arXiv subjects

Wanyu Lin

Publications and source records attributed to Wanyu Lin.

At least 19 recordsLinked to original sources

arXiv2026

Learning Generative Dynamics for 3D Molecule Generation via Sequential Neuro-Symbolic Constraints

Diffusion models have recently emerged as a powerful paradigm for generating 3D molecules. Despite showing remarkable promise, they learn atomic distributions implicitly from data, without explicitly encoding physicochemical rules into their generative dynamics. Consequently, generated samples can be statistically plausible yet chemically invalid. To address this problem, we propose Sequential Neuro-Symbolic Constrained Diffusion (SensDiff), a framework that integrates symbolic rules into generative modeling, enabling the model to internalize scientific priors during training. Central to SensDiff is the observation that the diffusion trajectory of 3D molecule generation exhibits a coarse-to-fine evolution: valid molecules arise by first mitigating macroscopic geometric conflicts and then refining structures toward microscopic chemical consistency through iterative denoising. Accordingly, SensDiff mirrors this dynamic by sequentially enforcing constraints with adaptive weighting during training, progressively steering generation toward valid molecular geometries. Moreover, SensDiff translates domain knowledge into interpretable generative controls, grounding opaque neural denoising in scientific principles. Experiments on molecular benchmarks and real-world constrained generation tasks confirm that SensDiff consistently improves chemical validity while adhering to physicochemical rules and user-specified targets. The code is publicly available at https://github.com/WanyuGroup/NeurIPS2026_SensDiff.

cs.LG↗

arXiv2026

Controllable Molecule Generation via Sparse Representation Editing: An Interpretability-Driven Perspective

Controllable molecule generation is crucial for diverse scientific applications, such as drug discovery and materials design. While large language models (LLMs) show great promise, their dense and entangled representations impede precise control over the generation of molecules with bespoke substructures or properties. To address this, we propose Sparse Representation Editing (SpaRE), an interpretability-driven framework for fine-grained and precise control in LLM-based molecule generation. The crux of SpaRE is to learn an overcomplete sparse feature space that disentangles LLM representations into a compact set of latent features corresponding to chemically meaningful concepts. Within this space, we can directly manipulate these concept-aligned latent features to achieve (1) local control, by generating target atoms and functional groups at specified positions; and (2) global control, by customizing the overall structural and physicochemical properties within defined ranges. In this way, our framework advances interpretability from post-hoc analysis to actionable generative control. Experiments show that SpaRE can generate chemically desirable molecules under complex constraints in real-world scenarios, while offering mechanistic insights for quantitative structure-property analysis. The code and demo are available at https://github.com/WanyuGroup/ICML2026_SpaRE.

cs.LG↗

arXiv2026

AgentSnare: Learning to Delay, Divert, and Defuse Autonomous Web Penetration Agents

Large language model agents automate web penetration testing through observation-action loops, enabling defenders to influence decisions with deceptive decoy observations. Advanced agents can recognize and bypass static decoys, then refocus on real targets. We propose AgentSnare, a trajectory-adaptive deception system that expands a decoy environment alongside exploration. For online deployment, ShadowOS combines an LLM-driven decoy constructor with a virtual kernel. Conditioned on exploration, it generates decoy artifacts that the kernel materializes and handles consistently across interactions, sustaining long-horizon exploration. Because deploying a powerful LLM for decoy construction is costly and slow, we train a small LLM using an offline adversarial interaction-simulation framework. It synthesizes multi-round trajectories by generating candidates and retaining those that sustain simulated interactions. Across 15 CVE-Bench web applications and four attacker models, AgentSnare absorbs 49.9% of tool calls to delay attacks, retains 57.2% of post-entry actions in the decoy to divert trajectories, and grounds 89.5% of completion reports in decoy evidence to defuse threats; across 60 attacker-CVE pairs, no real target is exploited at pass@3.

cs.CR↗

arXiv2026

Symmetry-Preserving Local Representation for Short-Range Interaction in Complex Materials

Modeling short-range interactions in highly disordered materials remains challenging due to the immense configuration space and the disruption of local symmetry in truncated theoretical models. Here, we propose a symmetry-preserving local representation framework that strictly enforces topological continuity and translational invariance via boundary passivation. Applied to high-entropy alloys, this approach autonomously unravels the nonlinear ligand effects modulating local reactivity and accurately captures coverage-dependent many-body interactions without explicit periodic expansion. By resolving the conflict between localized chemical environments and extended periodic boundaries, this framework not only reduces computational complexity by an order of magnitude but also provides a physically transparent paradigm for deciphering short-range order in complex condensed matter systems.

