Research map / Learning theory and algorithms / Fairness
SMiLE: Provably Enforcing Global Relational Properties in Neural Networks
Matteo Francobaldi, Michele Lombardi, Andrea Lodi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| enforce global robustness with SMiLE instead of property-agnostic training | base CNN · MNIST zero-vs-rest, delta=0.100 | PGD accuracy (%): 66.80 -> 92.17 | Improved |
| use SMiLE instead of CROWN-IBP for robustness | CROWN-IBP · MNIST zero-vs-rest, delta=0.100 | verified accuracy (%): 99.00 -> 90.57 | Hurt |
| use SMiLE instead of CROWN-IBP | CROWN-IBP · MNIST | training time (s): 14,138s -> 7,717s | Improved |
| use SMiLE instead of CROWN-IBP | CROWN-IBP · MNIST test-set predict and verify | inference time (s): 11.20s -> 0.03s | Improved |
| enforce fairness with SMiLE instead of CertiFair | CertiFair · Compas, Law, Crime | accuracy at equal or better fairness | Improved |
| enforce fairness with SMiLE instead of property-agnostic training | Agnostic · Compas, Law, Crime | counterfactual variation | Improved |
| enforce monotonicity with SMiLE on non-monotonic synthetic functions | SMiLE · 9 synthetic monotonic tasks | R2: 0.99 down to 0.83 | Mixed |
| replace original SMiLE pretraining with box-degeneracy-aware pretraining loss | SMiLE · relational properties | training stability | Improved |
| replace PGD training loop of original SMiLE with dual ascent and unrestricted projection | SMiLE · relational properties | convergence and generator calls | Improved |
| use timeout-extension scheme in counterexample generation instead of complete search | SMiLE generator · relational properties | training time | Improved |
Closest papers by meaning
- REGLO: Provable Neural Network Repair for Global Robustness Properties (AAAI 2024 · similarity 0.86)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.84)
- SMILE: Bridging Continuous Optimization and Discrete Symbolic Recovery (NeurIPS 2026 · similarity 0.84)
- Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules (NeurIPS 2025 · similarity 0.83)
- Achieving Domain-Independent Certified Robustness via Knowledge Continuity (NeurIPS 2024 · similarity 0.83)
- Relational Verification Leaps Forward with RABBit (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
