The rapid growth of video data has created an urgent need for efficient video understanding systems that can handle long and continuous visual inputs, posing significant challenges for Multimodal Large Language Models (MLLMs) due to severe memory bottlenecks. Moreover, existing compression methods often overlook the ex...
Yong-Hui Niu, Yu-Hao Ji, Chang Nie et al.· 0 citations
A deep reinforcement learning framework is introduced that combines Proximal Policy Optimization and Soft Actor-Critic with a Transformer-based state encoder to predict workload phase transitions rather than merely respond to them and indicates that temporally aware storage control can improve the efficiency of distrib...
H. Baghdasaryan, H. Astsatryan· Scientific Reports· 0 citations
The proposed GRASP algorithm outperforms Comyco and other cutting-edge DRL-based ABR methods under a variety of network situations and reduces the stall ratio compared to Comyco up to 67% and 70% for 5G and 4G networks respectively.
Mira is proposed, an algorithm-system co-design that enables high-capacity MoE inference on a single GPU by coupling predictive expert management with a tailored quantization format, and reduces expert-induced stalls.
CtrlCache is proposed, a training-free control-aware caching framework that adapts computation to the current control sequence that achieves DiT-backbone speedups without model retraining while improving WBench Overall scores over original inference across all three models.
Shang-Ye Song, Dong Gong, Hong Jia et al.· 0 citations
Experimental results demonstrate that the proposed ESM algorithm achieves well-balanced multi-objective performance for mobile video streaming under hardware resource constraints, and the achieved optimization gains will fluctuate slightly across different combinations of video content and network conditions.