arXiv subjects

Tianyue Wang

Publications and source records attributed to Tianyue Wang.

13 recordsLinked to original sources

arXiv2026

Routing Before Looking: Query-Adaptive Evidence Acquisition for Long-form Video Understanding

Long-form video understanding remains challenging for video agents due to the mismatch between query demands and evidence acquisition strategies. Although recent planning-before-perception methods outperform query-agnostic pipelines, they often rely on a single dominant strategy, either generation-based strategy or retrieval-based strategy, limiting their ability to handle diverse query demands. We propose Route2Look, a lightweight and model-agnostic framework for query-adaptive evidence acquisition in long-form video understanding. Route2Look operates in a Route-Look-Memorize loop with three tools: Global Browse for holistic context, Temporal Ground for explicit temporal cues, and Semantic Retrieve for semantic search. The core component is a routing policy that dynamically selects evidence acquisition tools based on the query. To build this policy, Route2Look adopts a two-stage design: first distilling the routing skill from differential contrastive analysis between generation-based and retrieval-based trajectories, and then applying the distilled skill with hard routing rules and continue-or-stop criteria during inference. Experiments on challenging long-video benchmarks show that Route2Look achieves state-of-the-art performance while maintaining strong frame efficiency across datasets and query types. Oracle routing analysis further reveals the potential of query-adaptive evidence acquisition for future long-form video understanding.

cs.CV↗

arXiv2026

LUT: Latent Utility Training for Visual Reasoning

Multimodal large language models have advanced visual understanding, yet perception-intensive reasoning remains challenging. Recent latent visual reasoning methods introduce hidden-space computation before answering, but they often rely on costly intermediate supervision, such as bounding boxes, sketches, or interleaved rationales. These strategies focus on how latent states should be shaped, but do not explicitly assess whether the latent is useful for the final answer. We propose LUT, a latent reasoning framework trained with only standard VQA pairs. LUT centers training on Latent Utility at two levels. At the trajectory level, we propose Utility-Aware Latent Distillation SFT, which explores answer-relevant latent trajectories, selects qualified trajectories by their information gain, and distills more reliable and learnable supervision through curriculum learning. At the step level, we propose Latent Attribution Policy Optimization, which uses answer-to-latent attribution to differentially optimize latent steps during reinforcement learning. Experiments on perception-intensive visual reasoning benchmarks show that LUT outperforms previous latent reasoning methods and remains competitive with latent-text interleaved methods with lower annotation cost.

cs.CV↗

arXiv2026

Don't Mix Rewards, Mix Policies: Policy Decomposition and Optimization for Multi-Reward RL

Modern large language models (LLMs) are expected not just to answer correctly, but to adapt their behavior to different human values and use cases. As a result, multi-reward reinforcement learning (RL) has become an increasingly important problem for LLMs, where each reward captures a different aspect of desired behavior. However, optimizing with multiple rewards suffers from a more severe alignment tax issue, where different optimization objectives can trade off or even conflict with each other, leading to unstable and inefficient post-training. In this work, we propose PRISM, a new multi-reward RL framework built upon the idea of policy-space decomposition and composition. Instead of compositing different rewards, PRISM optimizes a set of standalone positive policies and a global negative policy. This alleviates the potential conflict during multi-reward policy optimization, while enabling controllability during inference by flexible policy composition. Experiments on scientific reasoning, tool-use reasoning, and helpfulness-safety alignment show that PRISM consistently outperforms existing multi-reward RL baselines, with extra controllability for inference-time preference control.

cs.AI↗

arXiv2026

ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval

Composed Image Retrieval (CIR) aims to retrieve target images based on a hybrid query comprising a reference image and a modification text. Early dual-tower Vision-Language Models (VLMs) struggle with cross-modality compositional reasoning required for this task. While adapting generative Multimodal Large Language Models (MLLMs) for retrieval offers a promising direction, we identify that this strategy overlooks a fundamental issue: compressing a generative MLLM into a single-embedding discriminative retriever triggers a paradigm conflict, which leads to Capability Degradation - the deterioration of native fine-grained reasoning after retrieval adaptation. To address this challenge, we propose ReCALL, a model-agnostic framework that follows a diagnose-generate-refine pipeline: First, we diagnose cognitive blind spots of the retriever via self-guided informative instance mining. Next, we generate corrective instructions and triplets by prompting the foundation MLLM and conduct quality control with VQA-based consistency filtering. Finally, we refine the retriever through continual training on these triplets with a grouped contrastive scheme, thereby internalizing fine-grained visual-semantic distinctions and realigning the discriminative embedding space of retriever with intrinsic compositional reasoning within the MLLM. Extensive experiments on CIRR and FashionIQ show that ReCALL consistently recalibrates degraded capabilities and achieves state-of-the-art performance. Code is available at https://github.com/RemRico/Recall.

