Ning Yang

Large Language Models (LLMs) in long-context scenarios are severely constrained by the linear growth of Key-Value (KV) cache memory. Existing KV compression methods rely either on static thresholds and attention-only heuristics or on coarse memory budget allocation. Under tight memory budgets, these methods overlook two key factors: prompt-dependent variation in compression risk and functional heterogeneity across attention heads, which destabilize token selection and lead to tail failures. To address these challenges, we propose CompilerKV, a risk-adaptive and head-aware compression framework that compiles offline experience into reusable decision tables for prefill-only deployment. CompilerKV integrates two key synergistic components: (i) a Head Heterogeneity Table, learned via offline contextual bandits, which assigns head-specific reliability weights to govern functional differences across attention heads explicitly; and (ii) a Risk-Adaptive Threshold Gating mechanism that jointly models attention entropy and local perplexity, transforming prompt-level risk into deployable retention thresholds. Experiments on LongBench show CompilerKV dominates SOTA methods under a 512-token budget, recovering 97.7\% of FullKV performance while achieving up to +5.2 points gain over the strongest competitor.

Accurate spatiotemporal traffic forecasting is a critical prerequisite for proactive resource management in dense urban mobile networks. While Large Language Models (LLMs) have shown promise in time series analysis, they inherently struggle to model the complex spatial dependencies of grid-based traffic data. Effectively extending LLMs to this domain is challenging, as representing the vast amount of information from dense geographical grids can be inefficient and overwhelm the model's context. To address these challenges, we propose ST-Vision-LLM, a novel framework that reframes spatiotemporal forecasting as a vision-language fusion problem. Our approach leverages a Vision-LLM visual encoder to process historical global traffic matrices as image sequences, providing the model with a comprehensive global view to inform cell-level predictions. To overcome the inefficiency of LLMs in handling numerical data, we introduce an efficient encoding scheme that represents floating-point values as single tokens via a specialized vocabulary, coupled with a two-stage numerical alignment fine-tuning process. The model is first trained with Supervised Fine-Tuning (SFT) and then further optimized for predictive accuracy using Group Relative Policy Optimization (GRPO), a memory-efficient reinforcement learning method. Evaluations on real-world mobile traffic datasets demonstrate that ST-Vision-LLM outperforms existing methods by 15.6% in long-term prediction accuracy and exceeds the second-best baseline by over 30.04% in cross-domain few-shot scenarios. Our extensive experiments validate the model's strong generalization capabilities across various data-scarce environments.

Mobile Edge Computing (MEC) leverages computational heterogeneity between mobile devices and edge nodes to enable real-time applications requiring high information freshness. The Age-of-Information (AoI) metric serves as a crucial evaluator of information timeliness in such systems. Addressing AoI minimization in multi-user MEC environments presents significant challenges due to stochastic computing times. In this paper, we consider multiple users offloading tasks to heterogeneous edge servers in an MEC system, focusing on preemptive and non-preemptive task scheduling mechanisms. The problem is first reformulated as a Restless Multi-Arm Bandit (RMAB) problem, with a multi-layer Markov Decision Process (MDP) framework established to characterize AoI dynamics in the MEC system. Based on the multi-layer MDP, we propose a nested index framework and design a nested index policy with provably asymptotic optimality. This establishes a theoretical framework adaptable to various scheduling mechanisms, achieving efficient optimization through state stratification and index design in both preemptive and non-preemptive modes. Finally, the closed-form of the nested index is derived, facilitating performance trade-offs between computational complexity and accuracy while ensuring the universal applicability of the nested index policy across both scheduling modes. The experimental results show that in non-preemptive scheduling, compared with the benchmark method, the optimality gap is reduced by 25.43%, while in preemptive scheduling, the gap has reduced by 61.84%. As the system scale increases, it asymptotically converges in two scheduling modes and especially provides near-optimal performance in non-preemptive structure.

Safe Reinforcement Learning (RL) often faces significant issues such as constraint violations and instability, necessitating the use of constrained policy optimization, which seeks optimal policies while ensuring adherence to specific constraints like safety. Typically, constrained optimization problems are addressed by the Lagrangian method, a post-violation remedial approach that may result in oscillations and overshoots. Motivated by this, we propose a novel method named Proactive Constrained Policy Optimization (PCPO) that incorporates a preemptive penalty mechanism. This mechanism integrates barrier items into the objective function as the policy nears the boundary, imposing a cost. Meanwhile, we introduce a constraint-aware intrinsic reward to guide boundary-aware exploration, which is activated only when the policy approaches the constraint boundary. We establish theoretical upper and lower bounds for the duality gap and the performance of the PCPO update, shedding light on the method's convergence characteristics. Additionally, to enhance the optimization performance, we adopt a policy iteration approach. An interesting finding is that PCPO demonstrates significant stability in experiments. Experimental results indicate that the PCPO framework provides a robust solution for policy optimization under constraints, with important implications for future research and practical applications.