arXiv subjects

Yuhang Liu

Publications and source records attributed to Yuhang Liu.

At least 19 recordsLinked to original sources

arXiv2026

DoAtlas-2: A Foundation for Self-Evolving Causal Biomedical Discovery

We introduce DoAtlas-2, a foundation for self-evolving causal biomedical discovery that organizes knowledge around causal mechanisms and advances through external evidence from human populations. DoAtlas-2 integrates 771 research resources covering more than 720,000 participants in 48 countries, from longitudinal clinical phenotypes, medical imaging, and continuous physiological signals to eight molecular layers, together with an evidence network of approximately 4.7 million literature-derived records over 93,566 concepts and 149,383 candidate causal relations. DoAtlas-2 autonomously formulates research questions from evidence gaps and unresolved mechanisms, prespecifies their causal designs, and generates validated analyses. Supporting, challenging, and unresolved results continuously revise mechanistic interpretations, the causal evidence state, and the discovery frontier, so that DoAtlas-2 self-evolves within a closed loop of hypothesis generation, empirical testing, and renewed discovery. DoAtlas-2 has systematically evaluated 2,031 research questions. In the Human Phenotype Project (HPP), it formulated 4,014 candidate pathway questions across vascular, early-glycemic, and hepatic-metabolic systems, and screening of the first 1,079 yielded statistical support for 756. Representative studies identify blood pressure as a convergence node linking adiposity, hepatic, and lipid phenotypes to vascular outcomes, and show that an adiposity-inflammation-blood-pressure pathway is largely attenuated by joint adjustment for body mass index (BMI) and smoking. The discovered vascular network constitutes a completely interpretable predictive foundation, admitting exact attribution of every prediction and closed-form mediation effects. DoAtlas-2 thereby unifies causal mechanism discovery, population-evidence testing, and interpretable prediction within one continuously evolving foundation.

cs.AI↗

arXiv2026

Decision Transformer with Task-Aware Representations for Offline Meta-Reinforcement Learning

To enable effective generalization in offline meta-RL, it is essential to accurately infer the underlying task through latent task representations, which provide a compact and transferable summary of task-specific characteristics for guiding policy generation. In this paper, we propose Contrastive Task-Aware Decision Transformer (CoTA-DT), a DT-based offline meta-RL framework that learns task representations by jointly integrating four complementary components: (i) contrastive learning from offline trajectories, (ii) alignment with natural-language task descriptions, (iii) one-step supervision via a task-conditioned world model, and (iv) long-term supervision via a task-conditioned Q function. The learned task representations are used as conditioning signals for a DT-based policy, enabling task-aware meta-policy learning. To further support generalization in both zero-shot and few-shot settings, we design a language-grounded, self-adaptive strategy that enables effective and reliable adaptation to unseen tasks. Notably, in CoTA-DT, the pretrained LLM serves as a shared backbone for language encoding, world modeling, Q-function estimation, and policy learning. This unified design facilitates parameter sharing, improving both efficiency and generalization. Extensive experiments across diverse offline meta-RL benchmarks demonstrate that CoTA-DT consistently outperforms prior methods, particularly in challenging zero-shot generalization scenarios.

cs.LG↗

arXiv2026

Language Model Augmented Decision Transformer for Multi-Task Offline Reinforcement Learning

Offline multi-task reinforcement learning (MTRL) aims to learn a unified policy from heterogeneous offline datasets that generalizes across tasks. However, existing offline MTRL methods often struggle with two fundamental issues: severe per-task data scarcity and limited cross-task generalization. In this paper, we propose Language-Model Augmented Decision Transformer (LaMA-DT), a unified sequence-modeling framework that leverages a single large language model (LLM) for both trajectory augmentation and multi-task policy learning. LaMA-DT uses natural-language task descriptions to guide task-aligned trajectory augmentation through a set of supervised generation and alignment objectives, and incorporates a scoring-and-filtering mechanism to retain high-quality augmented trajectories. The same LLM is then repurposed as a language-conditioned Decision Transformer and fine-tuned on a mixture of real and augmented data. Experiments on Meta-World and MuJoCo multi-task benchmarks demonstrate that LaMA-DT consistently outperforms competitive baselines, particularly in low-data regimes, showing that language-guided data augmentation substantially improves cross-task generalization.

