arXiv subjects

Yao Lu

Publications and source records attributed to Yao Lu.

At least 109 records · Page 6Linked to original sources

arXiv2025

Federated Dialogue-Semantic Diffusion for Emotion Recognition under Incomplete Modalities

Multimodal Emotion Recognition in Conversations (MERC) enhances emotional understanding through the fusion of multimodal signals. However, unpredictable modality absence in real-world scenarios significantly degrades the performance of existing methods. Conventional missing-modality recovery approaches, which depend on training with complete multimodal data, often suffer from semantic distortion under extreme data distributions, such as fixed-modality absence. To address this, we propose the Federated Dialogue-guided and Semantic-Consistent Diffusion (FedDISC) framework, pioneering the integration of federated learning into missing-modality recovery. By federated aggregation of modality-specific diffusion models trained on clients and broadcasting them to clients missing corresponding modalities, FedDISC overcomes single-client reliance on modality completeness. Additionally, the DISC-Diffusion module ensures consistency in context, speaker identity, and semantics between recovered and available modalities, using a Dialogue Graph Network to capture conversational dependencies and a Semantic Conditioning Network to enforce semantic alignment. We further introduce a novel Alternating Frozen Aggregation strategy, which cyclically freezes recovery and classifier modules to facilitate collaborative optimization. Extensive experiments on the IEMOCAP, CMUMOSI, and CMUMOSEI datasets demonstrate that FedDISC achieves superior emotion classification performance across diverse missing modality patterns, outperforming existing approaches.

cs.CV↗

arXiv2025

A Multi-Stage Framework with Taxonomy-Guided Reasoning for Occupation Classification Using Large Language Models

Automatically annotating job data with standardized occupations from taxonomies, known as occupation classification, is crucial for labor market analysis. However, this task is often hindered by data scarcity and the challenges of manual annotations. While large language models (LLMs) hold promise due to their extensive world knowledge and in-context learning capabilities, their effectiveness depends on their knowledge of occupational taxonomies, which remains unclear. In this study, we assess the ability of LLMs to generate precise taxonomic entities from taxonomy, highlighting their limitations, especially for smaller models. To address these challenges, we propose a multi-stage framework consisting of inference, retrieval, and reranking stages, which integrates taxonomy-guided reasoning examples to enhance performance by aligning outputs with taxonomic knowledge. Evaluations on a large-scale dataset show that our framework not only enhances occupation and skill classification tasks, but also provides a cost-effective alternative to frontier models like GPT-4o, significantly reducing computational costs while maintaining strong performance. This makes it a practical and scalable solution for occupation classification and related tasks across LLMs.

cs.CL↗

arXiv2025

FlowMesh: A Service Fabric for Composable LLM Workflows

AI deployment increasingly resembles a pipeline of data transformation, fine-tuning, and agent interactions rather than a monolithic LLM job; recent examples include RLHF/RLAIF training and agentic workflows. To cope with this shift, we propose FlowMesh, a multi-tenant service fabric that executes and optimizes these workloads as one shared service instead of isolated pipelines. It decomposes workflows into fine-grained operators with recorded lineage, enabling de-duplication of work across users and batching requests on the same hardware while preserving per-workflow provenance. A global control plane maintains a cluster-wide pool of ready operators and uses a single utility function to pick both the batch and the worker, balancing throughput, cost, and data locality on heterogeneous GPUs. The data plane is an elastic fleet of stateless workers backed by a content-addressable store, enabling rapid, automatic scale-out, safe retry after preemption, and portability across managed clusters such as Kubernetes and geo-distributed GPU marketplaces such as Vast.ai. Compared with baseline solutions, FlowMesh achieves up to 3.8x cost reduction and 2.0x lower energy usage, provides a similar or better latency profile, and remains efficient under dynamic and failure-prone conditions.