cs.CV↗

arXiv2026

WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval

Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve target images given a multimodal query (comprising a reference image and a modification text), without training on annotated triplets. Existing methods typically convert the multimodal query into a single modality-either as an edited caption for Text-to-Image retrieval (T2I) or as an edited image for Image-to-Image retrieval (I2I). However, each paradigm has inherent limitations: T2I often loses fine-grained visual details, while I2I struggles with complex semantic modifications. To effectively leverage their complementary strengths under diverse query intents, we propose WISER, a training-free framework that unifies T2I and I2I via a "retrieve-verify-refine" pipeline, explicitly modeling intent awareness and uncertainty awareness. Specifically, WISER first performs Wider Search by generating both edited captions and images for parallel retrieval to broaden the candidate pool. Then, it conducts Adaptive Fusion with a verifier to assess retrieval confidence, triggering refinement for uncertain retrievals, and dynamically fusing the dual-path for reliable ones. For uncertain retrievals, WISER generates refinement suggestions through structured self-reflection to guide the next retrieval round toward Deeper Thinking. Extensive experiments demonstrate that WISER significantly outperforms previous methods across multiple benchmarks, achieving relative improvements of 45% on CIRCO (mAP@5) and 57% on CIRR (Recall@1) over existing training-free methods. Notably, it even surpasses many training-dependent methods, highlighting its superiority and generalization under diverse scenarios. Code will be released at https://github.com/Physicsmile/WISER.

cs.CV↗

arXiv2026

TRACE: Task-Adaptive Reasoning and Representation Learning for Universal Multimodal Retrieval

Universal Multimodal Retrieval requires unified embedding models capable of interpreting diverse user intents, ranging from simple keywords to complex compositional instructions. While Multimodal Large Language Models (MLLMs) possess strong reasoning capabilities, prevailing adaptations confine them to static encoders, underutilizing their generative potential. This encoder-only paradigm struggles with complex intents that demand logical deduction rather than superficial pattern matching. To address this, we introduce TRACE (Task-adaptive Reasoning And Compressing Embeddings). TRACE unifies generative reasoning with discriminative representation learning. It first generates a structured Chain-of-Thought (CoT) to explicitly reason about the query, and subsequently compresses this reasoning trace into a compact embedding via a dedicated token. To train this framework, we construct M-BEIR-CoT, a large-scale dataset featuring a difficulty-aware routing strategy. Experiments on the M-BEIR benchmark establish TRACE as the new state-of-the-art. Crucially, TRACE demonstrates a learned implicit routing behavior. It autonomously activates reasoning for complex queries while bypassing it for simpler ones, achieving an optimal balance between retrieval accuracy and inference throughput. Furthermore, by internalizing the deductive process, TRACE exhibits remarkable zero-shot transferability to unseen domains and novel constraints.

cs.CV↗

arXiv2026

Statistical Blendshape Calculation and Analysis for Graphics Applications

With the development of virtualization and AI, real-time facial avatar animation is widely used in entertainment, office, business and other fields. Against this background, blendshapes have become a common industry animation solution because of their relative simplicity and ease of interpretation. Aiming for real-time performance and low computing resource dependence, we independently developed an accurate blendshape prediction system for low-power VR applications using a standard webcam. First, blendshape feature vectors are extracted through affine transformation and segmentation. Through further transformation and regression analysis, we were able to identify models for most blendshapes with significant predictive power. Post-processing was used to further improve response stability, including smoothing filtering and nonlinear transformations to minimize error. Experiments showed the system achieved accuracy similar to ARKit 6. Our model has low sensor/hardware requirements and realtime response with a consistent, accurate and smooth visual experience.

