Mixture-of-Channels: Exploiting Sparse FFNs for Efficient LLMs Pre-Training and Inference

Large language models (LLMs) have demonstrated remarkable success across diverse artificial intelligence tasks, driven by scaling laws that correlate model size and training data with performance improvements. However, this scaling paradigm incurs substantial memory overhead, creating significant challenges for both training and inference. While existing research has primarily addressed parameter and optimizer state memory reduction, activation memory-particularly from feed-forward networks (FFNs)-has become the critical bottleneck, especially when FlashAttention is implemented. In this work, we conduct a detailed memory profiling of LLMs and identify FFN activations as the predominant source to activation memory overhead. Motivated by this, we introduce Mixture-of-Channels (MoC), a novel FFN architecture that selectively activates only the Top-K most relevant channels per token determined by SwiGLU's native gating mechanism. MoC substantially reduces activation memory during pre-training and improves inference efficiency by reducing memory access through partial weight loading into GPU SRAM. Extensive experiments validate that MoC delivers significant memory savings and throughput gains while maintaining competitive model performance.

An All-Reduce Compatible Top-K Compressor for Communication-Efficient Distributed Learning

Communication remains a central bottleneck in large-scale distributed machine learning, and gradient sparsification has emerged as a promising strategy to alleviate this challenge. However, existing gradient compressors face notable limitations: Rand-$K$\ discards structural information and performs poorly in practice, while Top-$K$\ preserves informative entries but loses the contraction property and requires costly All-Gather operations. In this paper, we propose ARC-Top-$K$, an {All-Reduce}-Compatible Top-$K$ compressor that aligns sparsity patterns across nodes using a lightweight sketch of the gradient, enabling index-free All-Reduce while preserving globally significant information. ARC-Top-$K$\ is provably contractive and, when combined with momentum error feedback (EF21M), achieves linear speedup and sharper convergence rates than the original EF21M under standard assumptions. Empirically, ARC-Top-$K$\ matches the accuracy of Top-$K$\ while reducing wall-clock training time by up to 60.7\%, offering an efficient and scalable solution that combines the robustness of Rand-$K$\ with the strong performance of Top-$K$.

An Efficient Subspace Algorithm for Federated Learning on Heterogeneous Data

This work addresses the key challenges of applying federated learning to large-scale deep neural networks, particularly the issue of client drift due to data heterogeneity across clients and the high costs of communication, computation, and memory. We propose FedSub, an efficient subspace algorithm for federated learning on heterogeneous data. Specifically, FedSub utilizes subspace projection to guarantee local updates of each client within low-dimensional subspaces, thereby reducing communication, computation, and memory costs. Additionally, it incorporates low-dimensional dual variables to mitigate client drift. We provide convergence analysis that reveals the impact of key factors such as step size and subspace projection matrices on convergence. Experimental results demonstrate its efficiency.

Bridging Video Quality Scoring and Justification via Large Multimodal Models

Classical video quality assessment (VQA) methods generate a numerical score to judge a video's perceived visual fidelity and clarity. Yet, a score fails to describe the video's complex quality dimensions, restricting its applicability. Benefiting from the linguistic output, adapting video large multimodal models (LMMs) to VQA via instruction tuning has the potential to address this issue. The core of the approach lies in the video quality-centric instruction data. Previous explorations mainly focus on the image domain, and their data generation processes heavily rely on human quality annotations and proprietary systems, limiting data scalability and effectiveness. To address these challenges, we propose the Score-based Instruction Generation (SIG) pipeline. Specifically, SIG first scores multiple quality dimensions of an unlabeled video and maps scores to text-defined levels. It then explicitly incorporates a hierarchical Chain-of-Thought (CoT) to model the correlation between specific dimensions and overall quality, mimicking the human visual system's reasoning process. The automated pipeline eliminates the reliance on expert-written quality descriptions and proprietary systems, ensuring data scalability and generation efficiency. To this end, the resulting Score2Instruct (S2I) dataset contains over 320K diverse instruction-response pairs, laying the basis for instruction tuning. Moreover, to advance video LMMs' quality scoring and justification abilities simultaneously, we devise a progressive tuning strategy to fully unleash the power of S2I. Built upon SIG, we further curate a benchmark termed S2I-Bench with 400 open-ended questions to better evaluate the quality justification capacity of video LMMs. Experimental results on the S2I-Bench and existing benchmarks indicate that our method consistently improves quality scoring and justification capabilities across multiple video LMMs.