cs.DC↗

arXiv2025

OmniVinci: Enhancing Architecture and Data for Omni-Modal Understanding LLM

Advancing machine intelligence requires developing the ability to perceive across multiple modalities, much as humans sense the world. We introduce OmniVinci, an initiative to build a strong, open-source, omni-modal LLM. We carefully study the design choices across model architecture and data curation. For model architecture, we present three key innovations: (i) OmniAlignNet for strengthening alignment between vision and audio embeddings in a shared omni-modal latent space; (ii) Temporal Embedding Grouping for capturing relative temporal alignment between vision and audio signals; and (iii) Constrained Rotary Time Embedding for encoding absolute temporal information in omni-modal embeddings. We introduce a curation and synthesis pipeline that generates 24M single-modal and omni-modal conversations. We find that modalities reinforce one another in both perception and reasoning. Our model, OmniVinci, outperforms Qwen2.5-Omni with +19.05 on DailyOmni (cross-modal understanding), +1.7 on MMAR (audio), and +3.9 on Video-MME (vision), while using just 0.2T training tokens - a 6 times reduction compared to Qwen2.5-Omni's 1.2T. We finally demonstrate omni-modal advantages in downstream applications spanning robotics, medical AI, and smart factory.

cs.CV↗

arXiv2025

The Structural Scalpel: Automated Contiguous Layer Pruning for Large Language Models

Although large language models (LLMs) have achieved revolutionary breakthroughs in many fields, their large model size and high computational cost pose significant challenges for practical deployment on resource-constrained edge devices. To this end, layer pruning has been proposed to reduce the computational overhead by directly removing redundant layers. However, existing layer pruning methods typically rely on hand-crafted metrics to evaluate and remove individual layers, while ignoring the dependencies between layers. This can disrupt the model's information flow and severely degrade performance. To address these issues, we propose CLP, a novel continuous layer pruning framework that introduces two key innovations: a differentiable concave gate algorithm that automatically identifies the best continuous layer segments for pruning via gradient-based optimization; and a cutoff endpoint tuning strategy that effectively restores model performance by fine-tuning only the layers adjacent to the pruned segments. Extensive experiments across multiple model architectures (including LLaMA2, LLaMA3 and Qwen) and sizes (from $7$B to $70$B parameters) show that CLP significantly outperforms existing state-of-the-art baselines. For example, at a pruning rate of $20\%$, CLP achieves an average performance retention of $95.34\%$ on LLaMA3-70B, outperforming baselines by $4.29\%$-$30.52\%$. Furthermore, CLP can be seamlessly combined with quantization to further compress the model with only a slight performance loss.

cs.LG↗

arXiv2025

BDiff: Block-aware and Accurate Text-based Code Differencing

Code differencing is a fundamental technique in software engineering practice and research. While researchers have proposed text-based differencing techniques capable of identifying line changes over the past decade, existing methods exhibit a notable limitation in identifying edit actions (EAs) that operate on text blocks spanning multiple lines. Such EAs are common in developers' practice, such as moving a code block for conditional branching or duplicating a method definition block for overloading. Existing tools represent such block-level operations as discrete sequences of line-level EAs, compelling developers to manually correlate them and thereby substantially impeding the efficiency of change comprehension. To address this issue, we propose BDiff, a text-based differencing algorithm capable of identifying two types of block-level EAs and five types of line-level EAs. Building on traditional differencing algorithms, we first construct a candidate set containing all possible line mappings and block mappings. Leveraging the Kuhn-Munkres algorithm, we then compute the optimal mapping set that can minimize the size of the edit script (ES) while closely aligning with the original developer's intent. To validate the effectiveness of BDiff, we selected five state-of-the-art tools, including large language models (LLMs), as baselines and adopted a combined qualitative and quantitative approach to evaluate their performance in terms of ES size, result quality, and running time. Experimental results show that BDiff produces higher-quality differencing results than baseline tools while maintaining competitive runtime performance. Our experiments also show the unreliability of LLMs in code differencing tasks regarding result quality and their infeasibility in terms of runtime efficiency. We have implemented a web-based visual differencing tool.

