Reproduce “Exponential Family Attention” 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
Kevin Christian Wibisono, Yixin Wang
The authors introduce exponential family attention (EFA) to estimate a missing city temperature from other cities’ readings at the same time. It learns attention weights over all other cities and uses a Gaussian distribution, rather than a fixed set of nearest neighbors. On held-out observations for 44 U.S. cities, one-layer EFA had lower test MSE than the best tested fixed-neighbor model, 17.135 versus 22.685, for reconstruction, not forecasting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.