New paper proposes solver-admitted metamorphic testing for learned optimization surrogates
An arXiv paper shows that equivalent encodings of an optimization problem can yield different predictions from learned surrogates, and proposes metamorphic testing that accounts for numerical factors when interpreting these differences.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.AI
Solver-Admitted Metamorphic Testing of Learned Network Optimization Surrogates