cs.SE↗

arXiv2025

SparseVILA: Decoupling Visual Sparsity for Efficient VLM Inference

Vision Language Models (VLMs) have rapidly advanced in integrating visual and textual reasoning, powering applications across high-resolution image understanding, long-video analysis, and multi-turn conversation. However, their scalability remains limited by the growing number of visual tokens that dominate inference latency. We present SparseVILA, a new paradigm for efficient VLM inference that decouples visual sparsity across the prefilling and decoding stages. SparseVILA distributes sparsity across stages by pruning redundant visual tokens during prefill and retrieving only query-relevant tokens during decoding. This decoupled design matches leading prefill pruning methods while preserving multi-turn fidelity by retaining most of the visual cache so that query-aware tokens can be retrieved at each conversation round. Built on an AWQ-optimized inference pipeline, SparseVILA achieves up to 4.0 times faster prefilling, 2.5 times faster decoding, and an overall 2.6 times end-to-end speedup on long-context video tasks -- while improving accuracy on document-understanding and reasoning tasks. By decoupling query-agnostic pruning and query-aware retrieval, SparseVILA establishes a new direction for efficient multimodal inference, offering a training-free, architecture-agnostic framework for accelerating large VLMs without sacrificing capability.

cs.CV↗

arXiv2025

LongLive: Real-time Interactive Long Video Generation

We present LongLive, a frame-level autoregressive (AR) framework for real-time and interactive long video generation. Long video generation presents challenges in both efficiency and quality. Diffusion and Diffusion-Forcing models can produce high-quality videos but suffer from low efficiency due to bidirectional attention. Causal attention AR models support KV caching for faster inference, but often degrade in quality on long videos due to memory challenges during long-video training. In addition, beyond static prompt-based generation, interactive capabilities, such as streaming prompt inputs, are critical for dynamic content creation, enabling users to guide narratives in real time. This interactive requirement significantly increases complexity, especially in ensuring visual consistency and semantic coherence during prompt transitions. To address these challenges, LongLive adopts a causal, frame-level AR design that integrates a KV-recache mechanism that refreshes cached states with new prompts for smooth, adherent switches; streaming long tuning to enable long video training and to align training and inference (train-long-test-long); and short window attention paired with a frame-level attention sink, shorten as frame sink, preserving long-range consistency while enabling faster generation. With these key designs, LongLive fine-tunes a 1.3B-parameter short-clip model to minute-long generation in just 32 GPU-days. At inference, LongLive sustains 20.7 FPS on a single NVIDIA H100, achieves strong performance on VBench in both short and long videos. LongLive supports up to 240-second videos on a single H100 GPU. LongLive further supports INT8-quantized inference with only marginal quality loss.

cs.CV↗

arXiv2025

QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs

We propose QeRL, a Quantization-enhanced Reinforcement Learning framework for large language models (LLMs). While RL is essential for LLMs' reasoning capabilities, it is resource-intensive, requiring substantial GPU memory and long rollout durations. QeRL addresses these issues by combining NVFP4 quantization with Low-Rank Adaptation (LoRA), accelerating rollout phase of RL while reducing memory overhead. Beyond efficiency, our findings show that quantization noise increases policy entropy, enhancing exploration, and enabling the discovery of better strategies during RL. To further optimize exploration, QeRL introduces an Adaptive Quantization Noise (AQN) mechanism, which dynamically adjusts noise during training. Experiments demonstrate that QeRL delivers over 1.5 times speedup in the rollout phase. Moreover, this is the first framework to enable RL training of a 32B LLM on a single H100 80GB GPU, while delivering overall speedups for RL training. It also achieves faster reward growth and higher final accuracy than 16-bit LoRA and QLoRA, while matching the performance of full-parameter fine-tuning on mathematical benchmarks such as GSM8K (90.8%) and MATH 500 (77.4%) in the 7B model. These results establish QeRL as an efficient and effective framework for RL training in LLMs.

