Research map / Architectures and efficiency / RNNs and normalization
Latent Weights Do Not Exist: Rethinking Binarized Neural Network Optimization
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 Adam with latent weights by Bop | BinaryNet · CIFAR-10 test | top-1 accuracy: 90.9% -> 91.3% | Helped |
| Replace latent weights with Bop | BinaryNet · ImageNet | top-1 accuracy: 40.1% -> 41.1% | Helped |
| Replace latent weights with Bop | BinaryNet · ImageNet | top-5 accuracy: 66.3% -> 65.4% | Hurt |
| Replace latent weights with Bop | XNOR-Net · ImageNet | top-1 accuracy: 44.2% -> 45.9% | Helped |
| Replace latent weights with Bop | XNOR-Net · ImageNet | top-5 accuracy: 69.2% -> 70.0% | Helped |
| Replace latent weights with Bop | BiReal-Net · ImageNet | top-1 accuracy: 56.4% -> 56.6% | Helped |
| Replace latent weights with Bop | BiReal-Net · ImageNet | top-5 accuracy: 79.5% -> 79.4% | Hurt |
| Use a very low adaptivity rate and high threshold (ablation) | Bop · CIFAR-10 | accuracy | Hurt |
| Use a high adaptivity rate and low threshold (ablation) | Bop · CIFAR-10 | training accuracy | Hurt |
| Set the threshold to zero (ablation) | Bop · CIFAR-10 | validation accuracy | Hurt |
Closest papers by meaning
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · similarity 0.89)
- Binarized Neural Networks (NeurIPS 2016 · similarity 0.89)
- How Do Adam and Training Strategies Help BNNs Optimization (ICML 2021 · similarity 0.88)
- Understanding weight-magnitude hyperparameters in training binary networks (ICLR 2023 · similarity 0.88)
- Resilient Binary Neural Network (AAAI 2023 · similarity 0.87)
- Fast and Slow Gradient Approximation for Binary Neural Network Optimization (AAAI 2025 · similarity 0.86)
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.