SC-LoRA: Balancing Efficient Fine-tuning and Knowledge Preservation via Subspace-Constrained LoRA

Parameter-Efficient Fine-Tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA), are indispensable for efficiently customizing Large Language Models (LLMs). However, vanilla LoRA suffers from slow convergence speed and knowledge forgetting problems. Recent studies have leveraged the power of designed LoRA initialization, to enhance the fine-tuning efficiency, or to preserve knowledge in the pre-trained LLM. However, none of these works can address the two cases at the same time. To this end, we introduce Subspace-Constrained LoRA (SC-LoRA), a novel LoRA initialization framework engineered to navigate the trade-off between efficient fine-tuning and knowledge preservation. We achieve this by constraining the output of trainable LoRA adapters in a low-rank subspace, where the context information of fine-tuning data is most preserved while the context information of preserved knowledge is least retained, in a balanced way. Such constraint enables the trainable weights to primarily focus on the main features of fine-tuning data while avoiding damaging the preserved knowledge features. We provide theoretical analysis on our method, and conduct extensive experiments including safety preservation and world knowledge preservation, on various downstream tasks. In our experiments, SC-LoRA succeeds in delivering superior fine-tuning performance while markedly diminishing knowledge forgetting, surpassing contemporary LoRA initialization methods.

ReinFlow: Fine-tuning Flow Matching Policy with Online Reinforcement Learning

We propose ReinFlow, a simple yet effective online reinforcement learning (RL) framework that fine-tunes a family of flow matching policies for continuous robotic control. Derived from rigorous RL theory, ReinFlow injects learnable noise into a flow policy's deterministic path, converting the flow into a discrete-time Markov Process for exact and straightforward likelihood computation. This conversion facilitates exploration and ensures training stability, enabling ReinFlow to fine-tune diverse flow model variants, including Rectified Flow [35] and Shortcut Models [19], particularly at very few or even one denoising step. We benchmark ReinFlow in representative locomotion and manipulation tasks, including long-horizon planning with visual input and sparse reward. The episode reward of Rectified Flow policies obtained an average net growth of 135.36% after fine-tuning in challenging legged locomotion tasks while saving denoising steps and 82.63% of wall time compared to state-of-the-art diffusion RL fine-tuning method DPPO [43]. The success rate of the Shortcut Model policies in state and visual manipulation tasks achieved an average net increase of 40.34% after fine-tuning with ReinFlow at four or even one denoising step, whose performance is comparable to fine-tuned DDIM policies while saving computation time for an average of 23.20%. Project webpage: https://reinflow.github.io/

Q-VDiT: Towards Accurate Quantization and Distillation of Video-Generation Diffusion Transformers

Diffusion transformers (DiT) have demonstrated exceptional performance in video generation. However, their large number of parameters and high computational complexity limit their deployment on edge devices. Quantization can reduce storage requirements and accelerate inference by lowering the bit-width of model parameters. Yet, existing quantization methods for image generation models do not generalize well to video generation tasks. We identify two primary challenges: the loss of information during quantization and the misalignment between optimization objectives and the unique requirements of video generation. To address these challenges, we present Q-VDiT, a quantization framework specifically designed for video DiT models. From the quantization perspective, we propose the Token-aware Quantization Estimator (TQE), which compensates for quantization errors in both the token and feature dimensions. From the optimization perspective, we introduce Temporal Maintenance Distillation (TMD), which preserves the spatiotemporal correlations between frames and enables the optimization of each frame with respect to the overall video context. Our W3A6 Q-VDiT achieves a scene consistency of 23.40, setting a new benchmark and outperforming current state-of-the-art quantization methods by 1.9$\times$. Code will be available at https://github.com/cantbebetter2/Q-VDiT.

