Official code of MonarchGRU (IEEE AICAS 2026): a compact GRU with Monarch weights for always-on keyword spotting, 97.9% on Google Speech Commands v2.
-
Updated
Oct 16, 2026 - Python
Official code of MonarchGRU (IEEE AICAS 2026): a compact GRU with Monarch weights for always-on keyword spotting, 97.9% on Google Speech Commands v2.
Compact deep neural networks via structured transforms (circulant, Toeplitz, skew-circulant, low-rank, ACDC) — a shared C++ core exposed as TensorFlow and Caffe layers.
[NeurIPS'24] BLAST: Block-Level Adaptive Structured Matrices for Efficient Deep Neural Network Inference
Caffe fork providing the plugin interface for Tamp's structured-transform layers (circulant, Toeplitz, skew-circulant, low-rank, ACDC) that compress deep networks — see github.com/Wenri/Tamp.
To associate your repository with the structured-matrices topic, visit your repo's landing page and select "manage topics."