1 min read
TFFM and Exponential Machines

Exponential Machines use a tensor-train representation to model high-order feature interactions without explicitly materializing an exponentially large parameter tensor. The project connects tensor methods with practical supervised learning.

TFFM is an open-source TensorFlow implementation of arbitrary-order factorization machines. It supports sparse and dense inputs, GPU execution, and experimentation with interaction orders beyond traditional pairwise factorization machines.

Read the paper or view TFFM on GitHub.