Research map / Probabilistic and scientific modeling / Bayesian optimization
Harmless Overfitting: Using Denoising Autoencoders in Estimation of Distribution Algorithms
Malte Probst, Franz Rothlauf
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 |
|---|---|---|---|
| use DAE-EDA instead of BOA on 5-Traps 100 bit | EDA model · 5-Traps 100 bit | unique evaluations: 190,011 -> 62,599 | Improved |
| use DAE-EDA instead of BOA on onemax | EDA model · ONEMAX100 | unique evaluations: 6,259 -> 7,005 | Hurt |
| use DAE-EDA instead of BOA on NK landscapes | EDA model · NK landscapes | unique evaluations: 32,015 -> 17,472 (NK n=30,k=4,i=1); 67,939 -> 189,613 (NK n=30,k=4,i=2) | Mixed |
| use DAE-EDA instead of BOA on HIFF | EDA model · HIFF | unique evaluations: 11,991 -> 22,066 (HIFF64); 51,008 -> 50,447 (HIFF128) | Mixed |
| use DAE-EDA instead of BOA | EDA model · All problem instances | CPU time (sec): 205,244 -> 2,633 (4-Traps 120 bit) | Improved |
| use DAE-EDA sampling instead of DAE-guided local search | HIFF 256 bit · HIFF 256 bit | average evaluations: 1,355,500 -> 220,650 | Improved |
| lower DAE corruption noise c (weaker regularization) (ablation) | DAE-EDA · 50 bit 5-Traps, N=200 | average generations to best solution: 22.0 -> 30.0 generations (c=0.2 to c=0.05) | Improved |
| lower DAE corruption noise c (weaker regularization) (ablation) | DAE-EDA · 50 bit 5-Traps, N=200 | fitness of best sample (avg best): 43.68 -> 46.52 | Improved |
| lower DAE corruption noise c (weaker regularization) (ablation) | DAE-EDA · 64 bit HIFF, N=200 | fitness of best sample (avg best): 305.10 -> 346.80 | Improved |
| lower DAE corruption noise c (weaker regularization) (ablation) | DAE-EDA · NK n=30, k=4, i=1, N=200 | fitness of best sample (avg best): 22.84 -> 22.92 | No clear effect |
| use DAE-EDA with c=0.05 instead of BOA | EDA model · 50 bit 5-Traps, N=200 | average generations to best solution: 18.7 -> 30.0 | Improved |
| lower DAE corruption noise c (ablation) | DAE-EDA · 50 bit 5-Traps | identity function degree tau | Improved |
Closest papers by meaning
- From Understanding Genetic Drift to a Smart-Restart Mechanism for Estimation-of-Distribution Algorithms (Journal Track) (AAAI 2025 · similarity 0.86)
- Beyond Benign Overfitting in Nadaraya-Watson Interpolators (NeurIPS 2025 · similarity 0.85)
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.84)
- A Unified View on Learning Unnormalized Distributions via Noise-Contrastive Estimation (ICML 2025 · similarity 0.84)
- Benign Overfitting in Out-of-Distribution Generalization of Linear Models (ICLR 2025 · similarity 0.83)
- The Bias-Variance Tradeoff in Data-Driven Optimization: A Local Misspecification Perspective (NeurIPS 2025 · 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.