cs.CL↗

arXiv2026

SemantiScene-Edit: Expert-Synthesized Supervision and Dual-Path Source Conditioning for Instruction-Based Video Editing

Instruction-guided video editing requires executing a requested change while preserving the source video's appearance and temporal structure. We present SemantiScene-Edit, a video editor that couples a multimodal encoder with a pretrained video diffusion transformer. The encoder jointly processes source frames and the editing instruction, and learnable queries convey their semantic context to the generator through a trainable connector and cross-attention. A complementary visual pathway concatenates source-video latents with noisy target latents before patch embedding, supplying dense spatial and temporal information for content preservation. We train the model with flow matching and a three-stage image--video curriculum comprising semantic alignment, joint adaptation, and high-resolution refinement. To support training, we construct over 1.2 million instruction-conditioned video-editing records covering object addition, removal, replacement, and attribute modification. We also construct a 100-case benchmark for evaluating instruction execution, source preservation, temporal consistency, and visual quality. Comparative experiments show that SemantiScene-Edit effectively follows editing instructions while retaining source content. At inference, the model requires only a source video and a natural-language instruction.

cs.AI↗

arXiv2026

On the Value of Cross-Modal Misalignment in Multimodal Representation Learning

Multimodal representation learning, exemplified by multimodal contrastive learning (MMCL) using image-text pairs, aims to learn powerful representations by aligning cues across modalities. This approach relies on the core assumption that the exemplar image-text pairs constitute two representations of an identical concept. However, recent research has revealed that real-world datasets often exhibit \emph{cross-modal misalignment}. There are two distinct viewpoints on how to address this issue: one suggests mitigating the misalignment, and the other leveraging it. We seek here to reconcile these seemingly opposing perspectives and to provide a practical guide for practitioners. Using latent variable models, we thus formalize cross-modal misalignment by introducing two specific mechanisms: \emph{selection bias}, where some semantic variables are absent in the text, and \emph{perturbation bias}, where semantic variables are altered---both leading to misalignment in data pairs. Our theoretical analysis shows that, under suitable assumptions, population global minimizers of an alignment--entropy surrogate yield representations that depend only on the semantic variables unaffected by selection and perturbation biases and locally identify this subset almost surely. Extensive empirical studies on synthetic data and real image-text datasets further examine how cross-modal misalignment shapes semantic predictability and downstream performance, shedding light on its nuanced impact on multimodal representation learning.

cs.LG↗

arXiv2026

A Principled Optimal Transport Framework for Frame Selection in Long Video Understanding

Long-video understanding with multimodal LLMs is fundamentally constrained by the mismatch between video length and the model's limited visual context budget, making frame selection essential for practical reasoning. Existing query-aware methods typically rely on heuristic pipelines: they assign one-dimensional query-conditioned scores to frames and then apply sampling, ranking, or extraction rules to form the final subset. While effective in practice, such methods are generally not derived from an explicit optimization objective and struggle to capture the heterogeneous evidence required for complex reasoning. In this paper, we revisit long-video frame selection from a principled optimization perspective. We formulate it as a structured evidence-allocation problem and propose OTFS, an optimal transport framework that maps multiple evidence sources onto video frames. To reflect the asymmetry of this problem---important evidence should be preserved, whereas redundant frames need not absorb mass---we instantiate this view with a semi-unbalanced entropic optimal transport objective and an efficient solver, OTFS-Sinkhorn. Combined with an exact dynamic program for coverage-aware subset extraction, OTFS yields a practical two-stage training-free pipeline. We further show that, in the single-source case, our formulation induces a Gibbs-type distribution over frames, making Q-Frame's temperature-scaled distribution a special case of our framework while also providing a broader perspective for understanding related prior methods. Experiments on three long-video understanding benchmarks show that OTFS consistently outperforms strong training-free baselines. Codes are available at https://github.com/barry664/OTFS.