cs.LG↗

arXiv2025

Few-shot Molecular Property Prediction: A Survey

AI-assisted molecular property prediction has become a promising technique in early-stage drug discovery and materials design in recent years. However, due to high-cost and complex wet-lab experiments, real-world molecules usually experience the issue of scarce annotations, leading to limited labeled data for effective supervised AI model learning. In light of this, few-shot molecular property prediction (FSMPP) has emerged as an expressive paradigm that enables learning from only a few labeled examples. Despite rapidly growing attention, existing FSMPP studies remain fragmented, without a coherent framework to capture methodological advances and domain-specific challenges. In this work, we present the first comprehensive and systematic survey of few-shot molecular property prediction. We begin by analyzing the few-shot phenomenon in molecular datasets and highlighting two core challenges: (1) cross-property generalization under distribution shifts, where each task corresponding to each property, may follow a different data distribution or even be inherently weakly related to others from a biochemical perspective, requiring the model to transfer knowledge across heterogeneous prediction tasks, and (2) cross-molecule generalization under structural heterogeneity, where molecules involved in different or same properties may exhibit significant structural diversity, making model difficult to achieve generalization. Then, we introduce a unified taxonomy that organizes existing methods into data, model, and learning paradigm levels, reflecting their strategies for extracting knowledge from scarce supervision in few-shot molecular property prediction. Next, we compare representative methods, summarize benchmark datasets and evaluation protocols. In the end, we identify key trends and future directions for advancing the continued research on FSMPP.

cs.CE↗

arXiv2025

A Set of Generalized Components to Achieve Effective Poison-only Clean-label Backdoor Attacks with Collaborative Sample Selection and Triggers

Poison-only Clean-label Backdoor Attacks aim to covertly inject attacker-desired behavior into DNNs by merely poisoning the dataset without changing the labels. To effectively implant a backdoor, multiple \textbf{triggers} are proposed for various attack requirements of Attack Success Rate (ASR) and stealthiness. Additionally, sample selection enhances clean-label backdoor attacks' ASR by meticulously selecting ``hard'' samples instead of random samples to poison. Current methods 1) usually handle the sample selection and triggers in isolation, leading to severely limited improvements on both ASR and stealthiness. Consequently, attacks exhibit unsatisfactory performance on evaluation metrics when converted to PCBAs via a mere stacking of methods. Therefore, we seek to explore the bidirectional collaborative relations between the sample selection and triggers to address the above dilemma. 2) Since the strong specificity within triggers, the simple combination of sample selection and triggers fails to substantially enhance both evaluation metrics, with generalization preserved among various attacks. Therefore, we seek to propose a set of components to significantly improve both stealthiness and ASR based on the commonalities of attacks. Specifically, Component A ascertains two critical selection factors, and then makes them an appropriate combination based on the trigger scale to select more reasonable ``hard'' samples for improving ASR. Component B is proposed to select samples with similarities to relevant trigger implanted samples to promote stealthiness. Component C reassigns trigger poisoning intensity on RGB colors through distinct sensitivity of the human visual system to RGB for higher ASR, with stealthiness ensured by sample selection, including Component B. Furthermore, all components can be strategically integrated into diverse PCBAs.

cs.CR↗

arXiv2025

Diffusion Low Rank Hybrid Reconstruction for Sparse View Medical Imaging

This work presents TV-LoRA, a novel method for low-dose sparse-view CT reconstruction that combines a diffusion generative prior (NCSN++ with SDE modeling) and multi-regularization constraints, including anisotropic TV and nuclear norm (LoRA), within an ADMM framework. To address ill-posedness and texture loss under extremely sparse views, TV-LoRA integrates generative and physical constraints, and utilizes a 2D slice-based strategy with FFT acceleration and tensor-parallel optimization for efficient inference. Experiments on AAPM-2016, CTHD, and LIDC datasets with $N_{\mathrm{view}}=8,4,2$ show that TV-LoRA consistently surpasses benchmarks in SSIM, texture recovery, edge clarity, and artifact suppression, demonstrating strong robustness and generalizability. Ablation studies confirm the complementary effects of LoRA regularization and diffusion priors, while the FFT-PCG module provides a speedup. Overall, Diffusion + TV-LoRA achieves high-fidelity, efficient 3D CT reconstruction and broad clinical applicability in low-dose, sparse-sampling scenarios.

