IEEE

MC#: Mixture Compressor for Mixture-of-Experts Large Models

Abstract: Mixture-of-Experts (MoE) has emerged as an effective and efficient scaling mechanism for large language models (LLMs) and vision-language models (VLMs). By expanding a single feed-forward ...
IEEE

Multi-Scale Feature Fusion Module for Change Detection in Remote Sensing Images

Abstract: Change detection refers to an identification of pixels that have undergone semantic changes among two remote sensing images captured at different times in a same area. The problem statement ...
GitHub

MOTIF: Learning Action Motifs for Few-shot Cross-Embodiment Transfer

MOTIF is a framework for few-shot cross-embodiment robotic transfer. It learns reusable action motifs that enable efficient policy generalization across different robot embodiments under limited ...
GitHub

models_with_mask_scale_multiprompt_moe_memory.py

# LICENSE file in the root directory of this source tree. self.memMatrix = nn.Parameter(torch.zeros(num_memory, memory_dim)) # M,C self.keyMatrix = nn.Parameter(torch.zeros(num_memory, memory_dim)) # ...