cs.CV↗

arXiv2026

The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-modal Divergence

While InfoNCE underlies modern contrastive learning, its geometric mechanisms remain under-characterized beyond the canonical alignment--uniformity decomposition. We develop a measure-theoretic framework in which representation measures evolve on a fixed embedding manifold. In the large-batch limit, we prove value and gradient consistency, linking the stochastic objective to explicit deterministic energy landscapes and revealing a geometric bifurcation between unimodal and symmetric multimodal regimes. In the unimodal case, the intrinsic energy is strictly convex and admits a unique Gibbs equilibrium, showing that entropy acts as a tie-breaker within the aligned basin. In the multimodal case, the intrinsic geometry becomes cross-coupled and contains a persistent negative symmetric divergence term: each modality's marginal reshapes the effective landscape of the other, allowing strong pairwise alignment to coexist with a persistent modality gap. Controlled synthetic experiments and analyses of pretrained CLIP representations support these predictions. Overall, our results shift the analytical lens from pointwise discrimination to population geometry, showing that pairwise alignment alone is insufficient to control cross-modal marginal structure.

cs.LG↗

arXiv2026

From Parameters to Behaviors: A Survey of Model Fusion for Large Language Models

Model fusion integrates the capabilities from source models into a single target model. As of June 2026, Hugging Face hosts more than 2M models. This growing pool provides a rich base for model reuse and capability integration. Yet existing surveys often cover only separate parts of this space, and they do not provide a unified definition or a systematic taxonomy. This survey defines model fusion and organizes prior work into three levels: parameter-level, representation-level, and behavior-level fusion. We also review related metrics, benchmarks, and applications, summarize current challenges, and identify future directions. Our goal is to provide a clear map of this area and support future work on model fusion. A comprehensive list of papers about model fusion is available at https://github.com/Baicaihaochi/Awesome-Model-Fusion-Survey.

cs.CL↗

arXiv2026

Orbit stratifications and smooth topology of positively curved six-manifolds with rank-one nonabelian symmetry

We study closed simply connected six-manifolds with positive sectional curvature and an effective isometric action of $SO(3)$ or $SU(2)$. Starting from Liu's corrected orbit-space analysis, we refine the stratification and recover smooth topology in several previously unresolved cases. For a boundaryless $SO(3)$ quotient, singular isotropy is connected, the weighted nonprincipal graph belongs to nine necessary families, and every action with only cyclic exceptional isotropy is classified equivariantly. The remaining branching families admit explicit rational intersection and Pontryagin formulas. For a three-ball $SO(3)$ quotient, the uniform-boundary configurations give sphere bundles over $S^4$ and a completely distinguished odd-integer family of non-spin six-manifolds. A doubled-quotient argument excludes the configuration with two boundary fixed points and an interior singular orbit, while the configuration with one fixed point and an exceptional interval has underlying manifold $\mathbb{CP}^3$. For a boundaryless $SU(2)$ quotient, an adjoint-weight matching argument reduces the exceptional three-orbit case to a unique action on the Eschenburg flag and identifies every two-orbit candidate with a weighted action on $S^2\times S^4$. These classifications imply rational ellipticity throughout the stated symmetry class. We also establish restricted curvature obstructions for the two diagonal product actions, without asserting nonexistence for arbitrary invariant metrics. Abstract graph restrictions, smooth realizability, and positive-curvature realizability are kept distinct. Some of the results in this paper are discovered by ChatGPT-6-astra and verified by the authors.

