HeadRouter: Dynamic Head-Weight Routing for Task-Adaptive Audio Token Pruning in Large Audio Language Models

Recent large audio language models (LALMs) demonstrate remarkable capabilities in processing extended multi-modal sequences, yet incur high inference costs. Token compression is an effective method that directly reduces redundant tokens in the sequence. Existing compression methods usually assume that all attention heads in LALMs contribute equally to various audio tasks and calculate token importance by averaging scores across all heads. However, our analysis demonstrates that attention heads exhibit distinct behaviors across diverse audio domains. We further reveal that only a sparse subset of attention heads actively responds to audio, with completely different performance when handling semantic and acoustic tasks. In light of this observation, we propose HeadRouter, a head-importance-aware token pruning method that perceives the varying importance of attention heads in different audio tasks to maximize the retention of crucial tokens. HeadRouter is training-free and can be applied to various LALMs. Extensive experiments on the AudioMarathon and MMAU-Pro benchmarks demonstrate that HeadRouter achieves state-of-the-art compression performance, exceeding the baseline model even when retaining 70% of the audio tokens and achieving 101.8% and 103.0% of the vanilla average on Qwen2.5-Omni-3B and Qwen2.5-Omni-7B, respectively.

RAICL: Retrieval-Augmented In-Context Learning for Vision-Language-Model Based EEG Seizure Detection

Electroencephalogram (EEG) decoding is a critical component of medical diagnostics, rehabilitation engineering, and brain-computer interfaces. However, contemporary decoding methodologies remain heavily dependent on task-specific datasets to train specialized neural network architectures. Consequently, limited data availability impedes the development of generalizable large brain decoding models. In this work, we propose a paradigm shift from conventional signal-based decoding by leveraging large-scale vision-language models (VLMs) to analyze EEG waveform plots. By converting multivariate EEG signals into stacked waveform images and integrating neuroscience domain expertise into textual prompts, we demonstrate that foundational VLMs can effectively differentiate between different patterns in the human brain. To address the inherent non-stationarity of EEG signals, we introduce a Retrieval-Augmented In-Context Learning (RAICL) approach, which dynamically selects the most representative and relevant few-shot examples to condition the autoregressive outputs of the VLM. Experiments on EEG-based seizure detection indicate that state-of-the-art VLMs under RAICL achieved better or comparable performance with traditional time series based approaches. These findings suggest a new direction in physiological signal processing that effectively bridges the modalities of vision, language, and neural activities. Furthermore, the utilization of off-the-shelf VLMs, without the need for retraining or downstream architecture construction, offers a readily deployable solution for clinical applications.

Mixing Importance with Diversity: Joint Optimization for KV Cache Compression in Large Vision-Language Models

Recent large vision-language models (LVLMs) demonstrate remarkable capabilities in processing extended multi-modal sequences, yet the resulting key-value (KV) cache expansion creates a critical memory bottleneck that fundamentally limits deployment scalability. While existing KV cache compression methods focus on retaining high-importance KV pairs to minimize storage, they often overlook the modality-specific semantic redundancy patterns that emerge distinctively in multi-modal KV caches. In this work, we first analyze how, beyond simple importance, the KV cache in LVLMs exhibits varying levels of redundancy across attention heads. We show that relying solely on importance can only cover a subset of the full KV cache information distribution, leading to potential loss of semantic coverage. To address this, we propose \texttt{MixKV}, a novel method that mixes importance with diversity for optimized KV cache compression in LVLMs. \texttt{MixKV} adapts to head-wise semantic redundancy, selectively balancing diversity and importance when compressing KV pairs. Extensive experiments demonstrate that \texttt{MixKV} consistently enhances existing methods across multiple LVLMs. Under extreme compression (budget=64), \texttt{MixKV} improves baseline methods by an average of \textbf{5.1\%} across five multi-modal understanding benchmarks and achieves remarkable gains of \textbf{8.0\%} and \textbf{9.0\%} for SnapKV and AdaKV on GUI grounding tasks, all while maintaining comparable inference efficiency. Furthermore, \texttt{MixKV} extends seamlessly to LLMs with comparable performance gains. Our code is available at \href{https://github.com/xuyang-liu16/MixKV}{\textcolor{citeblue}{https://github.com/xuyang-liu16/MixKV}}.