Research map / Architectures and efficiency / CNN design
ARMA Nets: Expanding Receptive Field for Dense Prediction
Jiahao Su, Shiqi Wang, Furong Huang
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 ConvLSTM convolutions with ARMA layers | ConvLSTM, kernel size 3 · Moving-MNIST-2 | PSNR: 18.24 -> 19.72 | Improved |
| Add non-local attention blocks (ablation) | ConvLSTM, kernel size 5 · Moving-MNIST-2 | PSNR: 19.58 -> 19.18 | Hurt |
| Add non-local attention blocks (ablation) | ConvLSTM, kernel size 3 · Moving-MNIST-2 | PSNR: 18.24 -> 19.45 | Improved |
| Add non-local attention blocks (ablation) | ARMA-LSTM, kernel size 3 · Moving-MNIST-2 | PSNR: 19.72 -> 19.62 | No clear effect |
| Replace convolution layers with ARMA layers | AlexNet · CIFAR10 | accuracy: 86.30 ± 0.29 -> 85.67 ± 0.19 | Hurt |
| Replace convolution layers with ARMA layers | VGG-11 · CIFAR10 | accuracy: 91.57 ± 0.59 -> 91.57 ± 0.73 | No clear effect |
| Replace convolution layers with ARMA layers | ResNet-18 · CIFAR10 | accuracy: 95.01 ± 0.15 -> 95.07 ± 0.13 | No clear effect |
| Replace convolution layers with ARMA layers | AlexNet · CIFAR100 | accuracy: 58.99 ± 0.37 -> 57.43 ± 0.24 | Hurt |
| Replace convolution layers with ARMA layers | VGG-11 · CIFAR100 | accuracy: 68.25± 0.11 -> 68.36 ± 1.67 | No clear effect |
| Replace convolution layers with ARMA layers | ResNet-18 · CIFAR100 | accuracy: 73.71 ± 0.23 -> 73.72 ± 0.52 | No clear effect |
| Use stability-guaranteeing re-parameterization (ablation) | ARMA-VGG-11 · CIFAR-10 | training convergence | Improved |
| Replace U-Net convolutions with ARMA layers | U-Net · ISIC 2018 lesion segmentation | segmentation metrics | Improved |
Closest papers by meaning
- Recurrence along Depth: Deep Convolutional Neural Networks with Recurrent Layer Aggregation (NeurIPS 2021 · similarity 0.85)
- Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions (ICLR 2018 · similarity 0.85)
- Learning Implicitly Recurrent CNNs Through Parameter Sharing (ICLR 2019 · similarity 0.85)
- ReX: An Efficient Approach to Reducing Memory Cost in Image Classification (AAAI 2022 · similarity 0.85)
- Progressive Supernet Training for Efficient Visual Autoregressive Modeling (CVPR 2026 · similarity 0.85)
- Scalable Pre-training of Large Autoregressive Image Models (ICML 2024 · 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.