math.DG↗

arXiv2026

SMAT: Simple and Efficient Merge-Aware Training

Model merging integrates the capabilities of multiple experts without joint retraining, but standard expert training optimizes task loss alone and does not guarantee good performance after merging. Merge-aware training (MAT) aims to improve merged performance, but existing methods do not fully account for common merging operations and add training cost. We observe that, from an expert's perspective, common merging methods can be described by three operations: Scale reweights its own update, Mask removes selected coordinates, and Perturb adds updates from other experts. Based on this view, we introduce SMAT (Simple MAT), which jointly optimizes expert loss and expected loss at simulated merged parameters generated by sampling scaling coefficients, masks, and additive noise. We further introduce periodic scheduling, kernel fusion, and parameter storage switching to make SMAT efficient, with one forward and one backward pass per step. Across four language and vision-language backbones, SMAT improves the mean score across five merging methods by 1.07-2.16 points over the strongest baseline for each backbone, with less than 2% training-time overhead over standard fine-tuning.

cs.LG↗

arXiv2026

FusionEval: Benchmarking Model Fusion for Large Language Models

Model fusion aims to integrate the capabilities of multi-expert models into a single model. Existing evaluations offer limited coverage of reasoning with long chains of thought and tool use, while strong average performance can mask substantial losses in individual expert capabilities. We introduce FusionEval, a benchmark that provides controlled expert pools, integrates fusion methods with and without additional training, and offers a unified evaluation suite. The pools contain domain experts independently trained from Qwen3 and Llama base models across math, code, science, instruction following, and tool use. The suite evaluates capability retention and computational cost on eight public benchmarks under comparable data access rules, reporting normalized average and worst performance, runtime, and peak GPU memory usage. Extensive experiments show that training-free methods preserve reasoning performance but substantially degrade tool use at both model scales. In contrast, MOPD preserves tool use and achieves a more balanced capability profile, but requires additional data preparation, inference, and training. Jointly retaining reasoning, instruction following, and tool use at low computational cost remains a challenge. We hope FusionEval supports future research on efficient model fusion that jointly preserves expert capabilities. Our code is available at https://github.com/Baicaihaochi/FusionEval_Benchmark.

cs.CL↗

arXiv2026

I Act Therefore I Am: When Is JEPA's Action-Conditioning Enough to Learn Causal Mechanisms?

Recent empirical and theoretical advances suggest that joint-embedding predictive architectures (JEPAs) may learn meaningful representations for action-conditioned prediction of future outcomes, thus becoming one of the foundational structures for world models. However, accurate prediction does not, in general, necessarily imply recovery of underlying causal states that give rise to the observed dynamics. This work investigates when and how JEPAs can recover the underlying causal states from observations. We first introduce a latent variable model, in which high-dimensional observations are generated from latent causal states whose dynamics are governed by action-conditioned transition mechanisms. Based on this formulation, we develop a general information-theoretic objective that combines conditional likelihood maximization for learning transition dynamics with entropy maximization for preserving latent state information. We then establish identifiability conditions under which representations learned by this general objective recover the underlying latent causal states up to component-wise invertible transformations and permutation. One key condition for such identifiability is sufficient action-induced variation in the transition mechanisms. Guided by this finding, we instantiate the general objective with an action-modulated Gaussian additive-noise model, yielding action-modulated JEPA (A-JEPA). Experiments on synthetic environments verify the theoretical findings under the identifiability conditions and robustness to moderate violations, while visual benchmarks demonstrate improved state recovery and transfer to unseen transition mechanisms.

cs.LG↗

arXiv2026

When and Where to Trust the Teacher: Unifying On-Policy Distillation and GRPO through Entropy-Calibrated Credit Assignment