Efficient Long-Context LLM Inference via KV Cache Clustering

Large language models (LLMs) with extended context windows have become increasingly prevalent for tackling complex tasks. However, the substantial Key-Value (KV) cache required for long-context LLMs poses significant deployment challenges. Existing approaches either discard potentially critical information needed for future generations or offer limited efficiency gains due to high computational overhead. In this paper, we introduce Chelsea, a simple yet effective framework for online KV cache clustering. Our approach is based on the observation that key states exhibit high similarity along the sequence dimension. To enable efficient clustering, we divide the sequence into chunks and propose Chunked Soft Matching, which employs an alternating partition strategy within each chunk and identifies clusters based on similarity. Chelsea then merges the KV cache within each cluster into a single centroid. Additionally, we provide a theoretical analysis of the computational complexity and the optimality of the intra-chunk partitioning strategy. Extensive experiments across various models and long-context benchmarks demonstrate that Chelsea achieves up to 80% reduction in KV cache memory usage while maintaining comparable model performance. Moreover, with minimal computational overhead, Chelsea accelerates the decoding stage of inference by up to 3.19$\times$ and reduces end-to-end latency by up to 2.72$\times$.

EndoVLA: Dual-Phase Vision-Language-Action Model for Autonomous Tracking in Endoscopy

In endoscopic procedures, autonomous tracking of abnormal regions and following circumferential cutting markers can significantly reduce the cognitive burden on endoscopists. However, conventional model-based pipelines are fragile for each component (e.g., detection, motion planning) requires manual tuning and struggles to incorporate high-level endoscopic intent, leading to poor generalization across diverse scenes. Vision-Language-Action (VLA) models, which integrate visual perception, language grounding, and motion planning within an end-to-end framework, offer a promising alternative by semantically adapting to surgeon prompts without manual recalibration. Despite their potential, applying VLA models to robotic endoscopy presents unique challenges due to the complex and dynamic anatomical environments of the gastrointestinal (GI) tract. To address this, we introduce EndoVLA, designed specifically for continuum robots in GI interventions. Given endoscopic images and surgeon-issued tracking prompts, EndoVLA performs three core tasks: (1) polyp tracking, (2) delineation and following of abnormal mucosal regions, and (3) adherence to circular markers during circumferential cutting. To tackle data scarcity and domain shifts, we propose a dual-phase strategy comprising supervised fine-tuning on our EndoVLA-Motion dataset and reinforcement fine-tuning with task-aware rewards. Our approach significantly improves tracking performance in endoscopy and enables zero-shot generalization in diverse scenes and complex sequential tasks.

ORQA: A Benchmark and Foundation Model for Holistic Operating Room Modeling

The real-world complexity of surgeries necessitates surgeons to have deep and holistic comprehension to ensure precision, safety, and effective interventions. Computational systems are required to have a similar level of comprehension within the operating room. Prior works, limited to single-task efforts like phase recognition or scene graph generation, lack scope and generalizability. In this work, we introduce ORQA, a novel OR question answering benchmark and foundational multimodal model to advance OR intelligence. By unifying all four public OR datasets into a comprehensive benchmark, we enable our approach to concurrently address a diverse range of OR challenges. The proposed multimodal large language model fuses diverse OR signals such as visual, auditory, and structured data, for a holistic modeling of the OR. Finally, we propose a novel, progressive knowledge distillation paradigm, to generate a family of models optimized for different speed and memory requirements. We show the strong performance of ORQA on our proposed benchmark, and its zero-shot generalization, paving the way for scalable, unified OR modeling and significantly advancing multimodal surgical intelligence. We will release our code and data upon acceptance.

