Reproduce “Taskonomy: Disentangling Task Transfer Learning” 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

Amir Zamir, Alexander Sax, William Shen, Leonidas Guibas, Jitendra Malik, Silvio Savarese

Taskonomy measures whether learned visual task representations can support other tasks, then selects transfers under a supervision budget. It trains small readouts on frozen source features and uses measured affinities to choose connections. For one set of 10 tasks, it uses roughly two-thirds fewer labeled datapoints with performance nearly the same as independent training; this finding concerns labeled data, not computation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.