Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning with Algorithmic Supervision via Continuous Relaxations
Felix Petersen, Christian Borgelt, Hilde Kuehne, Oliver Deussen
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 |
|---|---|---|---|
| Train with relaxed Bubble Sort rather than a task-specific sorting relaxation | sorting supervision · 4-digit MNIST sorting, n=3, n=5, and n=7 | exact matches: 0.928 -> 0.944; 0.811 -> 0.842; 0.656 -> 0.707 | Improved |
| Use the relaxed Bellman-Ford algorithm instead of Perturbed Opt. | Warcraft shortest-path supervision · Warcraft shortest-path | cost ratio: 1.00042 -> 1.00014 | Improved |
| Use the relaxed Bellman-Ford algorithm instead of Perturbed Opt. | Warcraft shortest-path supervision · Warcraft shortest-path | exact match (EM): 91.7% -> 88.4% | Hurt |
| Replace SoftRas with the relaxed silhouette renderer | single-view ShapeNet reconstruction · ShapeNet, batch size 2 | mean 3D IoU: 0.5455 -> 0.5256; 0.5455 -> 0.5349 | Hurt |
| Use relaxed Levenshtein distance supervision instead of the L1-encoding baseline | EMNIST character classification · AB, BC, CD, DE, EF, IL, OX, ACGT, OSXL | Top-1 accuracy: .616 -> .671; .651 -> .807; .768 -> .816; .739 -> .833; .701 -> .847; .550 -> .570; .893 -> .960; .403 -> .437; .448 -> | Improved |
| Use relaxed Levenshtein distance supervision instead of the L1-encoding baseline | EMNIST character classification · AB, BC, CD, DE, EF, IL, OX, ACGT, OSXL | F1-score: .581 -> .666; .629 -> .805; .759 -> .815; .711 -> .831; .674 -> .845; .490 -> .539; .890 -> .960; .336 -> .367; .384 -> | Improved |
| Use the directed Euclidean distance renderer instead of the three-edge renderer (ablation) | relaxed silhouette supervision · ShapeNet, batch size 2 | mean 3D IoU: 0.5256 -> 0.5349 | Improved |
| Use the three-edge renderer instead of the directed Euclidean distance renderer (ablation) | relaxed silhouette supervision · ShapeNet, batch size 2 | runtime: faster by a factor of 3 | Improved |
Closest papers by meaning
- Neural Algorithmic Reasoning Without Intermediate Supervision (NeurIPS 2023 · similarity 0.85)
- On the Theory of Continual Learning with Gradient Descent for Neural Networks (ICML 2026 · similarity 0.85)
- Discrete Neural Algorithmic Reasoning (ICML 2025 · similarity 0.85)
- Train and Test Tightness of LP Relaxations in Structured Prediction (ICML 2016 · similarity 0.85)
- Algorithm Development in Neural Networks: Insights from the Streaming Parity Task (ICML 2025 · similarity 0.84)
- On The Concurrence of Layer-wise Preconditioning Methods and Provable Feature Learning (ICML 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.