cs.CV↗

arXiv2025

Reward Model Routing in Alignment

Reinforcement learning from human or AI feedback (RLHF / RLAIF) has become the standard paradigm for aligning large language models (LLMs). However, most pipelines rely on a single reward model (RM), limiting alignment quality and risking overfitting. Recent work explores RM routing--dynamically selecting an RM from a candidate pool to exploit complementary strengths while maintaining $O(1)$ RM calls--but existing methods suffer from cold-start and insufficient exploration. We propose BayesianRouter, a hybrid routing framework that combines offline RM strengths learning with online Bayesian selection. In the offline stage, a multi-task router is trained on preference data to estimate per-RM reliability. In the online stage, a Bayesian Thompson sampling router performs per-query RM selection, initializing RM-specific weight vectors with offline embeddings as Gaussian priors and adaptively updating their posteriors with online rewards to adapt to the evolving policy distribution. Extensive experiments on instruction-following (AlpacaEval-2, Arena-Hard, MT-Bench) and reasoning (GSM8K, MMLU) benchmarks show that BayesianRouter consistently outperforms individual RMs, RM ensembling, and existing routing methods.

cs.AI↗

arXiv2025

Drawing Conclusions from Draws: Rethinking Preference Semantics in Arena-Style LLM Evaluation

In arena-style evaluation of large language models (LLMs), two LLMs respond to a user query, and the user chooses the winning response or deems the "battle" a draw, resulting in an adjustment to the ratings of both models. The prevailing approach for modeling these rating dynamics is to view battles as two-player game matches, as in chess, and apply the Elo rating system and its derivatives. In this paper, we critically examine this paradigm. Specifically, we question whether a draw genuinely means that the two models are equal and hence whether their ratings should be equalized. Instead, we conjecture that draws are more indicative of query difficulty: if the query is too easy, then both models are more likely to succeed equally. On three real-world arena datasets, we show that ignoring rating updates for draws yields a 1-3% relative increase in battle outcome prediction accuracy (which includes draws) for all four rating systems studied. Further analyses suggest that draws occur more for queries rated as very easy and those as highly objective, with risk ratios of 1.37 and 1.35, respectively. We recommend future rating systems to reconsider existing draw semantics and to account for query properties in rating updates.

cs.CL↗

arXiv2025

Floquet-engineered fast SNAP gates in weakly coupled circuit-QED systems

Superconducting cavities with high quality factors, coupled to a fixed-frequency transmon, provide a state-of-the-art platform for quantum information storage and manipulation. The commonly used selective number-dependent arbitrary phase (SNAP) gate faces significant challenges in ultra-high-coherence cavities, where the weak dispersive shifts necessary for preserving high coherence typically result in prolonged gate times. Here, we propose a protocol to achieve high-fidelity SNAP gates that are orders of magnitude faster than the standard implementation, surpassing the speed limit set by the bare dispersive shift. We achieve this enhancement by dynamically amplifying the dispersive coupling via sideband interactions, followed by quantum optimal control on the Floquet-engineered system. We also present a unified perturbation theory that explains both the gate acceleration and the associated benign drive-induced decoherence, corroborated by Floquet-Markov simulations. These results pave the way for the experimental realization of high-fidelity, selective control of weakly coupled, high-coherence cavities, and expanding the scope of optimal control techniques to a broader class of Floquet quantum systems.