cond-mat.dis-nn↗

arXiv2026

Dynamic Model Merging Made Slim

Model merging consolidates independently fine-tuned checkpoints without joint training or access to task data. Dynamic methods retain task-specific capacity, but must divide a limited storage budget between shared and expert parameters. We propose DiDi-Merging, which jointly allocates low-rank capacity to these components through differentiable rank selection, followed by data-free parameter-space refinement. On the eight-task ViT-B/32 benchmark, the small configuration reaches 89.3% average accuracy (98.7% normalized retention) at 1.24x the storage of one backbone, including that backbone; the medium configuration reaches 89.8% at 1.4x. Removing refinement reduces the medium configuration by 4.3 percentage points, while static-merge plus uniform-recovery controls are 4.7-8.9 points below the small configuration. Additional evaluations cover language and multimodal models at their separately reported budgets. The main results assume known task identity; practical routing and accuracy under larger task counts at fixed capacity remain distinct evaluation questions. Code is available at https://github.com/WanyuGroup/NeurIPS2026-DiDi-Merging.

cs.LG↗

arXiv2026

NestedVLA: Learning to Consolidate and Generate Skills for Vision-Language-Action Models

Vision-language-action (VLA) models can learn a wide range of robotic skills, but combining separately trained skills remains challenging. The original training data may be unavailable for joint retraining, while directly merging model weights can cause interference. We propose NestedVLA, which consolidates independently fine-tuned VLA policies without access to their original training data. Instead of merging all updates into one checkpoint, a hypernetwork embedded in the VLA generates action-expert weight updates for a selected task. Each task's compressed vision-language model update is stored separately, and both updates are added to a frozen base VLA to instantiate the corresponding policy. We evaluate NestedVLA by consolidating three policies fine-tuned with reinforcement learning on MetaWorld, ManiSkill, and CALVIN. It achieves 66.7% average success, improving on the strongest model-merging baseline by 41.1 percentage points. It performs within 1 percentage point of per-task fine-tuning on ManiSkill and substantially narrows the gap on CALVIN. In this three-policy setting, total storage is 1.5x that of a single full checkpoint. Code is available at https://github.com/WanyuGroup/NeurIPS2026-NestedVLA.

cs.RO↗

arXiv2026

Nucleus Speculative Decoding: Plausibility-Aware Verification Beyond Exact Distribution

Speculative decoding (SD) accelerates autoregressive generation by letting a lightweight draft model propose tokens that a target model verifies in parallel. However, its standard acceptance rule, designed for exact distribution correction, rejects tokens to which the draft assigns excess probability, even when the target finds them highly plausible: over half of the rejected tokens lie inside the target's own nucleus. Since verification dominates the cost of SD, these rejections waste its most expensive operation. We introduce Nucleus Speculative Decoding (NSD), a plausibility-aware verification rule that accepts a draft token if it passes the standard test or belongs to the target nucleus, while keeping SD's residual correction and bonus sampling unchanged. NSD requires no training or extra forward passes and is compatible with any drafting mechanism. Because it preserves SD's recovery rule, its deviation from the target distribution can be characterized exactly: we prove that the single-step total variation error equals the draft's excess probability inside the target nucleus, which is precisely NSD's gain in acceptance probability. NSD thus trades fidelity for acceptance at an exact one-to-one rate, and we further derive sequence-level fidelity bounds. Across six target models and five mathematical reasoning and code generation benchmarks, NSD achieves the highest throughput among all compared methods, with speedups of up to $5.16\times$ over autoregressive decoding and $3.15\times$ over standard SD, while improving accuracy over SD on five of the six models and nearly halving verification time per output token. Our code is available at https://github.com/EIT-NLP/Nucleus-Speculative-Decoding.

cs.CL↗

arXiv2026

Evolving Idea Graphs with Learnable Edits-and-Commits for Multi-Agent Scientific Ideation

LLM-empowered multi-agent systems offer new potential to accelerate scientific discovery by generating novel research ideas. However, existing methods typically coordinate agents through temporary texts, such as drafts or chat logs; it is difficult to pinpoint the weaknesses in the generated ideas and how the agents refine them. To this end, we introduce \textbf{Evolving Idea Graphs} (EIG), a graph-based multi-agent scientific ideation framework that can generate high-performance research ideas across various benchmark-native metrics, such as novelty, feasibility, and clarity. Instead of coordinating solely through texts, EIG represents a partially formed proposal as an evolving idea graph, where nodes capture scientific claims and edges encode relations (e.g., support and conflict), enabling unresolved weaknesses to remain identifiable throughout the idea evolving process. Specifically, a learned two-head controller operates over the evolving graph to guide the ideation: one head selects graph edits for agents to execute, while the other decides when the graph is ready for commit as final proposal synthesis. On AI Idea Bench 2025 and LiveIdeaBench, EIG outperforms all compared systems on both automatic benchmark scores and blinded expert ratings. Ablations show that the graph-based state, its typed relations, and the learned edit-and-commit control each contribute to final proposal quality.

