arXiv subjects

Shikun Feng

Publications and source records attributed to Shikun Feng.

At least 19 recordsLinked to original sources

arXiv2026

oHC: Orthogonal Hyper-Connections on SO(4) via Quaternions

Hyper-Connections (HC) replace the single residual stream of a Transformer with $n$ parallel ones, mixing them at every layer with a learned $n \times n$ residual matrix. Leaving that matrix unconstrained places no limit on the factor by which the mixing step rescales the residual streams, and that factor compounds across layers, which destabilizes training. Manifold-constrained Hyper-Connections (mHC) address this by restricting the matrix to the doubly stochastic matrices. That caps the factor at one, so the mixing can no longer amplify any direction, but nothing bounds it from below. We prove that inside this set the mixing step can reduce the norm of the residual streams only by shrinking the differences between the streams, while their mean is left unchanged; and since the reduction accumulates over layers, the streams grow more alike and their diversity is spent with depth. We therefore propose Orthogonal Hyper-Connections (oHC), restricting the residual matrix to the rotation group $SO(n)$, so that the mixing step can neither amplify nor attenuate the residual streams in any direction, which keeps training stable and no longer forces the differences between the streams to contract. Specifically, at the four streams used by recent HC models we parameterize the group in closed form by a pair of unit quaternions, which adds no parameters, replaces the iterative projection with a fixed pattern of signed additions, and can be constructed faster than mHC. We evaluate oHC across a comprehensive set of downstream tasks, where it outperforms the single-stream residual baseline, mHC and iHC, which fixes the residual matrix to the identity.

cs.CL↗

arXiv2026

LiveVVT: High-Fidelity Video Virtual Try-On in Real Time

Diffusion-based Video Virtual Try-On (VVT) achieves high visual fidelity through bidirectional spatio-temporal modeling, but complete-clip dependence incurs prohibitive latency and computational overhead in practical continuous deployment. Naively enforcing causality disrupts pretrained bidirectional priors and substantially degrades synthesis quality. We introduce LiveVVT, a rolling streaming diffusion framework that preserves bounded bidirectional modeling within causal recurrent generation. Within a fixed-size window, LiveVVT jointly denoises multiple video chunks under bounded look-ahead, preserving local bidirectional interactions while emitting one clean chunk per iteration. Beyond the window, two complementary memories sustain long-term consistency: a bounded temporal memory propagates recent dynamics and occlusion context, whereas a persistent global appearance memory, constructed once from the target garment and a frontal try-on keyframe, anchors garment details and dressed appearance throughout the stream. We further introduce a progressive distillation framework integrating bidirectional VVT learning, teacher-trajectory regression for causal few-step adaptation, and Collaborative Matching Distillation, which couples teacher-distribution matching with rolling flow matching on real videos to align optimization with recurrent inference. Experiments on paired and unpaired long-sequence benchmarks demonstrate superior generation quality over similarly sized models, with $26\times$ lower latency and $11\times$ higher throughput, enabling high-fidelity real-time streaming VVT.

cs.CV↗

arXiv2026

MCCE: A Framework for Multi-LLM Collaborative Search in Discrete Spaces with Similarity-Filtered Preference Learning

Multi-objective discrete optimization problems, such as molecular design, pose significant challenges due to their vast and unstructured combinatorial spaces. Traditional evolutionary algorithms often get trapped in local optima, while expert knowledge can provide crucial guidance for accelerating convergence. Large language models (LLMs) offer powerful priors and reasoning ability, making them natural optimizers when expert knowledge matters. However, closed-source LLMs, though strong in exploration, cannot update their parameters and thus cannot internalize experience. Conversely, smaller open models can be continually fine-tuned but lack broad knowledge and reasoning strength. We introduce Multi-LLM Collaborative Co-evolution (MCCE), a hybrid framework that unites a frozen closed-source LLM with a lightweight trainable model. The system maintains a trajectory memory of past search processes; the small model is progressively refined via reinforcement learning, with the two models jointly supporting and complementing each other in global exploration. Unlike model distillation, this process enhances the capabilities of both models through mutual inspiration. Experiments on multi-objective drug design benchmarks show that MCCE achieves state-of-the-art Pareto front quality and consistently outperforms baselines. These results highlight a new paradigm for enabling continual evolution in hybrid LLM systems, combining knowledge-driven exploration with experience-driven learning.

