Research map / Architectures and efficiency / Quantization
ShiftAddNet: A Hardware-Inspired Deep Network
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 |
|---|---|---|---|
| combine shift and add layers (ShiftAddNet) instead of add-only AdderNet | ResNet-20 · CIFAR-10 adaptation | accuracy (%): 79.79 -> 81.50 | Helped |
| combine shift and add layers instead of shift-only DeepShift | ResNet-20 · CIFAR-10 fine-tuning | accuracy (%): 51.31 -> 84.61 | Helped |
| freeze the shift layers in ShiftAddNet (ablation) | ShiftAddNet · CIFAR-10 adaptation | accuracy (%): 81.50 -> 85.10 | Helped |
| freeze the shift layers in ShiftAddNet (ablation) | ShiftAddNet · CIFAR-10 fine-tuning | accuracy (%): 84.61 -> 84.88 | No clear effect |
| freeze the shift layers in ShiftAddNet (ablation) | ShiftAddNet · CIFAR-10 adaptation/fine-tuning | training energy (MJ): 16.82 -> 11.04 | Helped |
| freeze the shift layers in ShiftAddNet (ablation) | ShiftAddNet · VGG19-small CIFAR-100, FIX8 | accuracy: up to +10.8% | Helped |
| include shift layers alongside add layers when pruning add layers by 50% (ablation) | ResNet-20 · CIFAR-10 | test accuracy (%): 51.47 -> 80.42 | Helped |
| randomly prune the fixed shift layers up to 70% (ablation) | ShiftAddNet · CIFAR-10 ResNet-20 convergence | convergence speed | No clear effect |
| ShiftAddNet with fixed shift layers vs AdderNet on accuracy (ablation) | ShiftAddNet (Fixed) · Fig. 4 classification/IoT | accuracy: -3.74% ~ +31.2% | Mixed |
Closest papers by meaning
- An Efficient Matrix Multiplication Algorithm for Accelerating Inference in Binary and Ternary Neural Networks (ICML 2025 · similarity 0.86)
- An Empirical Study of Distributed Deep Learning Training on Edge (Student Abstract) (AAAI 2024 · similarity 0.85)
- ShiftAddLLM: Accelerating Pretrained LLMs via Post-Training Multiplication-Less Reparameterization (NeurIPS 2024 · similarity 0.85)
- Practical Performance Guarantees for Pipelined DNN Inference (ICML 2024 · similarity 0.85)
- Towards training digitally-tied analog blocks via hybrid gradient computation (NeurIPS 2024 · similarity 0.85)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 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.