Research map / Architectures and efficiency / RNNs and normalization
Improving Simple Models with Confidence Profiles
Amit Dhurandhar, Karthikeyan Shanmugam, Ronny Luss, Peder A. Olsen
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 |
|---|---|---|---|
| Weight CIFAR-10 simple ResNets with ProfWeight confidence profiles | SM-3, SM-5, SM-7, and SM-9 · CIFAR-10 holdout test | accuracy: 73.15($\pm$ 0.7) -> 77.52 ($\pm$0.01); 75.78($\pm$0.5) -> 79.25($\pm$0.36); 78.76($\pm$0.35) -> 81.34($\pm$0.49); 79.9($ | Improved |
| Weight CART with ProfWeight AUC confidence profiles | CART · Manufacturing wafer classification test | accuracy: 74.3% -> 87.3% | Improved |
| Weight CART by final-layer confidence only | CART · Manufacturing wafer classification test | accuracy: 74.3% -> 77.1% | Improved |
| Train the simple model by distillation instead of unweighted training | CIFAR-10 simple ResNets · CIFAR-10 holdout test | accuracy: 73.15($\pm$ 0.7) -> 65.84($\pm$0.60); 75.78($\pm$0.5) -> 70.09 ($\pm$0.19); 78.76($\pm$0.35) -> 73.4($\pm$0.64); 79.9($\ | Hurt |
| Use probe confidence AUC starting at unit 13 rather than unit 14 (ablation) | ProfWeight AUC · CIFAR-10 holdout test, SM-3, SM-5, SM-7, and SM-9 | accuracy: 76.56 -> 75.93; 79.25 -> 78.85; 81.34 -> 81.22; 82.42 -> 82.54 | Mixed |
| Use probe confidence AUC starting at unit 15 rather than unit 14 (ablation) | ProfWeight AUC · CIFAR-10 holdout test, SM-3, SM-5, SM-7, and SM-9 | accuracy: 76.56 -> 73.64; 79.25 -> 77.81; 81.34 -> 80.81; 82.42 -> 80.97 | Hurt |
| Weight examples with ProfWeight ReLU rather than AUC (ablation) | CIFAR-10 simple ResNets · CIFAR-10 holdout test | accuracy: 77.52 vs 76.56; 78.24 vs 79.25; 80.16 vs 81.34; 81.65 vs 82.42 | Mixed |
Closest papers by meaning
- Initializing Models with Larger Ones (ICLR 2024 · similarity 0.84)
- Diffusion-based Neural Network Weights Generation (ICLR 2025 · similarity 0.84)
- The Effect of Weight Precision on the Neuron Count in Deep ReLU Networks (ICML 2024 · similarity 0.84)
- Beyond Slow Signs in High-fidelity Model Extraction (NeurIPS 2024 · similarity 0.83)
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.83)
- Towards Scalable and Versatile Weight Space Learning (ICML 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.