cs.MA↗

arXiv2026

HybridFlow: Resource-Adaptive Subtask Routing for Efficient Edge-Cloud LLM Inference

Edge-cloud collaborative inference is becoming a practical necessity for LLM-powered edge devices: on-device models often cannot afford the required reasoning capability, while cloud-only inference could be prohibitively costly and slow under strict latency and token/API budgets. However, existing edge-cloud collaboration methods often route per query or fixed steps simply based-on the estimated difficulty. Such coarse and static heuristics overlook subtask dependencies, missing opportunities for parallel execution and budget-adaptive routing. To this end, we propose \textbf{HybridFlow}, a resource-adaptive edge-cloud inference framework that (i) builds a dependency-aware DAG for each query and executes newly unlocked subtasks in parallel, reducing end-to-end latency; (ii) routes each subtask online to the edge or cloud via a learned benefit--cost utility model that dynamically trades accuracy gains against token/API and latency budgets, thereby reducing unnecessary cloud usage while preserving reasoning quality. Across GPQA, MMLU-Pro, AIME24, and LiveBench-Reasoning, HybridFlow improves the cost-accuracy trade-off, reducing latency and cloud API usage while maintaining competitive accuracy against strong structured reasoning baselines.

cs.DC↗

arXiv2026

REIN: Bridging the Gap between Reasoning and Reliability via Reflection and Abstention Alignment

Large reasoning models (LRMs) are prone to hallucination, which undermines their reliability and poses challenges for safe deployment. Hallucinations in LRMs arise from two distinct failure sources: reasoning hallucination, where flawed inference steps propagate to an incorrect conclusion, and knowledge hallucination, where the model lacks the requisite factual knowledge to answer the query. To address reasoning hallucination, we propose REIN, an alignment framework that trains LRMs to produce a structured reasoning sequence, $\texttt{ } $$\rightarrow$ $\texttt{ } $$\rightarrow$ $\texttt{ }$, enabling explicit self-reflection before committing to a final answer. To address knowledge hallucination, REIN introduces a reward mechanism that encourages explicit abstention (e.g., "I don't know") when none of the sampled reasoning chains yields a correct answer, allowing the model to refrain from unsupported predictions. Extensive evaluations on mathematical and commonsense reasoning benchmarks show that REIN consistently improves selective accuracy, reduces incorrect-but-self-endorsed responses, and maintains high coverage compared with competitive baselines. Notably, REIN achieves these gains within a single forward pass, without requiring process supervision, inference-time controllers, external search, or multi-round critiques. Experiments on multiple backbones show that REIN reduces the hallucination proxy by $58\sim72\%$ relative to the base models while maintaining $86\sim91\%$ average coverage, and improves selective accuracy on attempted questions by $6.6\sim14.2\%$.

cs.AI↗

arXiv2026

BANDMAS: Causality-Inspired Semantic Packet Scheduling for Bandwidth-Efficient Multi-Agent Collaboration

LLM-based multi-agent systems make decisions based on the aggregated information via exchanging messages across specialized agents. Forwarding every generated message among agents increases application-layer traffic. Yet, it introduces tremendous input tokens for agent processing, potentially raising inference latency and computational overhead. Existing approaches attempt to address the above issues by pruning agents or discarding redundant messages. Nevertheless, such agent-level or message-level optimization results in insufficient evidence supporting for final decisions or still containing redundant message transmissions. To address these challenges, we propose BANDMAS, a multi-agent collaboration framework that models inter-agent communications as task-oriented traffic, which enables efficient transmission via causality-inspired replay valuation. Specifically, we decompose messages into several data packets by analyzing their semantic features such as evidence and requests. The system only transmits these packets if their predicted replay-derived contribution exceeds their resource cost. Consequently, BANDMAS is able to adaptively schedule communication packets while adhering to bandwidth, latency, deadline, and receiver context constraints. On frozen Qwen3-4B traffic across SciFact, HotpotQA, and FanOutQA, our framework reduces application-layer bytes by 53.2\% to 77.3\% at selected caps and attains the highest mean task metric among constrained methods on all three workloads.

