A critical review of Simple Additive Scoring (SAS) and Simple Multiplicative Scoring (SMS) in education, healthcare, GIS, and human development, showing where simple aggregation creates dangerous logic failures.

A GPU vs. CPU performance study of the Sobel edge detection algorithm using OpenMP, CUDA, and OpenMP GPU offload on Perlmutter's A100 nodes, with CUDA achieving 91.4% occupancy at optimal configuration.

Using the LIKWID performance monitoring tool to correlate L2/L3 cache accesses and retired instruction counts with DGEMM runtime, revealing that naive implementations can incur up to 600x more L3 accesses than CBLAS.

Benchmarking basic, vectorized, OpenMP-parallel, and CBLAS vector matrix multiplication on Perlmutter, finding compiler vectorization delivers a 4-5x speedup and 4 threads is the optimal OpenMP thread count.

Evaluating naive, blocked-with-copy (BMMCO), and CBLAS DGEMM implementations on Perlmutter CPU nodes, showing how cache-aware blocking improves performance from 123 MFLOP/s to over 4600 MFLOP/s.

A performance study comparing direct, vector, and indirect summation on the Perlmutter supercomputer, showing serial array access is roughly 100x faster than random access.