Research map / Label-efficient and robust learning / Recognition and long-tail
Aggregated Learning: A Vector-Quantization Approach to Learning Neural Network Classifiers
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 |
|---|---|---|---|
| Aggregate two inputs rather than classify one input at a time | ResNet-18 · CIFAR-10 | error rate (%): 5.08 -> 4.89 | Helped |
| Aggregate two inputs rather than classify one input at a time | ResNet-18 · CIFAR-100 | error rate (%): 23.7 -> 23.03 | Helped |
| Add mutual-information regularization (α=0.3) to fold-2 (ablation) | ResNet-18 AgrLearn · CIFAR-10 and CIFAR-100 | error rate (%): 4.89 -> 4.73; 23.03 -> 22.94 | Helped |
| Add mutual-information regularization (α=0.7) to fold-1 (ablation) | ResNet-18 · CIFAR-10 and CIFAR-100 | error rate (%): 5.08 -> 4.92; 23.7 -> 23.7 | Mixed |
| Aggregate two inputs rather than classify one input at a time | WideResNet-22-10 · CIFAR-10 and CIFAR-100 | error rate (%): 4.3 -> 4.19; 21.13 -> 20.3 | Helped |
| Add mutual-information regularization (α=0.3) to fold-2 (ablation) | WideResNet-22-10 AgrLearn · CIFAR-10 and CIFAR-100 | error rate (%): 4.19 -> 4.18; 20.3 -> 20.28 | Helped |
| Aggregate two inputs rather than classify one input at a time | ResNet-34 · CIFAR-10 and CIFAR-100 | error rate (%): 4.94 -> 4.68; 23.86 -> 22.63 | Helped |
| Add mutual-information regularization (α=0.3) to fold-2 (ablation) | ResNet-34 AgrLearn · CIFAR-10 and CIFAR-100 | error rate (%): 4.68 -> 4.65; 22.63 -> 22.25 | Helped |
| Aggregate two sentences in each model input | CNN-sentence · Movie Review and Subjectivity | accuracy (%): 76.1 -> 79.3; 90.01 -> 93.5 | Helped |
| Aggregate two sentences in each model input | LSTM · Movie Review and Subjectivity | accuracy (%): 76.2 -> 77.8; 90.2 -> 92.1 | Helped |
| Double the number of filters per layer (ablation) | ResNet-18 AgrLearn, fold-2, α=0.3 · CIFAR-10 and CIFAR-100 | error rate (%): 4.73 -> 4.3; 22.94 -> 21.78 | Helped |
| Double the number of filters per layer (ablation) | ResNet-34 AgrLearn, fold-2, α=0.3 · CIFAR-10 and CIFAR-100 | error rate (%): 4.65 -> 4.45; 22.25 -> 21.68 | Helped |
Closest papers by meaning
- Information Bottleneck Learning Using Privileged Information for Visual Recognition (CVPR 2016 · similarity 0.86)
- Structured IB: Improving Information Bottleneck with Structured Feature Learning (AAAI 2025 · similarity 0.85)
- Learning Optimal Representations with the Decodable Information Bottleneck (NeurIPS 2020 · similarity 0.85)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.85)
- Information Bottleneck Analysis of Deep Neural Networks via Lossy Compression (ICLR 2024 · similarity 0.84)
- I-Con: A Unifying Framework for Representation Learning (ICLR 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.