cs.MA↗

arXiv2026

Adopting Reinforcement Learning with Verifiable Rewards for Molecular Generation

Leveraging large language models (LLMs) for molecular generation has shown remarkable potential in chemical and drug design. Current methods primarily rely on supervised training or fine-tuning with limited datasets, which are insufficient to capture complex molecular design objectives. While some approaches attempt to guide generation toward specific goals, they often lack direct optimization mechanisms, making it difficult to align generated molecules with desired properties. To tackle these challenges, we propose \textbf{LLMol}, a principled reinforcement learning framework that directly incorporates verifiable rewards for targeted molecule generation. The key insight is to formulate molecular design as a goal-conditioned sequence prediction task, where verifiable rewards serve as explicit supervision to drive generation toward desired objectives. LLMol follows a two-stage training paradigm combining supervised learning and reinforcement learning. In the first stage, large language models are supervised fine-tuned to capture chemical syntax and molecular distributions. In the second stage, we introduce Reinforcement Learning with Verifiable Rewards (RLVR), which directly integrates property-based reward signals to guide molecular generation toward task-specific objectives. To address the high variance and instability common in discrete sequence optimization, we adopt Group Relative Policy Optimization (GRPO), a stable on-policy algorithm that smooths reward signals and improves training robustness. This framework enables LLMol to effectively handle a range of molecular design tasks, including single-property targeting (e.g., penalized logP, QED) and structure-constrained optimization. Experimental results demonstrate that LLMol consistently outperforms existing methods, achieving higher success rates and improved efficiency across diverse molecular benchmarks.

cs.LG↗

arXiv2026

Post-Training Shifts Confidence: A Three-Stage Analysis of How SFT, RL, and OPD Shape CoT Calibration

Large language models have made strong reasoning gains through supervised fine-tuning, reinforcement learning, and on-policy distillation, yet these post-training methods are usually evaluated only by final-answer accuracy. We study how they reshape confidence during reasoning. We introduce a three-stage calibration framework that evaluates confidence before, during, and after chain-of-thought generation, corresponding to difficulty estimation, early termination, and answer aggregation. Through a controlled comparison on mathematical reasoning benchmarks, we find that OPD provides the most useful pre-reasoning confidence, SFT gives the strongest online signal for early stopping, and RL produces the most reliable trace-level signal for aggregation. We further show that confidence reliability is position-dependent: RL confidence becomes informative after a path-commitment phase, while OPD confidence is useful early but can become inversely calibrated later. Based on this observation, we propose PosConf, a position-aware confidence strategy that uses confidence only from reliable relative-position intervals. PosConf improves RL answer aggregation by 6.1 points over majority voting and consistently improves OPD early stopping under tight token budgets, with gains up to 4.3 points by avoiding its later inverse-calibration region, showing that \emph{confidence in reasoning models should be used both stage-wise and position-awarely}. Our code is available at https://github.com/EIT-NLP/Post-Training-Calibration.

cs.CL↗

arXiv2026

HPS: Hard Preference Sampling for Human Preference Alignment

Aligning Large Language Model (LLM) responses with human preferences is vital for building safe and controllable AI systems. While preference optimization methods based on Plackett-Luce (PL) and Bradley-Terry (BT) models have shown promise, they face challenges such as poor handling of harmful content, inefficient use of dispreferred responses, and, specifically for PL, high computational costs. To address these issues, we propose Hard Preference Sampling (HPS), a novel framework for robust and efficient human preference alignment. HPS introduces a training loss that prioritizes the most preferred response while rejecting all dispreferred and harmful ones. It emphasizes "hard" dispreferred responses -- those closely resembling preferred ones -- to enhance the model's rejection capabilities. By leveraging a single-sample Monte Carlo sampling strategy, HPS reduces computational overhead while maintaining alignment quality. Theoretically, HPS improves sample efficiency over existing PL methods and maximizes the reward margin between preferred and dispreferred responses, ensuring clearer distinctions. Experiments on HH-RLHF and PKU-Safety datasets validate HPS's effectiveness, achieving comparable BLEU and reward scores while greatly improving reward margins and thus reducing harmful content generation.

cs.AI↗

arXiv2026

Distributional Priors Guided Diffusion for Generating 3D Molecules in Low Data Regimes

