Research map / Optimization / Optimizers
Proximal Mean-Field for Neural Network Quantization
Thalaiyasingam Ajanthan, Puneet K. Dokania, Richard Hartley, Philip H. S. Torr
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 |
|---|---|---|---|
| use PMF instead of BinaryConnect | VGG-16 binary network · CIFAR-100 | top-1 test accuracy: 43.70 -> 61.52 | Improved |
| use PMF instead of BinaryConnect | VGG-16 binary network · CIFAR-10 | top-1 test accuracy: 86.40 -> 90.51 | Improved |
| use PMF instead of BinaryConnect | ResNet-18 binary network · TinyImageNet | top-1 test accuracy: 49.33 -> 51.00 | Improved |
| use PMF instead of ProxQuant | ResNet-18 binary network · CIFAR-100 | top-1 test accuracy: 68.35 -> 71.85 | Improved |
| use PMF instead of BinaryConnect | LeNet-5 · MNIST | top-1 test accuracy: 99.30 -> 99.44 | No clear effect |
| add softmax projection (noisy, annealed) instead of sparsemax projection (ablation) | PMF · CIFAR-100 VGG-16 | top-1 test accuracy: 57.83 -> 61.52 | Improved |
| add softmax projection instead of sparsemax projection (ablation) | PMF · CIFAR-10 ResNet-18 | top-1 test accuracy: 92.60 -> 92.73 | Improved |
| add softmax projection with annealed beta instead of hardmax (proximal ICM) (ablation) | PMF · CIFAR-100 VGG-16 | top-1 test accuracy: 45.65 -> 61.52 | Improved |
| add softmax projection instead of hardmax (ablation) | PMF · TinyImageNet ResNet-18 | top-1 test accuracy: 49.66 -> 51.00 | Improved |
| use proximal ICM (hardmax, u-space) instead of BinaryConnect | binary network · CIFAR-10 / CIFAR-100 / TinyImageNet | top-1 test accuracy: 86.40 -> 88.96 | Improved |
| use PMF on CIFAR-100 ResNet-18 instead of VGG-16 (gap to float reference) | PMF · CIFAR-100 | top-1 test accuracy: 61.52 -> 71.85 | Improved |
Closest papers by meaning
- And the Bit Goes Down: Revisiting the Quantization of Neural Networks (ICLR 2020 · similarity 0.89)
- Towards the Limit of Network Quantization (ICLR 2017 · similarity 0.89)
- Relaxed Quantization for Discretized Neural Networks (ICLR 2019 · similarity 0.89)
- Training Quantized Nets: A Deeper Understanding (NeurIPS 2017 · similarity 0.89)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.88)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.88)
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.
