MoE-GPS: Guidlines for Prediction Strategy for Dynamic Expert Duplication in MoE Load Balancing

In multi-GPU Mixture-of-Experts (MoE) network, experts are distributed across different GPUs, which creates load imbalance as each expert processes different number of tokens. Recent works improve MoE inference load balance by dynamically duplicating popular experts to more GPUs to process excessive tokens, which requires predicting the distribution before routing. In this paper, we discuss the tradeoff of prediction strategies, accuracies, overhead, and end-to-end system performance. We propose MoE-GPS, a framework that guides the selection of the optimal predictor design under various system configurations, by quantifying the performance impact to system-level model runtime. Specifically, we advocate for Distribution-Only Prediction, a prediction strategy that only predicts overall token distribution which significantly reduces overhead compared to the traditional Token-to-Expert Prediction. On Mixtral 8x7B MMLU dataset, MoE-GPS suggests Distribution-Only Prediction which improves end-to-end inference performance by more than 23% compared with Token-to-Expert Prediction.

CoreMatching: A Co-adaptive Sparse Inference Framework with Token and Neuron Pruning for Comprehensive Acceleration of Vision-Language Models

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 5x FLOPs reduction and a 10x overall speedup. Code is released at https://github.com/wangqinsi1/2025-ICML-CoreMatching/tree/main.

CONCORD: Concept-Informed Diffusion for Dataset Distillation

Dataset distillation (DD) has witnessed significant progress in creating small datasets that encapsulate rich information from large original ones. Particularly, methods based on generative priors show promising performance, while maintaining computational efficiency and cross-architecture generalization. However, the generation process lacks explicit controllability for each sample. Previous distillation methods primarily match the real distribution from the perspective of the entire dataset, whereas overlooking concept completeness at the instance level. The missing or incorrectly represented object details cannot be efficiently compensated due to the constrained sample amount typical in DD settings. To this end, we propose incorporating the concept understanding of large language models (LLMs) to perform Concept-Informed Diffusion (CONCORD) for dataset distillation. Specifically, distinguishable and fine-grained concepts are retrieved based on category labels to inform the denoising process and refine essential object details. By integrating these concepts, the proposed method significantly enhances both the controllability and interpretability of the distilled image generation, without relying on pre-trained classifiers. We demonstrate the efficacy of CONCORD by achieving state-of-the-art performance on ImageNet-1K and its subsets. The code implementation is released in https://github.com/vimar-gu/CONCORD.

Phi: Leveraging Pattern-based Hierarchical Sparsity for High-Efficiency Spiking Neural Networks

Spiking Neural Networks (SNNs) are gaining attention for their energy efficiency and biological plausibility, utilizing 0-1 activation sparsity through spike-driven computation. While existing SNN accelerators exploit this sparsity to skip zero computations, they often overlook the unique distribution patterns inherent in binary activations. In this work, we observe that particular patterns exist in spike activations, which we can utilize to reduce the substantial computation of SNN models. Based on these findings, we propose a novel \textbf{pattern-based hierarchical sparsity} framework, termed \textbf{\textit{Phi}}, to optimize computation. \textit{Phi} introduces a two-level sparsity hierarchy: Level 1 exhibits vector-wise sparsity by representing activations with pre-defined patterns, allowing for offline pre-computation with weights and significantly reducing most runtime computation. Level 2 features element-wise sparsity by complementing the Level 1 matrix, using a highly sparse matrix to further reduce computation while maintaining accuracy. We present an algorithm-hardware co-design approach. Algorithmically, we employ a k-means-based pattern selection method to identify representative patterns and introduce a pattern-aware fine-tuning technique to enhance Level 2 sparsity. Architecturally, we design \textbf{\textit{Phi}}, a dedicated hardware architecture that efficiently processes the two levels of \textit{Phi} sparsity on the fly. Extensive experiments demonstrate that \textit{Phi} achieves a $3.45\times$ speedup and a $4.93\times$ improvement in energy efficiency compared to state-of-the-art SNN accelerators, showcasing the effectiveness of our framework in optimizing SNN computation.

Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systems

Failure attribution in LLM multi-agent systems-identifying the agent and step responsible for task failures-provides crucial clues for systems debugging but remains underexplored and labor-intensive. In this paper, we propose and formulate a new research area: automated failure attribution for LLM multi-agent systems. To support this initiative, we introduce the Who&When dataset, comprising extensive failure logs from 127 LLM multi-agent systems with fine-grained annotations linking failures to specific agents and decisive error steps. Using the Who&When, we develop and evaluate three automated failure attribution methods, summarizing their corresponding pros and cons. The best method achieves 53.5% accuracy in identifying failure-responsible agents but only 14.2% in pinpointing failure steps, with some methods performing below random. Even SOTA reasoning models, such as OpenAI o1 and DeepSeek R1, fail to achieve practical usability. These results highlight the task's complexity and the need for further research in this area. Code and dataset are available at https://github.com/mingyin1/Agents_Failure_Attribution

HippoMM: Hippocampal-inspired Multimodal Memory for Long Audiovisual Event Understanding

Comprehending extended audiovisual experiences remains a fundamental challenge for computational systems. Current approaches struggle with temporal integration and cross-modal associations that humans accomplish effortlessly through hippocampal-cortical networks. We introduce HippoMM, a biologically-inspired architecture that transforms hippocampal mechanisms into computational advantages for multimodal understanding. HippoMM implements three key innovations: (i) hippocampus-inspired pattern separation and completion specifically designed for continuous audiovisual streams, (ii) short-to-long term memory consolidation that transforms perceptual details into semantic abstractions, and (iii) cross-modal associative retrieval pathways enabling modality-crossing queries. Unlike existing retrieval systems with static indexing schemes, HippoMM dynamically forms integrated episodic representations through adaptive temporal segmentation and dual-process memory encoding. Evaluations on our challenging HippoVlog benchmark demonstrate that HippoMM significantly outperforms state-of-the-art approaches (78.2% vs. 64.2% accuracy) while providing substantially faster response times (20.4s vs. 112.5s). Our results demonstrate that translating neuroscientific memory principles into computational architectures provides a promising foundation for next-generation multimodal understanding systems. The code and benchmark dataset are publicly available at https://github.com/linyueqian/HippoMM.

Keyframe-oriented Vision Token Pruning: Enhancing Efficiency of Large Vision Language Models on Long-Form Video Processing

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.

Proactive Privacy Amnesia for Large Language Models: Safeguarding PII with Negligible Impact on Model Utility

With the rise of large language models (LLMs), increasing research has recognized their risk of leaking personally identifiable information (PII) under malicious attacks. Although efforts have been made to protect PII in LLMs, existing methods struggle to balance privacy protection with maintaining model utility. In this paper, inspired by studies of amnesia in cognitive science, we propose a novel approach, Proactive Privacy Amnesia (PPA), to safeguard PII in LLMs while preserving their utility. This mechanism works by actively identifying and forgetting key memories most closely associated with PII in sequences, followed by a memory implanting using suitable substitute memories to maintain the LLM's functionality. We conduct evaluations across multiple models to protect common PII, such as phone numbers and physical addresses, against prevalent PII-targeted attacks, demonstrating the superiority of our method compared with other existing defensive techniques. The results show that our PPA method completely eliminates the risk of phone number exposure by 100% and significantly reduces the risk of physical address exposure by 9.8% - 87.6%, all while maintaining comparable model utility performance.

H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking

Large Reasoning Models (LRMs) have recently extended their powerful reasoning capabilities to safety checks-using chain-of-thought reasoning to decide whether a request should be answered. While this new approach offers a promising route for balancing model utility and safety, its robustness remains underexplored. To address this gap, we introduce Malicious-Educator, a benchmark that disguises extremely dangerous or malicious requests beneath seemingly legitimate educational prompts. Our experiments reveal severe security flaws in popular commercial-grade LRMs, including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking. For instance, although OpenAI's o1 model initially maintains a high refusal rate of about 98%, subsequent model updates significantly compromise its safety; and attackers can easily extract criminal strategies from DeepSeek-R1 and Gemini 2.0 Flash Thinking without any additional tricks. To further highlight these vulnerabilities, we propose Hijacking Chain-of-Thought (H-CoT), a universal and transferable attack method that leverages the model's own displayed intermediate reasoning to jailbreak its safety reasoning mechanism. Under H-CoT, refusal rates sharply decline-dropping from 98% to below 2%-and, in some instances, even transform initially cautious tones into ones that are willing to provide harmful content. We hope these findings underscore the urgent need for more robust safety mechanisms to preserve the benefits of advanced reasoning capabilities without compromising ethical standards.

A Bayesian Nonparametric Perspective on Mahalanobis Distance for Out of Distribution Detection

Bayesian nonparametric methods are naturally suited to the problem of out-of-distribution (OOD) detection. However, these techniques have largely been eschewed in favor of simpler methods based on distances between pre-trained or learned embeddings of data points. Here we show a formal relationship between Bayesian nonparametric models and the relative Mahalanobis distance score (RMDS), a commonly used method for OOD detection. Building on this connection, we propose Bayesian nonparametric mixture models with hierarchical priors that generalize the RMDS. We evaluate these models on the OpenOOD detection benchmark and show that Bayesian nonparametric methods can improve upon existing OOD methods, especially in regimes where training classes differ in their covariance structure and where there are relatively few data points per class.