cs.LG↗

arXiv2026

UniVVT: A Unified End-to-End Framework for High-Fidelity Video Virtual Try-on

Video Virtual Try-On (VVT) synthesizes a video of a person wearing a target garment while preserving identity, motion, and scene dynamics. Dominant approaches cast VVT as mask-conditioned video inpainting and rely on separate modules for human parsing, pose estimation, and garment warping. This multi-stage design complicates deployment and, more critically, allows errors in explicit geometric priors to propagate irreversibly into the generated video. We present UniVVT, a unified end-to-end framework that reframes VVT as semantically conditioned video generation, eliminating mask, pose, and warping modules at inference. At its core, a scene-task perceiver built on a Multimodal Large Language Model jointly encodes the source video, target garment, and task instruction into compact, task-aware latent tokens, implicitly capturing what to transfer and where and how to transfer it. A lightweight semantic bridge then aligns these tokens with the conditioning space of a diffusion-based video generator, enabling coherent garment transfer. To robustly couple the heterogeneous components, we devise a three-stage progressive training strategy comprising semantic alignment, joint task adaptation, and flexible-resolution refinement. Extensive experiments demonstrate that UniVVT achieves state-of-the-art performance across multiple benchmarks, validating implicit semantic guidance as a simple and effective alternative to fragile geometric preprocessing for end-to-end virtual try-on.

cs.CV↗

arXiv2026

OPTScientist: Multi-Agent Discovery of Typed Optimizer Programs for Transformer Pretraining

Designing optimizers for modern deep learning remains a challenging scientific problem, requiring the joint consideration of optimization geometry, state dynamics, numerical stability, implementation constraints, and empirical generalization. Existing automated optimizer discovery methods typically search either over unconstrained code spaces or within narrowly parameterized optimizer families. The former is flexible but often produces invalid or uninterpretable programs, while the latter is stable but limits novelty. We introduce OPTScientist, a theory-guided multi-agent framework for optimizer discovery in a typed domain-specific language (DSL). OPTScientist formulates optimizer design as a constrained scientific search process, where candidate updates are expressed through direction, scaling, preconditioning, regularization, state, and grouping modules. Four role agents, Theorist, Designer, Engineer, and Reviewer, collaborate within a single orchestration loop to propose hypotheses, synthesize DSL candidates, compile and evaluate optimizers, and critique results. To overcome the limitations of a fixed search space, OPTScientist combines evolutionary search over optimizer programs with a second-stage mechanism that proposes small DSL extensions when repeated failures reveal representational bottlenecks. Using this framework, we discover RS-MR, a reduced-state matrix optimizer that improves transformer pretraining over strong baselines under our native evaluation protocol. Our results suggest a path toward automated optimizer science grounded in theory, typed programs, compiler validation, and closed-loop experimentation.

cs.AI↗

arXiv2026

WMLLM: Self-Evolving Optimization Agents via Predict-Then-Act World Modeling

Black-box optimization problems remain challenging because of large, weakly structured, and high-dimensional search spaces. Existing methods often suffer from poor sample efficiency because they rely on direct candidate generation or trial-and-error refinement. A natural way to improve search efficiency is to use world modeling, which can help identify promising optimization directions before costly evaluation. Large language models can predict the outcomes of these candidates with nontrivial accuracy because of their implicit knowledge. Motivated by this observation, we propose WMLLM, a self-evolving optimization-agent framework based on predict-then-act world modeling. The agent first predicts promising directions and then acts to generate candidates. Combined with agentic multi-turn refinement, population-based search, and reinforcement learning, WMLLM refines both its implicit world model and its optimization strategy during search. Experiments on black-box optimization tasks, especially multi-objective molecular optimization, show that WMLLM improves sample efficiency and final optimization performance. On the multi-objective molecular optimization benchmark, WMLLM achieves state-of-the-art results under a limited evaluation budget.

cs.LG↗

arXiv2026

CodecCap: High-Fidelity Codec-Inspired Residual Modeling for Dense Video Captioning

