:2412.17747 (2024)

    work page arXiv 2024

  74. [74]

    arXiv preprint arXiv:2602.11164 (2026)

    Liu, W., Wu, H., Kuang, Y., Han, X., Zhong, T., Feng, J., Lu, W.: Automated optimization modeling via a localizable error-driven perspective. arXiv preprint arXiv:2602.11164 (2026)

    work page arXiv 2026

  75. [75]

    Liu, Y., Cheng, Z., Lai, L.: Improving the completeness and comparability of segmentdisclosures:Alargelanguagemodelapproach.AvailableatSSRN6720239 (2026)

    work page 2026

  76. [76]

    arXiv preprint arXiv:2407.207563(2024)

    Liu, Z., Liang, H., Huang, X., Xiong, W., Yu, Q., Sun, L., Chen, C., He, C., Cui, B., Zhang, W.: Synthvlm: High-efficiency and high-quality synthetic data for vision language models. arXiv preprint arXiv:2407.207563(2024)

    work page arXiv 2024

  77. [77]

    ChartVerse: Scaling Chart Reasoning via Reliable Programmatic Synthesis from Scratch

    Liu, Z., Lin, H., Qin, C., Wang, X., Gao, X., Li, Y., Cai, M., Zhu, Y., Zhong, Z., Pei, Q., et al.: Chartverse: Scaling chart reasoning via reliable programmatic synthesis from scratch. arXiv preprint arXiv:2601.13606 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  78. [78]

    FLARE: Fully Integration of Vision-Language Representations for Deep Cross-Modal Understanding

    Liu, Z., Liu, M., Chen, J., Xu, J., Cui, B., He, C., Zhang, W.: Fusion: Fully in- tegration of vision-language representations for deep cross-modal understanding. arXiv preprint arXiv:2504.09925 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  79. [79]

    JAMA Network Open7(12), e2447797 (2024)

    Lou, S.S., Lew, D., Xia, L., Baratta, L., Eiden, E., Kannampallil, T.: Secure mes- saging use and wrong-patient ordering errors among inpatient clinicians. JAMA Network Open7(12), e2447797 (2024)

    work page 2024

  80. [80]

    In: Machine learning for health (ML4H)

    Moor, M., Huang, Q., Wu, S., Yasunaga, M., Dalmia, Y., Leskovec, J., Zakka, C., Reis, E.P., Rajpurkar, P.: Med-flamingo: a multimodal medical few-shot learner. In: Machine learning for health (ML4H). pp. 353–367. PMLR (2023)

    work page 2023

Showing first 80 references.