Benchmark for matrix multiplications between dense and block sparse (BSR) matrix in TVM, blocksparse (Gray et al.) and cuSparse.

Benchmarking code for “PopSparse: Accelerated block sparse matrix multiplication on IPU”

Check Benchmark each operation from tridiagonal sparse size 2^5 to 2^15 with block 16.

PACE: Automated λ selection for SINDy/STLSQ using a six-stage Pareto–AICc consensus pipeline. Combines Pareto elbow detection, h-block cross-validation, and AICc arbitration within a candidate region. Demonstrates 71% exact sparsity recovery across 200 benchmark trials. MATLAB R2023a.