Research map / Architectures and efficiency / CNN design
TREC: Transient Redundancy Elimination-based Convolution
Jiawei Guan, Feng Zhang, Jiesong Liu, Hsin-Hsuan Sung, Ruofan Wu, Xiaoyong Du, Xipeng Shen
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 |
|---|---|---|---|
| replace Deep Reuse's random-hash clustering with TREC (differentiable LSH inside the convolution) | CifarNet · CIFAR-10 | top-1 accuracy (%): 73.2 ∼76.1 -> 76.5 | Improved |
| replace Deep Reuse with TREC | ZfNet (pruned) · CIFAR-10 | top-1 accuracy (%): 72.5 ∼76.6 -> 78.9 | Improved |
| replace Deep Reuse with TREC | Vanilla SqueezeNet · CIFAR-10 | top-1 accuracy (%): 79.8 ∼81.9 -> 83.0 | Improved |
| replace Deep Reuse with TREC | ResNet-34 · ImageNet 64x64 | top-1 accuracy (%): 46.7 ∼49.9 -> 52.2 | Improved |
| replace Deep Reuse with TREC | SqueezeNet + Complex Bypass · CIFAR-10 | top-1 accuracy (%): 80.5 ∼83.1 -> 84.6 | Improved |
| replace Deep Reuse with TREC | five CNNs · MCU inference time | average time per image: 154.44 -> 153.92 | No clear effect |
| replace conventional convolution with TREC | CifarNet · CIFAR-10 on MCU | average time per image (ms): 217.32 -> 153.92 | Improved |
| replace conventional convolution with TREC | ZfNet (pruned) · CIFAR-10 on MCU | average time per image (ms): 3557.32 -> 814.01 | Improved |
| replace conventional convolution with TREC | ResNet-34 · ImageNet 64x64 on MCU | average time per image (ms): 4242.77 -> 1378.75 | Improved |
| replace conventional convolution with TREC | ResNet-34 · ImageNet 64x64 | top-1 accuracy (%): 52.6 -> 52.2 | Hurt |
| replace conventional convolution with TREC | CifarNet · CIFAR-10 | top-1 accuracy (%): 78.2 -> 76.5 | Hurt |
| use dynamic sigma tied to H instead of static sigma 10^-4 in the Gaussian (ablation) | TREC · CIFAR-10 layers (CifarNet, ZfNet, SqueezeNet) | accuracy | Improved |
Closest papers by meaning
- ESPACE: Accelerating Convolutional Neural Networks via Eliminating Spatial and Channel Redundancy (AAAI 2017 · similarity 0.84)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.84)
- Accelerate CNN via Recursive Bayesian Pruning (ICCV 2019 · similarity 0.83)
- Convolutional Neural Network Pruning With Structural Redundancy Reduction (CVPR 2021 · similarity 0.83)
- Faster CNNs with Direct Sparse Convolutions and Guided Pruning (ICLR 2017 · similarity 0.83)
- Inference, Learning and Attention Mechanisms that Exploit and Preserve Sparsity in CNNs (IJCV 2020 · similarity 0.83)
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.
