Research map / Architectures and efficiency / Pruning and efficient tuning
Compressing Neural Networks using the Variational Information Bottleneck
Read the paperCheck 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 |
|---|---|---|---|
| Use VIBNet neuron pruning instead of L0-sep | LeNet-300-100 · MNIST | rW (%): 10.01 -> 3.59 | Helped |
| Use VIBNet neuron pruning instead of L0 | LeNet-300-100 · MNIST | error (%): 1.4 -> 1.6 | Hurt |
| Use VIBNet neuron pruning instead of BC-GHS | LeNet-5-Caffe · MNIST | FLOP (Mil): 0.153 -> 0.094 | Helped |
| Use VIBNet neuron pruning instead of SBP | LeNet-5-Caffe · MNIST | error (%): 0.9 -> 1.0 | Hurt |
| Use VIBNet instead of BC-GHS | VGG-16 · CIFAR10, BC protocol | FLOP (Mil): 121.9 -> 70.63 | Helped |
| Use VIBNet instead of PF | VGG-16 · CIFAR10, PF/SBP protocol | error (%): 6.6 -> 6.5 | Helped |
| Use VIBNet instead of NS-Best | VGG-16 · CIFAR10, NS protocol | error (%): 5.9 -> 6.2 | Hurt |
| Use VIBNet instead of RNP | VGG-16 · CIFAR100, RNP protocol | FLOP (Mil): 160 -> 133.6 | Helped |
| Use VIBNet instead of NS-Best | VGG-16 · CIFAR100, NS protocol | error (%): 26.0 -> 25.9 | No clear effect |
| Remove the ψi,j term from the KL approximation (ablation) | VIBNet · Compression experiments | compression performance | No clear effect |
| Increase the trade-off parameter γ | VIBNet · Compression experiments | compression | Helped |
Closest papers by meaning
- Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition (NeurIPS 2021 · similarity 0.90)
- Adaptive Estimators Show Information Compression in Deep Neural Networks (ICLR 2019 · similarity 0.90)
- Auto-Compressing Networks (NeurIPS 2025 · similarity 0.88)
- Information Bottleneck Analysis of Deep Neural Networks via Lossy Compression (ICLR 2024 · similarity 0.88)
- Does Compression Imply Generalization? A Minimum Description Length Perspective (NeurIPS 2026 · similarity 0.87)
- Forget the Data and Fine-Tuning! Just Fold the Network to Compress (ICLR 2025 · similarity 0.87)
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.