Reinforcement learning with verifiable rewards (RLVR) supervises mathematical reasoning through final-answer correctness, but provides little guidance on individual tokens. On-policy distillation (OPD) supplies dense feedback on student-generated responses, yet teacher preference need not reflect correctness. Recent hybrids combine OPD and verifier-derived advantages or reweight task credit using teacher ratios. However, teacher guidance enters after verifier-based group normalization, and token reweighting need not preserve the total task credit assigned to each response. We introduce Unified Entropy-Calibrated Credit Redistribution for GRPO (UECR-GRPO), which integrates verifier and teacher signals within a single GRPO-style update at both the response and token levels. \emph{Path-Utility Unification} (PUU) combines verifier reward and a teacher-to-anchor path log-ratio in a single KL-regularized objective. Its on-policy implementation uses a length-normalized teacher score and combines both rewards before group normalization and PPO clipping, allowing teacher evidence to influence the response ranking. \emph{Entropy-Calibrated Redistribution} (ECR) then uses the signed teacher--old-policy token gap to redistribute the verifier-derived component. Full-vocabulary teacher entropy attenuates uncertain guidance, while a response-wise zero-sum projection preserves the total task credit and its token-wise sign before clipping. Across five mathematical reasoning benchmarks, UECR-GRPO achieves average $\mathrm{Avg@12}$ accuracies of 17.21\% and 65.09\% with Qwen3-1.7B and Qwen3-4B students, respectively, exceeding the strongest baseline at each scale by 0.89 and 0.56 percentage points.

cs.LG↗

arXiv2026

Curvature Sign Rigidity and Sharp Pointwise Pinching Thresholds

We study connected Riemannian manifolds on which either the sectional curvature or the Ricci tensor is, at each point, strictly positive, strictly negative, or zero, and ask whether the two signs can coexist under pointwise pinching. A general support-rigidity theorem for positive semidefinite divergence-free symmetric tensors is the common analytic mechanism. For sectional curvature in dimension $n\ge 3$, any locally uniform positive lower bound for the absolute pointwise pinching ratio rules out a change of sign. With a fixed pinching constant $δ_0$, the flat set has no $C^1$ hypersurface piece; it is empty when $δ_0>1/2$, and is locally porous when $δ_0=1/2$. These conclusions are sharp in several senses: there are smooth conformally flat local metrics whose sectional curvature changes sign across a flat hypersurface when the pinching degenerates, and there are closed exactly $1/q$-pinched metrics on spheres with isolated flat points. For Ricci curvature, the sharp threshold is \[ δ_c=\frac1{n-1}. \] A locally uniform gap $δ_{\mathrm{Ric}}>δ_c$ forces one Ricci sign globally. Conversely, for every $0<δ<δ_c$ there are local sign-changing metrics with exact pointwise pinching $δ_{\mathrm{Ric}}\equivδ$, and exact local examples also exist at $δ=δ_c$. Pointwise strictness $δ_{\mathrm{Ric}}>δ_c$ is insufficient if the gap collapses at a Ricci-flat interface. For every subcritical $δ$ we give explicit closed metrics on $\mathbb{S}^1\times\mathbb{S}^{n-1}$, and complete periodic lifts to $\mathbb{R}\times\mathbb{S}^{n-1}$, whose optimal global lower pinching constant is exactly $δ$. Finally, we prove two ansatz-specific obstructions to compactifying the exact local constructions. The main content of the proof is generated by ChatGPT 5.6 sol and verified by the authors.

