Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA's compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA's latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA's design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5$\times$ prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.
Zeng You South China University of Technology, Yaofo Chen, Qiuwu Chen, Ying Sun AIGCode, Shuhai Zhang, Yingjian Li, Yaowei Wang Harbin Institute of Technology, Shenzhen and Pengcheng Laboratory, Mingkui Tan
Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA’s compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA’s latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA’s design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5× prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Niu Lian, Yuting Wang Aminer, Hanshu Yao, Jinpeng Wang Harbin Institute of Technology, Shenzhen, Bin Chen Harbin Institute of Technology, Shenzhen, Yaowei Wang Harbin Institute of Technology, Shenzhen and Pengcheng Laboratory, Min Zhang Harbin Institute of Technology, Shenzhen and Harbin Institute of Technology, Shu-Tao Xia Shenzhen International Graduate School, Tsinghua University
While multimodal large language models have demonstrated impressive short-term reasoning, they struggle with long-horizon video understanding due to limited context windows and static memory mechanisms that fail to mirror human cognitive efficiency. Existing paradigms typically fall into two extremes: vision-centric methods that incur high latency and redundancy through dense visual accumulation, or text-centric approaches that suffer from detail loss and hallucination via aggressive captioning. To bridge this gap, we propose MM-Mem, a pyramidal multimodal memory architecture grounded in Fuzzy-Trace Theory. MM-Mem structures memory hierarchically into a Sensory Buffer, Episodic Stream, and Symbolic Schema, enabling the progressive distillation of fine-grained perceptual traces (verbatim) into high-level semantic schemas (gist).Furthermore, to govern the dynamic construction of memory, we derive a Semantic Information Bottleneck objective and introduce SIB-GRPO to optimize the trade-off between memory compression and task-relevant information retention.In inference, we design an entropy-driven top-down memory retrieval strategy.Extensive experiments across 4 benchmarks confirm that MM-Mem achieves state-of-the-art performance on both offline and streaming tasks, demonstrating robust generalization and validating the effectiveness of cognition-inspired memory organization.Code and associated configurations are publicly available at ‘https://github.com/EliSpectre/MM-Mem‘.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表