Miao Wang

Yijiang Liu, Huanrui Yang, Youxin Chen et al.

Large language models (LLMs) excel in language tasks, especially with supervised fine-tuning after pre-training. However, their substantial memory and computational requirements hinder practical applications. Structural pruning, which reduces less significant weight dimensions, is one solution. Yet, traditional post-hoc pruning often leads to significant performance loss, with limited recovery from further fine-tuning due to reduced capacity. Since the model fine-tuning refines the general and chaotic knowledge in pre-trained models, we aim to incorporate structural pruning with the fine-tuning, and propose the Pruning-Aware Tuning (PAT) paradigm to eliminate model redundancy while preserving the model performance to the maximum extend. Specifically, we insert the innovative Hybrid Sparsification Modules (HSMs) between the Attention and FFN components to accordingly sparsify the upstream and downstream linear modules. The HSM comprises a lightweight operator and a globally shared trainable mask. The lightweight operator maintains a training overhead comparable to that of LoRA, while the trainable mask unifies the channels to be sparsified, ensuring structural pruning. Additionally, we propose the Identity Loss which decouples the transformation and scaling properties of the HSMs to enhance training robustness. Extensive experiments demonstrate that PAT excels in both performance and efficiency. For example, our Llama2-7b model with a 25\% pruning ratio achieves 1.33$\times$ speedup while outperforming the LoRA-finetuned model by up to 1.26\% in accuracy with a similar training cost. Code: https://github.com/kriskrisliu/PAT_Pruning-Aware-Tuning

Bowei Li, Yang Xu, Ran Zhang et al.

Deep reinforcement learning (DRL) has been extensively applied to Multi-Unmanned Aerial Vehicle (UAV) network (MUN) to effectively enable real-time adaptation to complex, time-varying environments. Nevertheless, most of the existing works assume a stationary user distribution (UD) or a dynamic one with predicted patterns. Such considerations may make the UD-specific strategies insufficient when a MUN is deployed in unknown environments. To this end, this paper investigates distributed user connectivity maximization problem in a MUN with generalization to arbitrary UDs. Specifically, the problem is first formulated into a time-coupled combinatorial nonlinear non-convex optimization with arbitrary underlying UDs. To make the optimization tractable, a multi-agent CNN-enhanced deep Q learning (MA-CDQL) algorithm is proposed. The algorithm integrates a ResNet-based CNN to the policy network to analyze the input UD in real time and obtain optimal decisions based on the extracted high-level UD features. To improve the learning efficiency and avoid local optimums, a heatmap algorithm is developed to transform the raw UD to a continuous density map. The map will be part of the true input to the policy network. Simulations are conducted to demonstrate the efficacy of UD heatmaps and the proposed algorithm in maximizing user connectivity as compared to K-means methods.

Ran Zhang, Miao Wang, Lin X. Cai

Energy-aware control for multiple unmanned aerial vehicles (UAVs) is one of the major research interests in UAV based networking. Yet few existing works have focused on how the network should react around the timing when the UAV lineup is changed. In this work, we study proactive self-remedy of energy-constrained UAV networks when one or more UAVs are short of energy and about to quit for charging. We target at an energy-aware optimal UAV control policy which proactively relocates the UAVs when any UAV is about to quit the network, rather than passively dispatches the remaining UAVs after the quit. Specifically, a deep reinforcement learning (DRL)-based self remedy approach, named SREC-DRL, is proposed to maximize the accumulated user satisfaction scores for a certain period within which at least one UAV will quit the network. To handle the continuous state and action space in the problem, the state-of-the-art algorithm of the actor-critic DRL, i.e., deep deterministic policy gradient (DDPG), is applied with better convergence stability. Numerical results demonstrate that compared with the passive reaction method, the proposed SREC-DRL approach shows a $12.12\%$ gain in accumulative user satisfaction score during the remedy period.