cs.GR↗

arXiv2025

GLip: A Global-Local Integrated Progressive Framework for Robust Visual Speech Recognition

Visual speech recognition (VSR), also known as lip reading, is the task of recognizing speech from silent video. Despite significant advancements in VSR over recent decades, most existing methods pay limited attention to real-world visual challenges such as illumination variations, occlusions, blurring, and pose changes. To address these challenges, we propose GLip, a Global-Local Integrated Progressive framework designed for robust VSR. GLip is built upon two key insights: (i) learning an initial coarse alignment between visual features across varying conditions and corresponding speech content facilitates the subsequent learning of precise visual-to-speech mappings in challenging environments; (ii) under adverse conditions, certain local regions (e.g., non-occluded areas) often exhibit more discriminative cues for lip reading than global features. To this end, GLip introduces a dual-path feature extraction architecture that integrates both global and local features within a two-stage progressive learning framework. In the first stage, the model learns to align both global and local visual features with corresponding acoustic speech units using easily accessible audio-visual data, establishing a coarse yet semantically robust foundation. In the second stage, we introduce a Contextual Enhancement Module (CEM) to dynamically integrate local features with relevant global context across both spatial and temporal dimensions, refining the coarse representations into precise visual-speech mappings. Our framework uniquely exploits discriminative local regions through a progressive learning strategy, demonstrating enhanced robustness against various visual challenges and consistently outperforming existing methods on the LRS2 and LRS3 benchmarks. We further validate its effectiveness on a newly introduced challenging Mandarin dataset.

cs.CV↗

arXiv2025

Realizing Blume-Capel Degrees of Freedom with Toroidal Moments in a Ruby Artificial Spin Ice

Realizing exotic Hamiltonians beyond the Ising model is a key pursuit in experimental statistical physics. One such example is the Blume-Capel model, a three-state spin model, whose phase diagram features a tricritical point where second-order and first-order transition lines converge, leading to a coexistence of paramagnetic, ferromagnetic, and disordered phases. Here, we realize an artificial crystal of single-domain nanomagnets, placed on the links of the Ruby lattice, enabling real-space observation of the Blume-Capel degrees of freedom. These Blume-Capel degrees of freedom are represented by the presence, sign and interactions of the toroidal moments that emerge naturally in plaquettes of nanomagnets in the Ruby artificial spin ice. By precisely tuning the lattice parameters of the Ruby artificial spin ice, we demonstrate control over the two-step ordering process of the toroidal moments, whereby there is a high-temperature crossover from a paramagnetic phase to an intermediate paratoroidic regime, followed by a second-order phase transition to a ferrotoroidic ground state. This sequence of toroidal phases and transitions is accurately captured by the Blume-Capel framework and provides a direct realization of a substantial portion of the phase diagram associated with the model. This establishes a new platform for exploring exotic Hamiltonians in terms of artificial spin ice superstructures, here with groups of nanomagnets forming toroidal moments. The success of this mapping paves the way for an entirely new frontier in artificial spin ice: intentionally engineering lattice designs whose effective Hamiltonians mediate unconventional forms of magnetic order, with new behaviors and functionalities.

cond-mat.mes-hall↗

arXiv2025

AutoLoop: a novel autoregressive deep learning method for protein loop prediction with high accuracy

Protein structure prediction is a critical and longstanding challenge in biology, garnering widespread interest due to its significance in understanding biological processes. A particular area of focus is the prediction of missing loops in proteins, which are vital in determining protein function and activity. To address this challenge, we propose AutoLoop, a novel computational model designed to automatically generate accurate loop backbone conformations that closely resemble their natural structures. AutoLoop employs a bidirectional training approach while merging atom- and residue-level embedding, thus improving robustness and precision. We compared AutoLoop with twelve established methods, including FREAD, NGK, AlphaFold2, and AlphaFold3. AutoLoop consistently outperforms other methods, achieving a median RMSD of 1.12 Angstrom and a 2-Angstrom success rate of 73.23% on the CASP15 dataset, while maintaining strong performance on the HOMSTARD dataset. It demonstrates the best performance across nearly all loop lengths and secondary structural types. Beyond accuracy, AutoLoop is computationally efficient, requiring only 0.10 s per generation. A post-processing module for side-chain packing and energy minimization further improves results slightly, confirming the reliability of the predicted backbone. A case study also highlights AutoLoop's potential for precise predictions based on dominant loop conformations. These advances hold promise for protein engineering and drug discovery.

