Research map / Learning theory and algorithms / Combinatorial and search optimization
A Solver-free Framework for Scalable Learning in Neural ILP Architectures
Yatin Nandwani, Rishabh Ranjan, Mausam ., Parag Singla
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 |
|---|---|---|---|
| replace RRN neural baseline with ILP-Loss | Visual Sudoku 9x9 · Visual Sudoku 9x9 | board accuracy (%): 71.1 -> 98.3 | Improved |
| replace RRN with ILP-Loss | Symbolic Sudoku 9x9 · Symbolic Sudoku 9x9 | board accuracy (%): 91.3 -> 100.0 | Improved |
| compare ILP-Loss with RRN on small boards | Visual Sudoku 4x4 · Visual Sudoku 4x4 | board accuracy (%): 99.8 -> 99.7 | No clear effect |
| replace SATNet reasoning layer with ILP-Loss | Visual Sudoku 9x9 · Visual Sudoku 9x9 | board accuracy (%): 17.8 -> 98.3 | Improved |
| replace CombOptNet with ILP-Loss | Sudoku · Symbolic Sudoku 9x9 | board accuracy (%): 0.0 -> 100.0 | Improved |
| replace CombOptNet with ILP-Loss | Random constraints Binary, 8 constraints · Random constraints Binary | vector accuracy (%): 63.4 -> 87.8 | Improved |
| replace CombOptNet with ILP-Loss | Random constraints Dense, 8 constraints · Random constraints Dense | vector accuracy (%): 2.0 -> 41.5 | Improved |
| replace CombOptNet with ILP-Loss | Random constraints Binary, 1 constraint · Random constraints Binary | vector accuracy (%): 97.6 -> 97.8 | No clear effect |
| replace CombOptNet with ILP-Loss | Random constraints Dense, 8 constraints · Random constraints Dense | training time (min): 48.2 -> 20.6 | Improved |
| replace CombOptNet with ILP-Loss | Knapsack N=30 · Knapsack from sentences | vector accuracy (%): 0.0 -> 28.4 | Improved |
| replace CombOptNet with ILP-Loss | Knapsack N=10 · Knapsack from sentences | training time (min): 41.0 -> 44.0 | No clear effect |
| replace CombOptNet with ILP-Loss | Knapsack N=30 · Knapsack from sentences | training time (min): 153.0 -> 111.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CombOptNet — CombOptNet: Fit the Right NP-Hard Problem by Learning Integer Programming Constraints (ICML 2021) · better on 10 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Constraints-Guided Diffusion Reasoner for Neuro-Symbolic Learning (AAAI 2026 · similarity 0.85)
- Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning (ICML 2026 · similarity 0.85)
- Beyond What Seems Necessary: Hidden Gains from Scaling Training-Time Reasoning Length under Outcome Supervision (NeurIPS 2026 · similarity 0.84)
- Causal language modeling can elicit search and reasoning capabilities on logic puzzles (NeurIPS 2024 · similarity 0.84)
- Primal-Dual Neural Algorithmic Reasoning (ICML 2025 · similarity 0.84)
- MLPs Learn In-Context on Regression and Classification Tasks (ICLR 2025 · similarity 0.84)
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.
