Research map / Label-efficient and robust learning / Recognition and long-tail
ClimbQ: Class Imbalanced Quantization Enabling Robustness on Efficient Inferences
Ting-An Chen, De-Nian Yang, Ming-syan Chen
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 |
|---|---|---|---|
| add HomoVar loss on top of ClimbQ (ablation) | ClimbQ · CIFAR-10-LT 2-bit MobileNet-V2, gamma=10 | accuracy (%): 69.94 -> 72.51 | Improved |
| add HomoVar loss on top of ClimbQ (ablation) | ClimbQ · CIFAR-10-LT 2-bit MobileNet-V2, gamma=200 | accuracy (%): 52.47 -> 56.89 | Improved |
| add HomoVar loss on top of ClimbQ (ablation) | ClimbQ · CIFAR-100-LT 2-bit ResNet-56, gamma=200 | accuracy (%): 23.91 -> 27.39 | Improved |
| add HomoVar loss on top of ClimbQ (ablation) | ClimbQ · CIFAR-10-LT 4-bit ResNet-20, gamma=10 | accuracy (%): 80.72 -> 80.58 | No clear effect |
| replace BatchQuant with ClimbQ | 2-bit MobileNet-V2 · CIFAR-10-LT, gamma=10 | accuracy (%): 16.24 -> 69.94 | Improved |
| replace BatchQuant with ClimbQ+ | 2-bit ResNet-56 · CIFAR-100-LT, gamma=10 | accuracy (%): 37.24 -> 49.12 | Improved |
| replace BatchQuant with ClimbQ+ | 2-bit MobileNet-V2 · CIFAR-100-LT, gamma=200 | accuracy (%): 10.69 -> 25.45 | Improved |
| replace Choi et al. with ClimbQ (no HomoVar) | 4-bit MobileNet-V2 · CIFAR-100-LT, gamma=200 | accuracy (%): 28.30 -> 22.58 | Hurt |
| replace Choi et al. with ClimbQ (no HomoVar) | 4-bit MobileNet-V2 · CIFAR-10-LT, gamma=10 | accuracy (%): 80.13 -> 78.91 | Hurt |
| replace best baseline with ClimbQ+ | 2-bit ResNet-20 · Syndigits-LT, gamma=200 | accuracy (%): 58.75 -> 80.55 | Improved |
| replace Focal loss rebalancing with HomoVar loss (ablation) | ClimbQ 2-bit ResNet-20 · CIFAR-10-LT, gamma=200 | accuracy (%): 37.69 -> 61.06 | Improved |
| replace Class-Balanced (CB) rebalancing with HomoVar loss (ablation) | ClimbQ 2-bit ResNet-20 · CIFAR-10-LT, gamma=50 | accuracy (%): 59.94 -> 71.73 | Improved |
Closest papers by meaning
- Understanding the Unfairness in Network Quantization (ICML 2025 · similarity 0.88)
- LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid (ICLR 2025 · similarity 0.87)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.86)
- CLIMB: Class-imbalanced Learning Benchmark on Tabular Data (NeurIPS 2025 · similarity 0.86)
- GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling (NeurIPS 2026 · similarity 0.86)
- Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment (NeurIPS 2025 · 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.
