Research map / Architectures and efficiency / Quantization
Simultaneously Optimizing Weight and Quantizer of Ternary Neural Network Using Truncated Gaussian Approximation
Check 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 |
|---|---|---|---|
| add gradient correctness scaling to the STE (ablation) | Ternary ResNet-20 · CIFAR-10 | accuracy: 87.89% -> 90.39% | Helped |
| use Adam rather than vanilla SGD for threshold optimization (ablation) | Ternary ResNet-20 · CIFAR-10 | accuracy: 90.39% -> 56.31% | Hurt |
| initialize thresholds to 0.05 max(|w|) rather than 0.1 max(|w|) (ablation) | Ternary ResNet-20 · CIFAR-10 | accuracy: 89.96% vs 90.24% | No clear effect |
| initialize thresholds to 0.15 max(|w|) rather than 0.1 max(|w|) (ablation) | Ternary ResNet-20 · CIFAR-10 | accuracy: 90.12% vs 90.24% | No clear effect |
| ternarize both the first and last layers using the proposed trainable-threshold method | ResNet-18b · ImageNet validation | top-1 accuracy: 69.75 -> 66.01 | Hurt |
| ternarize both the first and last layers using the proposed trainable-threshold method | ResNet-34b · ImageNet validation | top-1 accuracy: 73.31 -> 70.79 | Hurt |
| ternarize both the first and last layers using the proposed trainable-threshold method | ResNet-50b · ImageNet validation | top-1 accuracy: 76.13 -> 73.97 | Hurt |
Closest papers by meaning
- Trained Ternary Quantization (ICLR 2017 · similarity 0.90)
- Harmonious Coexistence of Structured Weight Pruning and Ternarization for Deep Neural Networks (AAAI 2020 · similarity 0.88)
- RTN: Reparameterized Ternary Network (AAAI 2020 · similarity 0.87)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.86)
- HADAMRNN: BINARY AND SPARSE TERNARY ORTHOGONAL RNNS (ICLR 2025 · similarity 0.86)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · 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.