IRS: Incremental Relationship-guided Segmentation for Digital Pathology

Continual learning is rapidly emerging as a key focus in computer vision, aiming to develop AI systems capable of continuous improvement, thereby enhancing their value and practicality in diverse real-world applications. In healthcare, continual learning holds great promise for continuously acquired digital pathology data, which is collected in hospitals on a daily basis. However, panoramic segmentation on digital whole slide images (WSIs) presents significant challenges, as it is often infeasible to obtain comprehensive annotations for all potential objects, spanning from coarse structures (e.g., regions and unit objects) to fine structures (e.g., cells). This results in temporally and partially annotated data, posing a major challenge in developing a holistic segmentation framework. Moreover, an ideal segmentation model should incorporate new phenotypes, unseen diseases, and diverse populations, making this task even more complex. In this paper, we introduce a novel and unified Incremental Relationship-guided Segmentation (IRS) learning scheme to address temporally acquired, partially annotated data while maintaining out-of-distribution (OOD) continual learning capacity in digital pathology. The key innovation of IRS lies in its ability to realize a new spatial-temporal OOD continual learning paradigm by mathematically modeling anatomical relationships between existing and newly introduced classes through a simple incremental universal proposition matrix. Experimental results demonstrate that the IRS method effectively handles the multi-scale nature of pathological segmentation, enabling precise kidney segmentation across various structures (regions, units, and cells) as well as OOD disease lesions at multiple magnifications. This capability significantly enhances domain generalization, making IRS a robust approach for real-world digital pathology applications.

Cold-Start Recommendation with Knowledge-Guided Retrieval-Augmented Generation

Cold-start items remain a persistent challenge in recommender systems due to their lack of historical user interactions, which collaborative models rely on. While recent zero-shot methods leverage large language models (LLMs) to address this, they often struggle with sparse metadata and hallucinated or incomplete knowledge. We propose ColdRAG, a retrieval-augmented generation approach that builds a domain-specific knowledge graph dynamically to enhance LLM-based recommendation in cold-start scenarios, without requiring task-specific fine-tuning. ColdRAG begins by converting structured item attributes into rich natural-language profiles, from which it extracts entities and relationships to construct a unified knowledge graph capturing item semantics. Given a user's interaction history, it scores edges in the graph using an LLM, retrieves candidate items with supporting evidence, and prompts the LLM to rank them. By enabling multi-hop reasoning over this graph, ColdRAG grounds recommendations in verifiable evidence, reducing hallucinations and strengthening semantic connections. Experiments on three public benchmarks demonstrate that ColdRAG surpasses existing zero-shot baselines in both Recall and NDCG. This framework offers a practical solution to cold-start recommendation by combining knowledge-graph reasoning with retrieval-augmented LLM generation.

R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing

Large Language Models (LLMs) achieve impressive reasoning capabilities at the cost of substantial inference overhead, posing substantial deployment challenges. Although distilled Small Language Models (SLMs) significantly enhance efficiency, their performance suffers as they fail to follow LLMs' reasoning paths. Luckily, we reveal that only a small fraction of tokens genuinely diverge reasoning paths between LLMs and SLMs. Most generated tokens are either identical or exhibit neutral differences, such as minor variations in abbreviations or expressions. Leveraging this insight, we introduce **Roads to Rome (R2R)**, a neural token routing method that selectively utilizes LLMs only for these critical, path-divergent tokens, while leaving the majority of token generation to the SLM. We also develop an automatic data generation pipeline that identifies divergent tokens and generates token-level routing labels to train the lightweight router. We apply R2R to combine R1-1.5B and R1-32B models from the DeepSeek family, and evaluate on challenging math, coding, and QA benchmarks. With an average activated parameter size of 5.6B, R2R surpasses the average accuracy of R1-7B by 1.6x, outperforming even the R1-14B model. Compared to R1-32B, it delivers a 2.8x wall-clock speedup with comparable performance, advancing the Pareto frontier of test-time scaling efficiency. Our code is available at https://github.com/thu-nics/R2R.

