Research map / Optimization / Optimizers
Stochastic Multiple Choice Learning for Training Diverse Deep Ensembles
Read the paperCheck 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 independent ensemble with sMCL | VGG16+LSTM captioning ensemble · MSCOCO M=5 | oracle CIDEr-D: 0.831 -> 0.922 | Helped |
| sMCL vs MCL | captioning ensemble · MSCOCO M=5 | oracle CIDEr-D: 0.852 -> 0.922 | Helped |
| sMCL vs Dey et al. | captioning ensemble · MSCOCO M=5 | oracle CIDEr-D: 0.910 -> 0.922 | Helped |
| fine-tune CNN jointly with sMCL | captioning ensemble · MSCOCO M=5 | oracle CIDEr-D: 0.922 -> 1.184 | Helped |
| sMCL vs independent ensemble diversity | captioning ensemble · MSCOCO M=5 | unique 4-grams: 10297 -> 15427 | Helped |
| increase pretraining iterations before MCL fine-tuning (ablation) | sMCL k=1 · CIFAR10-Quick | oracle accuracy: 94.21 -> 96.00 | Helped |
| increase pretraining iterations before MCL fine-tuning (ablation) | sMCL k=3 · CIFAR10-Quick | oracle accuracy: 92.25 -> 90.94 | Hurt |
| increase pretraining iterations (ablation) | sMCL k=2 · CIFAR10-Quick | oracle accuracy: 92.79 -> 93.00 | No clear effect |
| lower k (more specialization) from 4 to 1 (ablation) | sMCL fine-tuned · ILSVRC-Alex | oracle accuracy: 70.84 -> 74.67 | Helped |
| lower k from 4 to 1 (ablation) | sMCL fine-tuned · ILSVRC-Alex | ensemble-mean accuracy: 60.09 -> 55.22 | Hurt |
| initialize all members identically then fine-tune with MCL k=1 (ablation) | sMCL · ILSVRC-Alex | oracle accuracy: 56.90 -> 72.67 | Helped |
| sMCL vs hand-designed random label split | sMCL · CIFAR10 | oracle accuracy: 91.83 -> 94.65 | Helped |
Closest papers by meaning
- Confident Multiple Choice Learning (ICML 2017 · similarity 0.85)
- Versatile Multiple Choice Learning and Its Application to Vision Computing (CVPR 2019 · similarity 0.85)
- Partially Stochastic Infinitely Deep Bayesian Neural Networks (ICML 2024 · similarity 0.84)
- A Convergence Theory for Deep Learning via Over-Parameterization (ICML 2019 · similarity 0.84)
- DeepOBS: A Deep Learning Optimizer Benchmark Suite (ICLR 2019 · similarity 0.84)
- Annealed Multiple Choice Learning: Overcoming limitations of Winner-takes-all with annealing (NeurIPS 2024 · 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.