quant-ph↗

arXiv2025

Scaling RL to Long Videos

We introduce a full-stack framework that scales up reasoning in vision-language models (VLMs) to long videos, leveraging reinforcement learning. We address the unique challenges of long video reasoning by integrating three critical components: (1) a large-scale dataset, LongVideo-Reason, comprising 104K long video QA pairs with high-quality reasoning annotations across diverse domains such as sports, games, and vlogs; (2) a two-stage training pipeline that extends VLMs with chain-of-thought supervised fine-tuning (CoT-SFT) and reinforcement learning (RL); and (3) a training infrastructure for long video RL, named Multi-modal Reinforcement Sequence Parallelism (MR-SP), which incorporates sequence parallelism and a vLLM-based engine tailored for long video, using cached video embeddings for efficient rollout and prefilling. In our experiments, LongVILA-R1-7B achieves strong performance on video benchmarks, reaching 65.1% and 71.1% accuracy on VideoMME without and with subtitles, respectively, and consistently outperforming LongVILA-7B across multiple benchmarks. Moreover, LongVILA-R1-7B supports processing up to 8,192 video frames per video, and configurable FPS settings. Notably, our MR-SP system achieves up to 2.1x speedup on long video RL training. In addition, we release our training system for public availability that supports RL training on various modalities (video, text, and audio), various models (VILA and Qwen series), and even image and video generation models. On a single A100 node (8 GPUs), it supports RL training on hour-long videos (e.g., 3,600 frames).

cs.CV↗

arXiv2025

STORM: Token-Efficient Long Video Understanding for Multimodal LLMs

Recent advances in video-based multimodal large language models (Video-LLMs) have significantly improved video understanding by processing videos as sequences of image frames. However, many existing methods treat frames independently in the vision backbone, lacking explicit temporal modeling, which limits their ability to capture dynamic patterns and efficiently handle long videos. To address these limitations, we introduce STORM (Spatiotemporal TOken Reduction for Multimodal LLMs), a novel architecture incorporating a dedicated temporal encoder between the image encoder and the LLM. Our temporal encoder leverages the Mamba State Space Model to integrate temporal information into image tokens, generating enriched representations that preserve inter-frame dynamics across the entire video sequence. This enriched encoding not only enhances video reasoning capabilities but also enables effective token reduction strategies, including test-time sampling and training-based temporal and spatial pooling, substantially reducing computational demands on the LLM without sacrificing key temporal information. By integrating these techniques, our approach simultaneously reduces training and inference latency while improving performance, enabling efficient and robust video understanding over extended temporal contexts. Extensive evaluations show that STORM achieves state-of-the-art results across various long video understanding benchmarks (more than 5% improvement on MLVU and LongVideoBench) while reducing the computation costs by up to $8\times$ and the decoding latency by 2.4-2.9$\times$ for the fixed numbers of input frames. Project page is available at https://research.nvidia.com/labs/lpr/storm

cs.CV↗

arXiv2025

DSPC: Dual-Stage Progressive Compression Framework for Efficient Long-Context Reasoning

Large language models (LLMs) have achieved remarkable success in many natural language processing (NLP) tasks. To achieve more accurate output, the prompts used to drive LLMs have become increasingly longer, which incurs higher computational costs. To address this prompt inflation problem, prompt compression has been proposed. However, most existing methods require training a small auxiliary model for compression, incurring a significant amount of additional computation. To avoid this, we propose a two-stage, training-free approach, called Dual-Stage Progressive Compression (DSPC). In the coarse-grained stage, semantic-related sentence filtering removes sentences with low semantic value based on TF-IDF. In the fine-grained stage, token importance is assessed using attention contribution, cross-model loss difference, and positional importance, enabling the pruning of low-utility tokens while preserving semantics. We validate DSPC on LLaMA-3.1-8B-Instruct and GPT-3.5-Turbo under a constrained token budget and observe consistent improvements. For instance, in the FewShot task of the Longbench dataset, DSPC achieves a performance of 49.17 by using only 3x fewer tokens, outperforming the best state-of-the-art baseline LongLLMLingua by 7.76.

cs.CL↗