What Can RL Bring to VLA Generalization? An Empirical Study

Large Vision-Language Action (VLA) models have shown significant potential for embodied AI. However, their predominant training via supervised fine-tuning (SFT) limits generalization due to susceptibility to compounding errors under distribution shifts. Reinforcement learning (RL) offers a path to overcome these limitations by optimizing for task objectives via trial-and-error, yet a systematic understanding of its specific generalization benefits for VLAs compared to SFT is lacking. To address this, our study introduces a comprehensive benchmark for evaluating VLA generalization and systematically investigates the impact of RL fine-tuning across diverse visual, semantic, and execution dimensions. Our extensive experiments reveal that RL fine-tuning, particularly with PPO, significantly enhances generalization in semantic understanding and execution robustness over SFT, while maintaining comparable visual robustness. We identify PPO as a more effective RL algorithm for VLAs than LLM-derived methods like DPO and GRPO. We also develop a simple recipe for efficient PPO training on VLAs, and demonstrate its practical utility for improving VLA generalization. The project page is at https://rlvla.github.io

AgentRecBench: Benchmarking LLM Agent-based Personalized Recommender Systems

The emergence of agentic recommender systems powered by Large Language Models (LLMs) represents a paradigm shift in personalized recommendations, leveraging LLMs' advanced reasoning and role-playing capabilities to enable autonomous, adaptive decision-making. Unlike traditional recommendation approaches, agentic recommender systems can dynamically gather and interpret user-item interactions from complex environments, generating robust recommendation strategies that generalize across diverse scenarios. However, the field currently lacks standardized evaluation protocols to systematically assess these methods. To address this critical gap, we propose: (1) an interactive textual recommendation simulator incorporating rich user and item metadata and three typical evaluation scenarios (classic, evolving-interest, and cold-start recommendation tasks); (2) a unified modular framework for developing and studying agentic recommender systems; and (3) the first comprehensive benchmark comparing 10 classical and agentic recommendation methods. Our findings demonstrate the superiority of agentic systems and establish actionable design guidelines for their core components. The benchmark environment has been rigorously validated through an open challenge and remains publicly available with a continuously maintained leaderboard~\footnote[2]{https://tsinghua-fib-lab.github.io/AgentSocietyChallenge/pages/overview.html}, fostering ongoing community engagement and reproducible research. The benchmark is available at: \hyperlink{https://huggingface.co/datasets/SGJQovo/AgentRecBench}{https://huggingface.co/datasets/SGJQovo/AgentRecBench}.

Adaptive Location Hierarchy Learning for Long-Tailed Mobility Prediction

Human mobility prediction is crucial for applications ranging from location-based recommendations to urban planning, which aims to forecast users' next location visits based on historical trajectories. Despite the severe long-tailed distribution of locations, the problem of long-tailed mobility prediction remains largely underexplored. Existing long-tailed learning methods primarily focus on rebalancing the skewed distribution at the data, model, or class level, neglecting to exploit the spatiotemporal semantics of locations. To address this gap, we propose the first plug-and-play framework for long-tailed mobility prediction in an exploitation and exploration manner, named \textbf{A}daptive \textbf{LO}cation \textbf{H}ier\textbf{A}rchy learning (ALOHA). First, we construct city-tailored location hierarchy based on Large Language Models (LLMs) by exploiting Maslow's theory of human motivation to design Chain-of-Thought (CoT) prompts that captures spatiotemporal semantics. Second, we optimize the location hierarchy predictions by Gumbel disturbance and node-wise adaptive weights within the hierarchical tree structure. Experiments on state-of-the-art models across six datasets demonstrate the framework's consistent effectiveness and generalizability, which strikes a well balance between head and tail locations. Weight analysis and ablation studies reveal the optimization differences of each component for head and tail locations. Furthermore, in-depth analyses of hierarchical distance and case study demonstrate the effective semantic guidance from the location hierarchy. Our code will be made publicly available.

A Graph Perspective to Probe Structural Patterns of Knowledge in Large Language Models

Large language models have been extensively studied as neural knowledge bases for their knowledge access, editability, reasoning, and explainability. However, few works focus on the structural patterns of their knowledge. Motivated by this gap, we investigate these structural patterns from a graph perspective. We quantify the knowledge of LLMs at both the triplet and entity levels, and analyze how it relates to graph structural properties such as node degree. Furthermore, we uncover the knowledge homophily, where topologically close entities exhibit similar levels of knowledgeability, which further motivates us to develop graph machine learning models to estimate entity knowledge based on its local neighbors. This model further enables valuable knowledge checking by selecting triplets less known to LLMs. Empirical results show that using selected triplets for fine-tuning leads to superior performance.