Wenkai Wang, Xiyun Li, Hongcan Guo, Wenhao Yu, Tianqing Fang, Haitao Mi, Dong Yu, Shengyu Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large language models (LLMs) face significant challenges in long-context modeling due to increased inference costs, higher latency, and performance degradation caused by information redundancy. Context compression offers a promising solution, but existing methods often rely on fixed strategies that don’t adapt to variations in information density. To address this, we propose the Integrate-and-Fire Compressor (IFC), an adaptive method inspired by the neural integrate-and-fire mechanism. IFC assesses token importance more accurately and dynamically adjusts compression, producing compact and effective context representations for LLMs, even at high compression rates. Experiments on three open-domain QA datasets show that IFC consistently outperforms existing baselines in both compression rate and task performance, highlighting its potential to improve the efficiency and effectiveness of LLMs in handling long contexts.

Despite their success in specific scenarios, existing foundation agents still struggle to generalize across various virtual scenarios, mainly due to the dramatically different encapsulations of environments with manually designed observation and action spaces. To handle this issue, we propose the General Computer Control (GCC) setting to restrict foundation agents to interact with software through the most unified and standardized interface, i.e., using screenshots as input and keyboard and mouse actions as output. We introduce Cradle, a modular and flexible LMM-powered framework, as a preliminary attempt towards GCC. Enhanced by six key modules, Information Gathering, Self-Reflection, Task Inference, Skill Curation, Action Planning, and Memory, Cradle is able to understand input screenshots and output executable code for low-level keyboard and mouse control after high-level planning and information retrieval, so that Cradle can interact with any software and complete long-horizon complex tasks without relying on any built-in APIs. Experimental results show that Cradle exhibits remarkable generalizability and impressive performance across four previously unexplored commercial video games (Red Dead Redemption 2, Cities:Skylines, Stardew Valley and Dealer’s Life 2), five software applications (Chrome, Outlook, Feishu, Meitu and CapCut), and a comprehensive benchmark, OSWorld. With a unified interface to interact with any software, Cradle greatly extends the reach of foundation agents thus paving the way for generalist agents.

Comprehending natural language and following human instructions are critical capabilities for intelligent agents. However, the flexibility of linguistic instructions induces substantial ambiguity across language-conditioned tasks, severely degrading algorithmic performance. To address these limitations, we present a novel method named DAIL (Distributional Aligned Learning), featuring two key components: distributional policy and semantic alignment. Specifically, we provide theoretical results that the value distribution estimation mechanism enhances task differentiability. Meanwhile, the semantic alignment module captures the correspondence between trajectories and linguistic instructions. Extensive experimental results on both structured and visual observation benchmarks demonstrate that DAIL effectively resolves instruction ambiguities, achieving superior performance to baseline methods. Our implementation is available at https://github.com/RunpengXie/Distributional-Aligned-Learning.

Enhancing the theory of mind (ToM) ability of agents is becoming more and more critical in the research area of cooperative multi-agent reinforcement learning (MARL). ToM describes the ability of agents to understand their partners’ logic first, and then reason their intentions and behaviors accurately. In cognitive science, dual-reasoning pathway theory (DRPT) is a statistical ToM process, which claims that humans can achieve accurate and rapid reasoning by combining long-term and short-term reasoning (LTR and STR) pathways that cover different brain regions. However, most existing works focus on the quick decision-making ability of the STR, while overlooking the significance of the long-term ToM reasoning ability from the LTR. To emphasize such ability, we propose a long short-term reasoning (LSTR) algorithm which contains a large language model (LLM) for long-term reasoning and an additional augmenting module to decode the semantic space in LLM as the action space in MARL. Experimental results demonstrate that our LSTR algorithm has achieved significant improvement over competitive MARL methods (e.g., value-based QMIX and policy-based COMA) from the perspective of reward scores, convergence speed, and scalability.

Recently, our proposed recurrent neural network (RNN) based all deep learning minimum variance distortionless response (ADL-MVDR) beamformer method yielded superior performance over the conventional MVDR by replacing the matrix inversion and eigenvalue decomposition with two RNNs.In this work, we present a self-attentive RNN beamformer to further improve our previous RNN-based beamformer by leveraging on the powerful modeling capability of self-attention.Temporal-spatial self-attention module is proposed to better learn the beamforming weights from the speech and noise spatial covariance matrices.The temporal self-attention module could help RNN to learn global statistics of covariance matrices.The spatial self-attention module is designed to attend on the cross-channel correlation in the covariance matrices.Furthermore, a multi-channel input with multi-speaker directional features and multi-speaker speech separation outputs (MIMO) model is developed to improve the inference efficiency.The evaluations demonstrate that our proposed MIMO self-attentive RNN beamformer improves both the automatic speech recognition (ASR) accuracy and the perceptual estimation of speech quality (PESQ) against prior arts.