Existing video captioning methods struggle to balance visual fidelity and redundancy: holistic captions are compact but lose fine-grained evidence, whereas segment-wise captions improve coverage but introduce heavy redundancy. We propose CodecCap, a codec-inspired framework for high-fidelity dense video captioning. Analogous to video codecs, CodecCap represents videos using keyframe and residual captions. Keyframe captions exhaustively encode stable visual context, while residual captions capture temporally only localized actions, motions and changes. This effectively preserves fine-grained visual evidence while reducing redundant descriptions. To quantify the fidelity of captions, we introduce VidCapQA, a caption-then-QA benchmark with 1,000 questions across 14 capability dimensions. Results on VidCapQA show that captions directly generated by strong VLMs still miss many visual details, highlighting caption representation as a critical bottleneck. Experiments show that CodecCap significantly surpasses direct captioning with the same underlying VLMs, suggesting keyframe-residual captioning a way for high-fidelity video-language supervision. We further use CodecCap to construct CodecVDC-100K, a large-scale dense captioning dataset with anchor, residual, scene-level, and video-level supervision.

cs.CV↗

arXiv2026

LLaVA-OneVision-2: Towards Next-Generation Perceptual Intelligence

We introduce LLaVA-OneVision-2 (LLaVA-OV-2), the most capable vision-language model in the LLaVA-OneVision series to date, achieving superior performance across a broad range of multimodal benchmarks. The model builds on a native OneVision-Encoder and incorporates Windowed Attention for efficient local computation while maintaining native resolution. Its key advance is codec-stream tokenization: it treats compressed video as a continuous bit-cost stream, where bit-cost dynamics determine adaptive temporal groups, and motion-residual cues select salient spatial evidence into compact visual canvases. This allocation concentrates a limited token budget on event-bearing content, enabling more stable long-video token compression than fixed groups of pictures. A shared 3D RoPE further places codec canvases, sampled frames, and images in a unified spatiotemporal coordinate system. Furthermore, we build the LLaVA-OV-2 data and training stack around large-scale open supervision: approximately 8M re-captioned video samples for pretraining, a 4M-sample spatial corpus for fine-tuning. We also introduce JumpScore, a temporal-localization benchmark targeting fine-grained grounding in high-frequency, densely repeated motion, a regime underrepresented by existing video evaluations. A standout capability of LLaVA-OV-2 is its unified perception across video understanding, temporal grounding, spatial grounding, and manipulation-trace reasoning. On JumpScore, LLaVA-OneVision-2-8B reaches 74.9 JumpScore mAP, surpassing Qwen3-VL-8B (30.1) by +44.8 points; under matched visual-token budgets on the same benchmark, codec-stream inputs improve temporal grounding over frame sampling by +9.7 points. Across standard benchmarks, LLaVA-OneVision-2-8B further outperforms Qwen3-VL-8B by +4.3 average points on video tasks, +5.3 on spatial tasks, and +15.6 average J&F on tracking tasks.

cs.CV↗

arXiv2026

OneVision-Encoder: Codec-Aligned Sparsity as a Foundational Principle for Multimodal Intelligence

Hypothesis. Artificial general intelligence is, at its core, a compression problem. Effective compression demands resonance: deep learning scales best when its architecture aligns with the fundamental structure of the data. These are the fundamental principles. Yet, modern vision architectures have strayed from these truths: visual signals are highly redundant, while discriminative information, the surprise, is sparse. Current models process dense pixel grids uniformly, wasting vast compute on static background rather than focusing on the predictive residuals that define motion and meaning. We argue that to solve visual understanding, we must align our architectures with the information-theoretic principles of video, i.e., Codecs. Method. OneVision-Encoder encodes video by compressing predictive visual structure into semantic meaning. By adopting Codec Patchification, OV-Encoder abandons uniform computation to focus exclusively on the 3.1%-25% of regions rich in signal entropy. To unify spatial and temporal reasoning under irregular token layouts, OneVision-Encoder employs a shared 3D RoPE and is trained with a large-scale cluster discrimination objective over more than one million semantic concepts, jointly capturing object permanence and motion dynamics. Evidence. The results validate our core hypothesis: efficiency and accuracy are not a trade-off; they are positively correlated. When integrated into LLM, it consistently outperforms strong vision backbones such as Qwen3-ViT and SigLIP2 across 16 image, video, and document understanding benchmarks, despite using substantially fewer visual tokens and pretraining data. Notably, on video understanding tasks, OV-Encoder achieves an average improvement of 4.1% over Qwen3-ViT. Codec-aligned, patch-level sparsity is a foundational principle, enabling OV-Encoder as a scalable engine for next-generation visual generalists.

