Singular Value Decomposition (SVD) has recently gained traction as an effective compression technique for large language models (LLMs), with many studies reporting 20-80% parameter reduction at minimal accuracy cost. However, despite reducing weight memory, existing SVD-based approaches still rely on standard dense CUDA kernels during inference, which incur substantial-and ultimately unnecessary-activation memory overhead. Our analysis reveals that this kernel-induced cost, which grows with sequence length and hidden size, in worst case prevents any real reduction in peak inference memory, limiting the practical impact of SVD compression for on-device deployment. To address this bottleneck, we propose FlashSVD, an end-to-end, rank-aware streaming inference framework for SVD-compressed LLMs. FlashSVD integrates seamlessly with any SVD-based model and directly fuses low-rank projection kernels into self-attention and feed-forward pipelines. This design avoids materializing large activation buffers by streaming small tiles of truncated factors through on-chip SRAM, performing on-the-fly multiplication and reduction, and immediately evicting results–thus preserving high GPU occupancy without introducing latency. On standard benchmarks (e.g., BERT-Base), FlashSVD reduces peak activation memory by up to 70.2% and transient memory by 75%, with zero accuracy loss against low-rank baselines, enabling truly memory-efficient deployment of low-rank LLMs.

Vision-Language Models (VLMs) have demonstrated strong performance on tasks such as video captioning and visual question answering. However, their growing scale and video-level inputs lead to significant computational and memory overhead, posing challenges for real-time deployment on hardware accelerators. While prior work attempts to reduce redundancy via token pruning or merging, these methods typically operate at coarse granularity and incur high runtime overhead due to global token-level operations. In this study, we propose Focus, a Streaming Concentration Architecture that efficiently accelerates VLM inference through progressive, fine-grained redundancy elimination. Focus introduces a multilevel concentration paradigm that hierarchically compresses vision-language inputs at three levels: (1) semantic-guided token pruning based on textual prompts, (2) spatial-temporal blocklevel concentration using localized comparisons, and (3) vectorlevel redundancy removal via motion-aware matching. All concentration steps are tightly co-designed with the architecture to support streaming-friendly, on-chip execution. Focus leverages GEMM tiling, convolution-style layout, and cross-modal attention to minimize off-chip access while enabling high throughput. Implemented as a modular unit within a systolic-array accelerator, Focus achieves$2.4 \times$speedup and$3.3 \times$reduction in energy, significantly outperforming state-of-the-art accelerator in both performance and energy efficiency. Full-stack implementation of Focus is open-sourced at https://github.com/dubcyfor3/Focus.

Effective human-AI collaboration on complex reasoning tasks requires that users understand and interact with the model’s process, not just receive an output. However, the monolithic text from methods like Chain-of-Thought (CoT) prevents this, as current interfaces lack real-time verbalization and robust user barge-in. We present AsyncVoice Agent, a system whose asynchronous architecture decouples a streaming LLM backend from a conversational voice frontend. This design allows narration and inference to run in parallel, empowering users to interrupt, query, and steer the model’s reasoning process at any time. Objective benchmarks show this approach reduces interaction latency by more than 600 $\times$ compared to monolithic baselines while ensuring high fidelity and competitive task accuracy. By enabling a two-way dialogue with a model’s thought process, AsyncVoice Agent offers a new paradigm for building more effective, steerable, and trustworthy human-AI systems for high-stakes tasks.1

Vision language models (VLMs) demonstrate strong capabilities in jointly processing visual and textual data. However, they often incur substantial computational overhead due to redundant visual information, particularly in long-form video scenarios. Existing approaches predominantly focus on either vision token pruning, which may overlook spatio-temporal dependencies, or keyframe selection, which identifies informative frames but discards others, thus disrupting contextual continuity. In this work, we propose KVTP (Keyframe-oriented Vision Token Pruning), a novel framework that overcomes the drawbacks of token pruning and keyframe selection. By adaptively assigning pruning rates based on frame relevance to the query, KVTP effectively retains essential contextual information while significantly reducing redundant computation. To thoroughly evaluate the long-form video understanding capacities of VLMs, we curated and reorganized subsets from VideoMME, EgoSchema, and NextQA into a unified benchmark named SparseKV-QA that highlights real-world scenarios with sparse but crucial events. Our experiments with VLMs of various scales show that KVTP can reduce token usage by 80% without compromising spatiotemporal and contextual consistency, significantly cutting computation while maintaining the performance. These results demonstrate our approach's effectiveness in efficient long-video processing, facilitating more scalable VLM deployment.

