Reproduce “UniDEC : Unified Dual Encoder and Classifier Training for Extreme Multi-Label Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siddhant Kharbanda, Devaansh Gupta, Gururaj K, Pankaj Malhotra, Amit Singh, Cho-Jui Hsieh, Rohit Babbar
UniDEC jointly trains a dual encoder and classifier for extreme multi-label retrieval, computing loss over a selected batch-level label shortlist instead of every label. On LF-AmazonTitles-1.3M, it uses one 48-GB GPU and 3,000 labels, versus 16 GPUs and 1.3 million labels for DEXML-Full. It improves top-five scores, but not top-one scores, and takes longer: 78 versus 66 hours.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.