cs.CV↗

arXiv2026

Eureka-Audio: Triggering Audio Intelligence in Compact Language Models

We present Eureka-Audio, a compact yet high-performance audio language model that achieves competitive performance against models that are 4 to 18 times larger across a broad range of audio understanding benchmarks. Despite containing only 1.7B parameters, Eureka-Audio demonstrates strong performance on automatic speech recognition (ASR), audio understanding, and dense audio captioning, matching or surpassing multiple 7B to 30B audio and omni-modal baselines. The model adopts a unified end-to-end architecture composed of a lightweight language backbone, a Whisper-based audio encoder, and a sparsely activated Mixture-of-Experts (MoE) adapter that explicitly accounts for audio heterogeneity and alleviates cross-modal optimization conflicts under limited capacity. To further enhance paralinguistic reasoning, we introduce DataFlux, a closed loop audio instruction data synthesis and verification pipeline that constructs high quality, logically consistent supervision from raw audio. Extensive evaluations across ASR, knowledge reasoning, safety, instruction following, and paralinguistic benchmarks, demonstrate that Eureka-Audio achieves an efficient balance between computational cost and performance. These results establish Eureka Audio as a strong and practical baseline for lightweight audio understanding models.

cs.SD↗

arXiv2026

ERNIE 5.0 Technical Report

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

cs.CL↗

arXiv2026

CORD: Bridging the Audio-Text Reasoning Gap via Weighted On-policy Cross-modal Distillation

Large Audio Language Models (LALMs) have garnered significant research interest. Despite being built upon text-based large language models (LLMs), LALMs frequently exhibit a degradation in knowledge and reasoning capabilities. We hypothesize that this limitation stems from the failure of current training paradigms to effectively bridge the acoustic-semantic gap within the feature representation space. To address this challenge, we propose CORD, a unified alignment framework that performs online cross-modal self-distillation. Specifically, it aligns audio-conditioned reasoning with its text-conditioned counterpart within a unified model. Leveraging the text modality as an internal teacher, CORD performs multi-granularity alignment throughout the audio rollout process. At the token level, it employs on-policy reverse KL divergence with importance-aware weighting to prioritize early and semantically critical tokens. At the sequence level, CORD introduces a judge-based global reward to optimize complete reasoning trajectories via Group Relative Policy Optimization (GRPO). Empirical results across multiple benchmarks demonstrate that CORD consistently enhances audio-conditioned reasoning and substantially bridges the audio-text performance gap with only 80k synthetic training samples, validating the efficacy and data efficiency of our on-policy, multi-level cross-modal alignment approach.

cs.SD↗

arXiv2026

MoE Adapter for Large Audio Language Models: Sparsity, Disentanglement, and Gradient-Conflict-Free

Extending the input modality of Large Language Models~(LLMs) to the audio domain is essential for achieving comprehensive multimodal perception. However, it is well-known that acoustic information is intrinsically \textit{heterogeneous}, entangling attributes such as speech, music, and environmental context. Existing research is limited to a dense, parameter-shared adapter to model these diverse patterns, which induces \textit{gradient conflict} during optimization, as parameter updates required for distinct attributes contradict each other. To address this limitation, we introduce the \textit{\textbf{MoE-Adapter}}, a sparse Mixture-of-Experts~(MoE) architecture designed to decouple acoustic information. Specifically, it employs a dynamic gating mechanism that routes audio tokens to specialized experts capturing complementary feature subspaces while retaining shared experts for global context, thereby mitigating gradient conflicts and enabling fine-grained feature learning. Comprehensive experiments show that the MoE-Adapter achieves superior performance on both audio semantic and paralinguistic tasks, consistently outperforming dense linear baselines with comparable computational costs. Furthermore, we will release the related code and models to facilitate future research.

