Research map / Architectures and efficiency / CNN design
More Is Less: A More Complicated Network With Less Inference Complexity
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 LCCLs to ResNet-18 convolutional layers | ResNet-18 · ImageNet validation | Speed-up: 34.6% | Helped |
| Add LCCLs to ResNet-18 convolutional layers | ResNet-18 · ImageNet validation | Top-1 error: 30.02 -> 33.67 | Hurt |
| Add LCCLs to selected ResNet-34 convolutional layers | ResNet-34 · ImageNet validation | Speed-up: 24.8% | Helped |
| Add LCCLs to selected ResNet-34 convolutional layers | ResNet-34 · ImageNet validation | Top-1 error: 26.58 -> 27.01 | Hurt |
| Add weight-sharing LCCLs | WRN-52-1 · CIFAR-100 | Top-1 error: 29.88 -> 29.55 | Helped |
| Use after-activation inputs for both collaborative layers (ablation) | ResNet-20 LCCN · CIFAR-10 | Top-1 error: 8.32 vs 8.71; 11.62; 12.85 | Helped |
| Use after-activation inputs for both collaborative layers (ablation) | ResNet-20 LCCN · CIFAR-10 | Speed-up: 34.9% vs 24.1%; 39.8%; 55.4% | Mixed |
| Use a shared k×k×C×1 kernel instead of a 1×1×C×T kernel (ablation) | ResNet-20 LCCN · CIFAR-10 | Top-1 error: 8.57 -> 8.32 | Helped |
| Use a shared k×k×C×1 kernel instead of a 1×1×C×T kernel (ablation) | ResNet-20 LCCN · CIFAR-10 | Speed-up: 20.3% -> 34.9% | Helped |
| Add BN before ReLU in the LCCL (ablation) | ResNet-20 LCCN · LCCL feature maps | Sparsity: 0.0% -> 38.8% | Helped |
| Add an L1L2 regularizer to LCCL outputs (ablation) | CIFAR-10 ResNet-20 LCCN · CIFAR-10 | Classification performance | Hurt |
| Use before-activation rather than after-activation LCCL inputs (ablation) | ResNet-20 LCCN · CIFAR-10 | Top-1 error: 8.32 -> 11.62; 12.85 | Hurt |
Closest papers by meaning
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.89)
- Doubly Convolutional Neural Networks (NeurIPS 2016 · similarity 0.88)
- LCNN: Lookup-Based Convolutional Neural Network (CVPR 2017 · similarity 0.88)
- Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear Units (ICML 2016 · similarity 0.88)
- Faster CNNs with Direct Sparse Convolutions and Guided Pruning (ICLR 2017 · similarity 0.88)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · 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.