Linear Algebraic Abduction with Partial Evaluation
summary
Recasts propositional Horn abduction as matrix operations and applies partial evaluation (unfolding via powers of a reductive program matrix) to collapse multi-step derivations into fewer multiplications; a running example reduces iterations from four to two. Implementation uses Python numpy (dense) and scipy.sparse with sparse CSR/COO formats, and minimal-hitting-set enumeration via an off‑the‑shelf solver. Benchmarks on artificial and FMEA-based datasets show large speedups for sparse cases but costly matrix powering in dense mode.
AI-written summary · about
chapters (8)
- 0:00Intro
- 0:41Outline
- 1:32Overview and Preliminaries
- 5:08Linear Algebraic Computation of Abduction
- 10:37Partial evaluation
- 20:24Experimental Results - Original benchmark
- 21:55Experimental Results Enhanced benchmark datasets
- 24:11Conclusion
related talks
29mPartial Evaluation of Automatic Differentiation for Differential-Algebraic Equati...
18mLearning Abstraction Selection for Bayesian Program Analysis
17mAlgebro-geometric Algorithms for Template-Based Synthesis of Polynomial Program...
15mGauss: Program Synthesis by Reasoning Over Graphs
47mCausal Matrix Completion: Applications to Offline Causal Reinforcement Learning
22mTowards Verified Linear Algebra Programs Through Equivalence
19mCompiler Support for Sparse Tensor Convolutions
01h 05mIntroduction to Relational Algorithms