cs.SD↗

arXiv2025

MatryoshkaThinking: Recursive Test-Time Scaling Enables Efficient Reasoning

Test-time scaling has emerged as a promising paradigm in language modeling, wherein additional computational resources are allocated during inference to enhance model performance. Recent approaches, such as DeepConf, have demonstrated the efficacy of this strategy, however, they often incur substantial computational overhead to achieve competitive results. In this work, we propose MatryoshkaThinking, a novel method that significantly reduces computational cost while maintaining state-of-the-art performance. Specifically, MatryoshkaThinking attains a score of 99.79 on AIME2025 using only 4% of the computation required by DeepConf. The core of our approach lies in the recursive exploitation of the model's intrinsic capabilities in reasoning, verification, and summarization, which collectively enhance the retention of correct solutions and reduce the disparity between Pass@k and Pass@1. Comprehensive evaluations across multiple open-source models and challenging multi-modal reasoning benchmarks validate the effectiveness and generality of our method. These findings offer new insights into the design of efficient and scalable test-time inference strategies for advanced language models.

cs.CL↗

arXiv2025

Revisiting Sampling Strategies for Molecular Generation

Sampling strategies in diffusion models are critical to molecular generation yet remain relatively underexplored. In this work, we investigate a broad spectrum of sampling methods beyond conventional defaults and reveal that sampling choice substantially affects molecular generation performance. In particular, we identify a maximally stochastic sampling (StoMax), a simple yet underexplored strategy, as consistently outperforming default sampling methods for generative models DDPM and BFN. Our findings highlight the pivotal role of sampling design and suggest promising directions for advancing molecular generation through principled and more expressive sampling approaches.

physics.chem-ph↗

arXiv2025

Straight-Line Diffusion Model for Efficient 3D Molecular Generation

Diffusion-based models have shown great promise in molecular generation but often require a large number of sampling steps to generate valid samples. In this paper, we introduce a novel Straight-Line Diffusion Model (SLDM) to tackle this problem, by formulating the diffusion process to follow a linear trajectory. The proposed process aligns well with the noise sensitivity characteristic of molecular structures and uniformly distributes reconstruction effort across the generative process, thus enhancing learning efficiency and efficacy. Consequently, SLDM achieves state-of-the-art performance on 3D molecule generation benchmarks, delivering a 100-fold improvement in sampling efficiency.

cs.LG↗

arXiv2025

UniGEM: A Unified Approach to Generation and Property Prediction for Molecules

Molecular generation and molecular property prediction are both crucial for drug discovery, but they are often developed independently. Inspired by recent studies, which demonstrate that diffusion model, a prominent generative approach, can learn meaningful data representations that enhance predictive tasks, we explore the potential for developing a unified generative model in the molecular domain that effectively addresses both molecular generation and property prediction tasks. However, the integration of these tasks is challenging due to inherent inconsistencies, making simple multi-task learning ineffective. To address this, we propose UniGEM, the first unified model to successfully integrate molecular generation and property prediction, delivering superior performance in both tasks. Our key innovation lies in a novel two-phase generative process, where predictive tasks are activated in the later stages, after the molecular scaffold is formed. We further enhance task balance through innovative training strategies. Rigorous theoretical analysis and comprehensive experiments demonstrate our significant improvements in both tasks. The principles behind UniGEM hold promise for broader applications, including natural language processing and computer vision.

cs.LG↗

arXiv2024

UniMAP: Universal SMILES-Graph Representation Learning

Molecular representation learning is fundamental for many drug related applications. Most existing molecular pre-training models are limited in using single molecular modality, either SMILES or graph representation. To effectively leverage both modalities, we argue that it is critical to capture the fine-grained 'semantics' between SMILES and graph, because subtle sequence/graph differences may lead to contrary molecular properties. In this paper, we propose a universal SMILE-graph representation learning model, namely UniMAP. Firstly, an embedding layer is employed to obtain the token and node/edge representation in SMILES and graph, respectively. A multi-layer Transformer is then utilized to conduct deep cross-modality fusion. Specially, four kinds of pre-training tasks are designed for UniMAP, including Multi-Level Cross-Modality Masking (CMM), SMILES-Graph Matching (SGM), Fragment-Level Alignment (FLA), and Domain Knowledge Learning (DKL). In this way, both global (i.e. SGM and DKL) and local (i.e. CMM and FLA) alignments are integrated to achieve comprehensive cross-modality fusion. We evaluate UniMAP on various downstream tasks, i.e. molecular property prediction, drug-target affinity prediction and drug-drug interaction. Experimental results show that UniMAP outperforms current state-of-the-art pre-training methods.We also visualize the learned representations to demonstrate the effect of multi-modality integration.

cs.LG↗