Diffusion Learning with Partial Agent Participation and Local Updates

Diffusion learning is a framework that endows edge devices with advanced intelligence. By processing and analyzing data locally and allowing each agent to communicate with its immediate neighbors, diffusion effectively protects the privacy of edge devices, enables real-time response, and reduces reliance on central servers. However, traditional diffusion learning relies on communication at every iteration, leading to communication overhead, especially with large learning models. Furthermore, the inherent volatility of edge devices, stemming from power outages or signal loss, poses challenges to reliable communication between neighboring agents. To mitigate these issues, this paper investigates an enhanced diffusion learning approach incorporating local updates and partial agent participation. Local updates will curtail communication frequency, while partial agent participation will allow for the inclusion of agents based on their availability. We prove that the resulting algorithm is stable in the mean-square error sense and provide a tight analysis of its Mean-Square-Deviation (MSD) performance. Various numerical experiments are conducted to illustrate our theoretical findings.

NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement: KwaiSR Dataset and Study

In this work, we build the first benchmark dataset for short-form UGC Image Super-resolution in the wild, termed KwaiSR, intending to advance the research on developing image super-resolution algorithms for short-form UGC platforms. This dataset is collected from the Kwai Platform, which is composed of two parts, i.e., synthetic and wild parts. Among them, the synthetic dataset, including 1,900 image pairs, is produced by simulating the degradation following the distribution of real-world low-quality short-form UGC images, aiming to provide the ground truth for training and objective comparison in the validation/testing. The wild dataset contains low-quality images collected directly from the Kwai Platform, which are filtered using the quality assessment method KVQ from the Kwai Platform. As a result, the KwaiSR dataset contains 1800 synthetic image pairs and 1900 wild images, which are divided into training, validation, and testing parts with a ratio of 8:1:1. Based on the KwaiSR dataset, we organize the NTIRE 2025 challenge on a second short-form UGC Video quality assessment and enhancement, which attracts lots of researchers to develop the algorithm for it. The results of this competition have revealed that our KwaiSR dataset is pretty challenging for existing Image SR methods, which is expected to lead to a new direction in the image super-resolution field. The dataset can be found from https://lixinustc.github.io/NTIRE2025-KVQE-KwaSR-KVQ.github.io/.

Can DeepSeek Reason Like a Surgeon? An Empirical Evaluation for Vision-Language Understanding in Robotic-Assisted Surgery

DeepSeek series have demonstrated outstanding performance in general scene understanding, question-answering (QA), and text generation tasks, owing to its efficient training paradigm and strong reasoning capabilities. In this study, we investigate the dialogue capabilities of the DeepSeek model in robotic surgery scenarios, focusing on tasks such as Single Phrase QA, Visual QA, and Detailed Description. The Single Phrase QA tasks further include sub-tasks such as surgical instrument recognition, action understanding, and spatial position analysis. We conduct extensive evaluations using publicly available datasets, including EndoVis18 and CholecT50, along with their corresponding dialogue data. Our comprehensive evaluation results indicate that, when provided with specific prompts, DeepSeek-V3 performs well in surgical instrument and tissue recognition tasks However, DeepSeek-V3 exhibits significant limitations in spatial position analysis and struggles to understand surgical actions accurately. Additionally, our findings reveal that, under general prompts, DeepSeek-V3 lacks the ability to effectively analyze global surgical concepts and fails to provide detailed insights into surgical scenarios. Based on our observations, we argue that the DeepSeek-V3 is not ready for vision-language tasks in surgical contexts without fine-tuning on surgery-specific datasets.