Research map / Label-efficient and robust learning / Semi-supervised learning
Generating Accurate Pseudo-Labels in Semi-Supervised Learning and Avoiding Overconfident Predictions via Hermite Polynomial Activations
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · SVHN, 1K labeled | maximum pseudo-label accuracy: 93.3% -> 94.2% | Helped |
| Replace ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · CIFAR10, 4K labeled | maximum pseudo-label accuracy: 84.4% -> 85.5% | Helped |
| Replace ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · SmallNORB, 1K labeled | maximum pseudo-label accuracy: 90.4% -> 92.6% | Helped |
| Replace ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · MNIST, 1K labeled | maximum pseudo-label accuracy: 98.2% -> 98.2% | No clear effect |
| Replace ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · CIFAR10, 4K labeled | test error rate: 10.94 -> 10.65 | Helped |
| Replace ReLU activations with Hermite activations in SaaS | SaaS with ResNet-18 · SVHN, 1K labeled | test error rate: 3.82 -> 3.57 ± 0.04 | Helped |
| Replace ReLU with Hermite activations | ResNet152 · CIFAR10 | best test accuracy: 94.5% -> 95.48% | Helped |
| Replace ReLU with Hermite activations | ResNet152 · CIFAR10 | epochs to reach 90% test accuracy: 80 -> 30 | Helped |
| Replace sigmoid with 4 Hermite polynomials | Deep autoencoder · MNIST | test loss: 7.91 ± 0.14 -> 3.16 ± 0.02 | Helped |
| Use Hermite-SaaS instead of ReLU-SaaS | SaaS with ResNet-18 · CIFAR10, 30% label noise | epochs to best accuracy: ≥600 -> 95 | Helped |
| Add the Tanaka noisy-label post-processing step (ablation) | Hermite-SaaS · CIFAR10, 30% label noise | epochs to best accuracy: 95 -> 61 | Helped |
| Add the Tanaka noisy-label post-processing step (ablation) | Hermite-SaaS · CIFAR10, 10% label noise | epochs to best accuracy: 224 -> 244 | Hurt |
Closest papers by meaning
- Polynomial, trigonometric, and tropical activations (ICLR 2026 · similarity 0.87)
- Neural Networks are Convex Regularizers: Exact Polynomial-time Convex Optimization Formulations for Two-layer Networks (ICML 2020 · similarity 0.85)
- How Universal Polynomial Bases Enhance Spectral Graph Neural Networks: Heterophily, Over-smoothing, and Over-squashing (ICML 2024 · similarity 0.85)
- Identifiability of Deep Polynomial Neural Networks (NeurIPS 2025 · similarity 0.84)
- ReLUs Are Sufficient for Learning Implicit Neural Representations (ICML 2024 · similarity 0.84)
- Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (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.