Research map / Architectures and efficiency / CNN design
A Discriminative Gaussian Mixture Model with Sparsity
Hideaki Hayashi, Seiichi Uchida
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 SDGM instead of the compared kernel Bayesian classifiers | SDGM · Ripley, Waveform, Banana, Titanic, Breast Cancer | error rate (%): 9.1 vs 9.1; 10.1 vs 9.6; 10.6 vs 10.5; 22.6 vs 22.6; 29.4 vs 29.9 | Mixed |
| Use sparse Bayesian learning to obtain sparse weights | SDGM · Ripley, Waveform, Banana, Titanic, Breast Cancer | number of nonzero weights: SDGM vs RVM: 6 vs 4; 11.0 vs 14.6; 11.1 vs 11.4; 74.5 vs 65.3; 15.7 vs 6.3 | Mixed |
| Replace CNN softmax with SDGM | CNN · MNIST (D=2) | error rate (%): 3.19 -> 1.81 | Improved |
| Replace CNN softmax with SDGM | CNN · MNIST (D=10) | error rate (%): 1.01 -> 0.86 | Improved |
| Replace CNN softmax with SDGM | CNN · Fashion-MNIST | error rate (%): 8.78 -> 8.05 | Improved |
| Replace CNN softmax with SDGM | DenseNet-40 · CIFAR-10 | error rate (%): 11.07 -> 10.46 | Improved |
| Replace CNN softmax with SDGM | DenseNet-40 · CIFAR-100 | error rate (%): 22.99 -> 21.32 | Improved |
| Replace MobileNet softmax with SDGM | MobileNet · ImageNet | error rate (%): 33.45 -> 32.40 | Improved |
| Add sparse Bayesian learning to the discriminative GMM (ablation) | CNN classifier · MNIST (D=10), CIFAR-10 | error rate (%): 0.72 -> 0.86; 10.05 -> 10.46 | Hurt |
| Add sparse Bayesian learning to the discriminative GMM (ablation) | CNN classifier · MNIST (D=2), Fashion-MNIST, CIFAR-100, ImageNet | error rate (%): 2.43 -> 1.81; 8.30 -> 8.05; 21.93 -> 21.32; 34.75 -> 32.40 | Improved |
| Increase the initial number of mixture components (ablation) | SDGM · Synthetic data | error rate | Improved |
| Increase initial components beyond eight (ablation) | SDGM · Synthetic data | test error rate | No clear effect |
Closest papers by meaning
- Learning Gaussian Mixtures with Generalized Linear Models: Precise Asymptotics in High-dimensions (NeurIPS 2021 · similarity 0.88)
- Efficient Deep Approximation of GMMs (NeurIPS 2019 · similarity 0.86)
- Bayesian Sparse Gaussian Mixture Model for Clustering in High Dimensions (JMLR 2025 · similarity 0.85)
- Robust Model Selection and Nearly-Proper Learning for GMMs (NeurIPS 2022 · similarity 0.85)
- Minimax Optimal Early-Stopped Gradient Descent for Gaussian Mixture Classification (NeurIPS 2026 · similarity 0.83)
- Toward Global Convergence of Gradient EM for Over-Paramterized Gaussian Mixture Models (NeurIPS 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.