Large language models (LLMs) have sparked a new wave of AI applications; however, their substantial computational costs and memory demands pose significant challenges to democratizing access to LLMs for a broader audience. Singular Value Decomposition (SVD), a technique studied for decades, offers a hardware-independent and flexibly tunable solution for LLM compression. In this paper, we present new directions using SVD: we first theoretically analyze the optimality of truncating weights and truncating activations, then we further identify three key issues on SVD-based LLM compression, including (1) How can we determine the optimal truncation position for each weight matrix in LLMs? (2) How can we efficiently update the weight matrices based on truncation position? (3) How can we address the inherent "injection" nature that results in the information loss of the SVD? We propose an effective approach, **Dobi-SVD**, to tackle the three issues. First, we propose a **differentiable** truncation-value learning mechanism, along with gradient-robust backpropagation, enabling the model to adaptively find the optimal truncation positions. Next, we utilize the Eckart-Young-Mirsky theorem to derive a theoretically **optimal** weight update formula through rigorous mathematical analysis. Lastly, by observing and leveraging the quantization-friendly nature of matrices after SVD decomposition, we reconstruct a mapping between truncation positions and memory requirements, establishing a **bijection** from truncation positions to memory. Experimental results show that with a 40\% parameter-compression rate, our method achieves a perplexity of 9.07 on the Wikitext2 dataset with the compressed LLama-7B model, a 78.7\% improvement over the state-of-the-art SVD for LLM compression method. We emphasize that Dobi-SVD is the first to achieve such a high-ratio LLM compression with minimal performance drop. We also extend our Dobi-SVD to VLM compression, achieving a 20\% increase in throughput with minimal performance degradation. We hope that the inference speedup—up to 12.4x on 12GB NVIDIA Titan Xp GPUs and 3x on 80GB A100 GPUs for LLMs, and 1.2x on 80GB A100 GPUs for VLMs—will bring significant benefits to the broader community such as robotics.

Vision-Language Models (VLMs) excel across diverse tasks but suffer from high inference costs in time and memory. Token sparsity mitigates inefficiencies in token usage, while neuron sparsity reduces high-dimensional computations, both offering promising solutions to enhance efficiency. Recently, these two sparsity paradigms have evolved largely in parallel, fostering the prevailing assumption that they function independently. However, a fundamental yet underexplored question remains: Do they truly operate in isolation, or is there a deeper underlying interplay that has yet to be uncovered? In this paper, we conduct the first comprehensive investigation into this question. By introducing and analyzing the matching mechanism between Core Neurons and Core Tokens, we found that key neurons and tokens for inference mutually influence and reinforce each other. Building on this insight, we propose CoreMatching, a co-adaptive sparse inference framework, which leverages the synergy between token and neuron sparsity to enhance inference efficiency. Through theoretical analysis and efficiency evaluations, we demonstrate that the proposed method surpasses state-of-the-art baselines on ten image understanding tasks and three hardware devices. Notably, on the NVIDIA Titan Xp, it achieved 5$\times$ FLOPs reduction and a 10$\times$ overall speedup. Code is released at https://github.com/wangqinsi1/2025-ICML-CoreMatching/tree/main.

Current Reinforcement Fine-tuning (RFT) paradigms for Large Language Models (LLMs) suffer from sample inefficiency due to the redundant exposure of identical queries under uniform data sampling. While previous work has explored curriculum learning via heuristic difficulty metrics, these strategies exhibit limitations by neglecting the intrinsic learning signals generated by the model itself, thus leading to suboptimal training regimes. In this paper, we identify a model-inherent signal termed *angle concentration* that effectively reflects an LLM's capacity to learn from specific data. We theoretically and empirically demonstrate a correlation between the angular distribution of token hidden state vectors and the resulting gradient, revealing a learning preference for data exhibiting higher angle concentration. Inspired by this finding, we propose GAIN-RL, a Gradient-driven Angle-Informed Navigated RL framework. By leveraging the model's intrinsic angle concentration signal, GAIN-RL dynamically selects training data in each epoch, ensuring consistently impactful gradient updates and thus significantly enhancing overall training efficiency. Empirical evaluations show that GAIN-RL (GRPO) achieves over a 2.5$\times$ acceleration in training efficiency across diverse mathematical and coding tasks and varying model scales. Furthermore, GAIN-RL (GRPO)'s efficient sampling yields data-efficient training, achieving better performance with half the original data compared to vanilla GRPO with full training data.

Multi-agent large language model (LLM) systems are increasingly adopted for complex language processing tasks that require communication and coordination among agents. However, these systems often suffer substantial overhead from repeated reprocessing of overlapping contexts across agents. In typical pipelines, once an agent receives a message from its predecessor, the full context-including prior turns-must be reprocessed from scratch, leading to inefficient processing. While key-value (KV) caching is an effective solution for avoiding redundant computation in single-agent settings where prefixes remain unchanged, it cannot be directly reused in multi-agent scenarios due to diverging prefixes introduced by agent-specific context extensions. We identify that the core challenge lies in the offset variance of KV-caches across agents. To address this, we propose **KVCOMM**, a training-free framework that enables efficient prefilling in multi-agent inference by reusing KV-caches and aligning cache offsets of overlapping contexts under diverse prefix contexts. KVCOMM estimates and adjusts KV-caches for shared content by referencing a pool of cached examples—termed *anchors*—that store observed cache deviations under varying prefixes. The anchor pool is maintained and updated online, allowing dynamic adaptation to distinct user requests and context structures. KVCOMM achieves over 70% reuse rate across diverse multi- agent workloads, including retrieval-augmented generation, math reasoning, and collaborative coding tasks, all without quality degradation. Particularly, when each fully-connected agent receives 1K input tokens with 512 prefix tokens and 512 output tokens under a five-agent setting, KVCOMM achieves up to 7.8× speedup compared to the standard prefill pipeline, reducing TTFT from ∼430ms to ∼55ms. Code is available at https://github.com/FastMAS/KVCOMM.

