Tag
The paper introduces MESH-FL, an entropy-guided matrix product state compression framework for multimodal federated learning on edge devices. It adaptively allocates compression ranks per layer and modality, achieving up to 56.8× compression and 2.01% final accuracy improvement over uncompressed FedAvg on a heterogeneous Raspberry Pi cluster.
SeKV is a resolution-adaptive KV cache method that organizes context into entropy-guided semantic spans stored across a GPU-CPU hierarchy, enabling selective token-level reconstruction during decoding while reducing GPU memory by 53.3% versus full caching at 128K context.
Proposes EntMTP, a training-free scheduler that adapts tree-based attention topologies for speculative decoding based on local entropy estimates, achieving 1.09-1.15x speedup over Hydra and up to 1.36x over Medusa.