Announcement_6

Our paper “ARC-Encoder: Learning Compressed Text Representations for Large Language Models” has been accepted at TMLR, with a Featured Certification awarded to very high quality papers (approx. top 3%)! It will be presented in NeurIPS 2026. We introduce ARC-Encoder, a context compression module that produces continuous representations directly understandable by a target decoder. The project page, code, and resources are available here.