q-bio.BM↗

arXiv2025

Collaborative Expert LLMs Guided Multi-Objective Molecular Optimization

Molecular optimization is a crucial yet complex and time-intensive process that often acts as a bottleneck for drug development. Traditional methods rely heavily on trial and error, making multi-objective optimization both time-consuming and resource-intensive. Current AI-based methods have shown limited success in handling multi-objective optimization tasks, hampering their practical utilization. To address this challenge, we present MultiMol, a collaborative large language model (LLM) system designed to guide multi-objective molecular optimization. MultiMol comprises two agents, including a data-driven worker agent and a literature-guided research agent. The data-driven worker agent is a large language model being fine-tuned to learn how to generate optimized molecules considering multiple objectives, while the literature-guided research agent is responsible for searching task-related literature to find useful prior knowledge that facilitates identifying the most promising optimized candidates. In evaluations across six multi-objective optimization tasks, MultiMol significantly outperforms existing methods, achieving a 82.30% success rate, in sharp contrast to the 27.50% success rate of current strongest methods. To further validate its practical impact, we tested MultiMol on two real-world challenges. First, we enhanced the selectivity of Xanthine Amine Congener (XAC), a promiscuous ligand that binds both A1R and A2AR, successfully biasing it towards A1R. Second, we improved the bioavailability of Saquinavir, an HIV-1 protease inhibitor with known bioavailability limitations. Overall, these results indicate that MultiMol represents a highly promising approach for multi-objective molecular optimization, holding great potential to accelerate the drug development process and contribute to the advancement of pharmaceutical research.

q-bio.BM↗

arXiv2025

Discovery of novel antimicrobial peptides with notable antibacterial potency by a LLM-based foundation model

Large language models (LLMs) have shown remarkable advancements in chemistry and biomedical research, acting as versatile foundation models for various tasks. We introduce AMP-Designer, an LLM-based approach for swiftly designing novel antimicrobial peptides (AMPs) with desired properties. Within 11 days, AMP-Designer achieved the de novo design of 18 AMPs with broad-spectrum activity against Gram-negative bacteria. In vitro validation revealed a 94.4% success rate, with two candidates demonstrating exceptional antibacterial efficacy, minimal hemotoxicity, stability in human plasma, and low potential to induce resistance, as evidenced by significant bacterial load reduction in murine lung infection experiments. The entire process, from design to validation, concluded in 48 days. AMP-Designer excels in creating AMPs targeting specific strains despite limited data availability, with a top candidate displaying a minimum inhibitory concentration of 2.0 μg/ml against Propionibacterium acnes. Integrating advanced machine learning techniques, AMP-Designer demonstrates remarkable efficiency, paving the way for innovative solutions to antibiotic resistance.

q-bio.BM↗

arXiv2023

Highly accurate and efficient deep learning paradigm for full-atom protein loop modeling with KarmaLoop

Protein loop modeling is the most challenging yet highly non-trivial task in protein structure prediction. Despite recent progress, existing methods including knowledge-based, ab initio, hybrid and deep learning (DL) methods fall significantly short of either atomic accuracy or computational efficiency. Moreover, an overarching focus on backbone atoms has resulted in a dearth of attention given to side-chain conformation, a critical aspect in a host of downstream applications including ligand docking, molecular dynamics simulation and drug design. To overcome these limitations, we present KarmaLoop, a novel paradigm that distinguishes itself as the first DL method centered on full-atom (encompassing both backbone and side-chain heavy atoms) protein loop modeling. Our results demonstrate that KarmaLoop considerably outperforms conventional and DL-based methods of loop modeling in terms of both accuracy and efficiency, with the average RMSD improved by over two-fold compared to the second-best baseline method across different tasks, and manifests at least two orders of magnitude speedup in general. Consequently, our comprehensive evaluations indicate that KarmaLoop provides a state-of-the-art DL solution for protein loop modeling, with the potential to hasten the advancement of protein engineering, antibody-antigen recognition, and drug design.

q-bio.BM↗