The low-cost Neural Architecture Search (NAS) for lightweight networks working on massive mobile devices is essential for fast-developing ICT technology. Current NAS work can not search on unseen devices without latency sampling, which is a big obstacle to the implementation of NAS on mobile devices. In this paper, we overcome this challenge by proposing the Device Generic Latency (DGL) model. By absorbing processor modeling technology, the proposed DGL formula maps the parameters in the interval theory to the seven static configuration parameters of the device. And to make the formula more practical, we refine it to low-cost form by decreasing the number of configuration parameters to four. Then based on this formula, the DGL model is proposed which introduces the network parameters predictor and accuracy predictor to work with the DGL formula to predict the network latency. We propose the DGL-based NAS framework to enable fast searches without latency sampling. Extensive experiments results validate that the DGL model can achieve more accurate latency predictions than existing NAS latency predictors on unseen mobile devices. When configured with current state-of-the-art predictors, DGL-based NAS can search for architectures with higher accuracy that meet the latency limit than other NAS implementations, while using less training time and prediction time. Our work shed light on how to adopt domain knowledge into NAS topic and play important role in low-cost NAS on mobile devices.

Neural Architecture Search (NAS) has emerged as a favoured method for unearthing effective neural architectures. Recent development of large models has intensified the demand for faster search speeds and more accurate search results. However, designing large models by NAS is challenging due to the dramatical increase of search space and the associated huge performance evaluation cost. Consider a typical modular search space widely used in NAS, in which a neural architecture consists of $m$ block nodes and a block node has $n$ alternative blocks. Facing the space containing $n^m$ candidate networks, existing NAS methods attempt to find the best one by searching and evaluating candidate networks directly. Different from the general strategy that takes architecture search as a whole problem, we propose a novel divide-and-conquer strategy by making use of the modular nature of the search space. Here, we introduce MathNAS, a general NAS framework based on mathematical programming. In MathNAS, the performances of all possible building blocks in the search space are calculated first, and then the performance of a network is directly predicted based on the performances of its building blocks. Although estimating block performances involves network training, just as what happens for network performance evaluation in existing NAS methods, predicting network performance is completely training-free and thus extremely fast. In contrast to the $n^m$ candidate networks to evaluate in existing NAS methods, which requires training and a formidable computational burden, there are only $m*n$ possible blocks to handle in MathNAS. Therefore, our approach effectively reduces the complexity of network performance evaluation. The superiority of MathNAS is validated on multiple large-scale CV and NLP benchmark datasets. Notably on ImageNet-1k, MathNAS achieves 82.5\% top-1 accuracy, 1.2\% and 0.96\% higher than Swin-T and LeViT-256, respectively. In addition, when deployed on mobile device, MathNAS achieves real-time search and dynamic network switching within 1s (0.4s on TX2 GPU), surpassing baseline dynamic networks in on-device performance.

In this paper, we survey recent progress in CyberCardia project, a CPS Frontier project funded by the National Science Foundation. The CyberCardia project will lead to significant advances in the state of the art for system verification and cardiac therapies based on the use of formal methods and closed-loop control and verification. The animating vision for the work is to enable the development of a true in silico design methodology for medical devices that can be used to speed the development of new devices and to provide greater assurance that their behavior matches designer intentions, and to pass regulatory muster more quickly so that they can be used on patients needing their care. The acceleration in medical-device innovation achievable as a result of the CyberCardia research will also have long-term and sustained societal benefits, as better diagnostic and therapeutic technologies enter into the practice of medicine more quickly.

In this work, we developed a methodology to analyze a bacteria model that mimics the stages through which bacteria change when phage therapy is applied. Due to the widespread misuse and overuse of antibiotics, drug resistant bacteria now pose significant risks to health, agriculture and the environment. Therefore, we were interested in an alternative to conventional antibiotics, a phage therapy. Our model was designed according to an experimental procedure to engineer a temperate phage, Lambda (λ), and then kill bacteria via light-activated production of superoxide. We applied formal analysis to our model and the results show that such an approach can speed up evaluation of the system, which would be impractical or possibly not even feasible to study in a wet lab.