Full Professor, University of Jena
1 paper at NeurIPS 2025
Evaluating real-world einsum expressions can lead to hyper-sparse intermediate tensors that are hard to predict, we exploit this dynamic sparsity in a hybrid algorithm.