Can we train a 3D molecule generator using data from dense regions to generate samples in sparse regions? This challenge can be framed as an out-of-distribution (OOD) generation problem. While prior research on OOD generation predominantly targets property shifts, structural shifts -- such as differences in molecular scaffolds or functional groups -- represent an equally critical source of distributional shifts. This work introduces the Geometric OOD Diffusion Model (GODD), a novel diffusion-based framework that enables training on data-abundant molecular distributions while generalizing to data-scarce distributions under distributional structural shifts. Central to our approach is a designated equivariant asymmetric autoencoder to capture distributional structural priors. The asymmetric design allows the model to generalize to unseen structural variations by capturing distributional priors representing distinct distributions. The encoded structural-grained priors guide generation toward sparse regions without requiring explicit training on such data. Evaluated across standard benchmarks encompassing OOD structural shifts (e.g., scaffolds, rings), GODD achieves an improvement of 12.6% in success rate, defined based on molecular validity, uniqueness, and novelty. Furthermore, the framework demonstrates promising performance and generalization on canonical fragment-based drug design tasks, highlighting its utility in learning-based molecular discovery.

cs.LG↗

arXiv2026

Knowledge Fusion of Large Language Models Via Modular SkillPacks

Cross-capability transfer is a key challenge in large language model (LLM) research, with applications in multi-task integration, model compression, and continual learning. Recent works like FuseLLM and FuseChat have demonstrated the potential of transferring multiple model capabilities to lightweight models, enhancing adaptability and efficiency, which motivates our investigation into more efficient cross-capability transfer methods. However, existing approaches primarily focus on small, homogeneous models, limiting their applicability. For large, heterogeneous models, knowledge distillation with full-parameter fine-tuning often overlooks the student model's intrinsic capacity and risks catastrophic forgetting, while PEFT methods struggle to effectively absorb knowledge from source LLMs. To address these issues, we introduce GraftLLM, a novel method that stores source model capabilities in a target model with SkillPack format. This approach preserves general capabilities, reduces parameter conflicts, and supports forget-free continual learning and model fusion. We employ a module-aware adaptive compression strategy to compress parameter updates, ensuring efficient storage while maintaining task-specific knowledge. The resulting SkillPack serves as a compact and transferable knowledge carrier, ideal for heterogeneous model fusion and continual learning. Experiments across various scenarios demonstrate that GraftLLM outperforms existing techniques in knowledge transfer, knowledge fusion, and forget-free learning, providing a scalable and efficient solution for cross-capability transfer. The code is publicly available at: https://github.com/duguodong7/GraftLLM.

cs.AI↗

arXiv2026

Exploiting Latent Linearity in LLMs Improves Explainable Molecular Representation Learning

Large language models (LLMs) have demonstrated broad utility across molecular domains, spanning drug discovery and materials design. Analyzing LLMs' latent representations is crucial for elucidating their underlying mechanisms, improving explainability, and ultimately advancing downstream performance. We propose MoleX, a simple yet effective framework that decomposes molecular embeddings within LLM representations into a concept-aligned space for explainable molecular representation learning. We further show that these high-dimensional embeddings admit a linear mapping onto chemically consistent concepts. Our analysis suggests that the uncovered linearity aligns with established chemical principles, indicating a mechanistically explainable latent structure in LLM representations for scientific applications. When applied to downstream tasks, this latent linearity improves both predictive and explanatory performance. Extensive experiments demonstrate that MoleX outperforms existing approaches in accuracy, explainability, and efficiency, achieving CPU inference on large-scale datasets 300 times faster with 100,000 fewer parameters than LLMs.

cs.LG↗

arXiv2026

Explainable Molecular Property Prediction: Aligning Chemical Concepts with Predictions via Language Models

Providing explainable molecular property predictions is critical for many scientific domains, such as drug discovery and material science. Though transformer-based language models have shown great potential in accurate molecular property prediction, they neither provide chemically meaningful explanations nor faithfully reveal the molecular structure-property relationships. In this work, we develop a framework for explainable molecular property prediction based on language models, dubbed as Lamole, which can provide chemical concepts-aligned explanations. We take a string-based molecular representation -- Group SELFIES -- as input tokens to pretrain and fine-tune our Lamole, as it provides chemically meaningful semantics. By disentangling the information flows of Lamole, we propose combining self-attention weights and gradients for better quantification of each chemically meaningful substructure's impact on the model's output. To make the explanations more faithfully respect the structure-property relationship, we then carefully craft a marginal loss to explicitly optimize the explanations to be able to align with the chemists' annotations. We bridge the manifold hypothesis with the elaborated marginal loss to prove that the loss can align the explanations with the tangent space of the data manifold, leading to concept-aligned explanations. Experimental results over six mutagenicity datasets and one hepatotoxicity dataset demonstrate Lamole can achieve comparable classification accuracy and boost the explanation accuracy by up to 14.3%, being the state-of-the-art in explainable molecular property prediction.

cs.LG↗