Research map / Architectures and efficiency / CNN design
Deep Lookup Network
Yulan Guo, Longguang Wang, Wendong Mao, Xiaoyu Dong, Yingqian Wang, Li Liu, Wei An
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 |
|---|---|---|---|
| make lookup tables learnable instead of fixed (ablation) | LookupNet · CIFAR-10/CIFAR-100 (ResNet-20) | top-1 accuracy: 91.80/67.77 -> 92.68/68.96 | Improved |
| construct tables with cumulative softmax distributions instead of independent random-init values (ablation) | LookupNet · CIFAR-10 (ResNet-20) | top-1 accuracy: 19.17 -> 92.68 | Improved |
| construct tables with cumulative softmax instead of independent step-initialized values (ablation) | LookupNet · CIFAR-10 (ResNet-20) | top-1 accuracy: 43.52 -> 92.68 | Improved |
| increase table granularity from 17 to 33 (ablation) | LookupNet · CIFAR-10/CIFAR-100 (ResNet-20) | top-1 accuracy: 92.21/67.85 -> 92.68/68.96 | Improved |
| increase table granularity from 33 to 65 (ablation) | LookupNet · CIFAR-10/CIFAR-100 (ResNet-20) | top-1 accuracy: 92.68/68.96 vs. 92.70/68.97 | No clear effect |
| use exponential formulation of scale parameters (ablation) | LookupNet · CIFAR-10/CIFAR-100 (ResNet-20) | top-1 accuracy: 92.35/68.12 -> 92.47/68.51 | Improved |
| add gradient re-scaling across table cells (ablation) | LookupNet · CIFAR-10/CIFAR-100 (ResNet-20) | top-1 accuracy: 92.47/68.51 -> 92.68/68.96 | Improved |
| replace convolutions with lookup layers | ResNet-20 · CIFAR-10/CIFAR-100 | top-1 accuracy: 92.25 -> 92.68; 68.14 -> 68.97 | Improved |
| use lookup operation instead of addition-based AdderNet | ResNet-20 · CIFAR-10/CIFAR-100 | top-1 accuracy: 91.84/67.60 -> 92.68/68.97 | Improved |
| add 4-bit quantization to LookupNet | ResNet-20 · CIFAR-10 and Cortex-A15 energy | accuracy / energy: 92.68 -> 92.51; 75.0 -> 34.5 mJ | Mixed |
| add pruning to LookupNet | ResNet-20 · CIFAR-10 and Cortex-A15 energy | accuracy / energy: 92.68 -> 92.55; 75.0 -> 36.0 mJ | Mixed |
| replace convolutions with lookup layers | VGG-Small · CIFAR-100 | top-1 accuracy: 72.73 -> 75.75 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdderNet — AdderNet: Do We Really Need Multiplications in Deep Learning? (CVPR 2020) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LCNN: Lookup-Based Convolutional Neural Network (CVPR 2017 · similarity 0.87)
- Learning Series-Parallel Lookup Tables for Efficient Image Super-Resolution (ECCV 2022 · similarity 0.85)
- XiNet: Efficient Neural Networks for tinyML (ICCV 2023 · similarity 0.85)
- Learnable Lookup Table for Neural Network Quantization (CVPR 2022 · similarity 0.85)
- Designing Extremely Memory-Efficient CNNs for On-device Vision and Audio Tasks (IJCV 2026 · similarity 0.85)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.85)
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.