math.DG↗

arXiv2026

GUI-PRA: Process Reward Agent for GUI Tasks

Long-horizon GUI automation remains challenging due to error accumulation over extended interaction sequences. Process Reward Models (PRMs) provide dense step-level supervision for mitigating error accumulation, yet standard PRMs are poorly suited to GUI verification. Standard PRM judgments often rely on superficial visual alignment rather than functional correctness, reflecting an evaluative knowledge gap caused by missing domain-specific adjudication logic. Standard PRMs also perform passive, single-pass visual assessment, which creates Visual Ambiguity when reliable judgment requires actively locating, parsing, or inspecting task-relevant UI evidence. We introduce GUI-PRA, a Process Reward Agent that transforms GUI process evaluation from passive scoring into active investigation. GUI-PRA couples Experience-Injected Criterion Synthesis, which distills generalized verification principles into state-specific criteria, with Criterion-Guided Autoregressive Perception, which uses these criteria to navigate multi-granularity visual tools and gather grounded evidence. On AndroidWorld and Mobile-MiniWoB++, GUI-PRA achieves improvements of 5.0% and 6.5% over standard PRMs on the Qwen-VL series, with Qwen3-VL attaining 54.74% success rate on AndroidWorld. On the offline OS-Critic Bench, GUI-PRA demonstrates strong competitiveness against fully trained critic models.

cs.AI↗

arXiv2026

InjecMEM: Memory Injection Attack on LLM Agent Memory Systems

Memory is becoming a default subsystem in deployed LLM agents to provide persistent personalization and continuity. This naturally prompts a question: will memory system introduce new vulnerabilities into agents? Thus we propose InjecMEM, a novel memory injection attack paradigm that requires only a single interaction (no read/edit access to memory store) to steer later responses of related queries toward a pre-specified output. Guided by the retrieval-then-generate mechanism of memory systems, we craft the injection with a retriever-agnostic anchor and an adversarial command. The anchor contains high-recall topical cues so that downstream retrieval consistently associates the record with the target topic. The command is a short sequence optimized to remain effective under uncertain fused contexts, variable placements, and long prompts so that it reliably steers outputs once retrieved. We learn the command via gradient-based coordinate search, averaging over synthetic prompt templates and insertion positions, and extend it to joint optimization across backbones to study transfer. Evaluated across multiple memory systems and backbone models, InjecMEM achieves reliable topic-conditioned retrieval and targeted generation, remains effective under memory drift, and leaves non-target queries unaffected. Our results underscore the need to harden memory systems and provide a reproducible framework for studying agent memory.

cs.CR↗

arXiv2026

On Closed 6-Manifolds Admitting Riemannian Metrics with Positive Sectional Curvature and Non-Abelian Symmetry

We study the topology of closed, simply-connected, 6-dimensional Riemannian manifolds of positive sectional curvature which admit isometric actions by $SU(2)$ or $SO(3)$. We show that their Euler characteristic agrees with that of the known examples, i.e., $S^6$, $\mathbb{CP}^3$, the Wallach space $SU(3)/T^2$ and the biquotient $SU(3)//T^2$. We determine the non-trivial principal isotropy actions, classify the fixed-point-free $SU(2)$-actions without exceptional orbits, and give restrictions on the exceptional strata. In particular, an exceptional isotropy group for an $SU(2)$-action is cyclic of odd order, while tetrahedral exceptional isotropy occurs for the irreducible linear $SO(3)$-action on the round $6$-sphere.

math.DG↗

arXiv2026

On the Positivity of the Products of Positive Primitive Forms

Let $(V_i,ω_i)$ be real symplectic vector spaces and let $Ω_i\in\mathcal U^+(V_i)$ in the sense of Haiden. We prove that $p_1^*Ω_1\wedge p_2^*Ω_2$ belongs to $\mathcal U^+(V_1\oplus V_2)$ whenever one factor has real dimension at most six. There are forms $Ω\in\mathcal{U}^+(\mathbb{R}^6)\setminus\mathcal{U}_{\mathrm{ag}}(\mathbb{R}^6)$ for which $p_1^*Ω\wedge\cdots\wedge p_k^*Ω$ belongs to $\mathcal{U}^+$ for every $k\geq 1$; hence a conjecture of Kontsevich does not hold in complex dimension three. We also give a criterion for a product to belong to $\mathcal{U}$ and show that, for every $N\geq 26$, there are two forms in $\mathcal{U}^+(\mathbb{C}^N)$ whose exterior product does not belong to $\mathcal{U}$. The main content of this paper is generated by ChatGPT 5.6 and verified by the author.

math.SG↗