Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a groundbreaking programming language designed to revolutionize tensor computation in this 15-minute conference presentation from OOPSLA 2025. Learn how Finch addresses the fundamental limitations of existing tensor frameworks that can only handle dense rectilinear integer grids, while real-world tensors often contain complex underlying structures like sparsity, runs of repeated values, or symmetry. Discover how this innovative language combines flexible control flow with diverse data structures into a unified representation that enables co-optimization of both elements. Understand the programming model that allows developers to write familiar code using loops, statements, conditionals, and breaks while working with a wide variety of tensor structures including sparsity, run-length-encoding, symmetry, triangles, padding, and blocks. Examine how Finch automatically specializes control flow to data structures, enabling performance engineers to focus on algorithm experimentation rather than low-level optimization details. See demonstrated performance improvements in critical operations such as sparse matrix-vector multiplication (SpMV), sparse general matrix multiplication (SpGEMM), image processing, and graph analytics. Gain insights into how the language reliably utilizes key structural properties like structural zeros, repeated values, and clustered non-zeros to achieve dramatic speedups. Access comprehensive research materials including the full academic paper, supplementary archives with reproducible artifacts, and detailed experimental results that validate the approach's effectiveness across multiple computational domains.
