Research map / Architectures and efficiency / Pruning and efficient tuning
MST-compression: Compressing and Accelerating Binary Neural Networks with Minimum Spanning Tree
Quang Hieu Vo, Linh-Tam Tran, Sung-Ho Bae, Lok-Won Kim, Choong Seon Hong
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 K-mean reordering with MST reordering in the hardware design (ablation) | BNN FPGA accelerator · CIFAR-10 FPGA | FPS/LUTs: 1.01 -> 1.27 | Improved |
| use MST instead of shortest Hamiltonian path to reorder computation (ablation) | ResNet-20 binary conv layers · ResNet-20 layer 1 | exploration time: 31.3 s -> 0.021 s | Improved |
| use MST instead of K-mean for reordering (ablation) | ResNet-20 binary conv layers · ResNet-20 layer 1 | Bit-Ops: 1.33 M -> 1.07 M | Improved |
| apply MST reordering directly to pretrained BNN (no retraining) | ReCU VGG-small · CIFAR-10 | Top-1 accuracy: 92.4 -> 93.3 | No clear effect |
| add MST distance loss during training (Ours 1) on ImageNet ResNet-18 | ReCU ResNet-18 · ImageNet | Top-1 accuracy: 61.0 -> 57.0 | Hurt |
| increase MST loss weight lambda from 1e-7 to 1e-5 (ablation) | VGG-small · CIFAR-10 | Top-1 accuracy: 92.17 -> 90.17 | Hurt |
| increase lambda from 1e-7 to 1e-5 (ablation) | VGG-small · CIFAR-10 | Params (Mbit): 1.391 -> 0.422 | Improved |
| use a single initial center instead of 8 (ablation) | MST learning · CIFAR-10 VGG-small | MST depth: 73.0 -> 22.3 | Improved |
| vary number of centers from 1 to 8 (ablation) | MST learning · CIFAR-10 VGG-small | Top-1 accuracy: 91.49 -> 91.49 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- ReCU — ReCU: Reviving the Dead Weights in Binary Neural Networks (ICCV 2021) · better on 4 of 4 shared results
- Adabin — AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets (ECCV 2022) · better on 1 of 2 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
- Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition (NeurIPS 2021 · similarity 0.86)
- "Lossless" Compression of Deep Neural Networks: A High-dimensional Neural Tangent Kernel Approach (NeurIPS 2022 · similarity 0.86)
- Coreset-Based Neural Network Compression (ECCV 2018 · similarity 0.86)
- Data-Dependent Coresets for Compressing Neural Networks with Applications to Generalization Bounds (ICLR 2019 · similarity 0.85)
- Sub-Bit Neural Networks: Learning To Compress and Accelerate Binary Neural Networks (ICCV 2021 · similarity 0.85)
- Data-Independent Neural Pruning via Coresets (ICLR 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.
