Research map / Architectures and efficiency / Pruning and efficient tuning
MP-ISMoE: Mixed-Precision Interactive Side Mixture-of-Experts for Efficient Transfer Learning
Yutong Zhang, Zimeng Wu, Shengcai Liao, Shujiang Wu, Jiaxin 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 GNP-IQ (mixed-precision backbone) alone (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 534.8 | Hurt |
| add GNP-IQ alone (ablation) | UniPT · Flickr30K | training memory (G): 24.4 -> 18.4 | Improved |
| add ISMoE alone (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 544.3 | Improved |
| add ISMoE alone (ablation) | UniPT · Flickr30K | training memory (G): 24.4 -> 32.4 | Hurt |
| combine GNP-IQ and ISMoE (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 543.1 | Improved |
| use low unified precision backbone (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 529.3 | Hurt |
| use mixed precision backbone (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 531.4 | Hurt |
| add Gaussian noise perturbation to mixed precision (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 531.4 -> 534.8 | Improved |
| add sparse MoE structure (ablation) | UniPT · Flickr30K (UniPT base, VSE∞) | Rsum: 537.4 -> 543.1 | Improved |
| add cross-network correlation guided expert selection (ablation) | MoE side network · Flickr30K (UniPT base, VSE∞) | Rsum: 543.1 -> 544.3 | Improved |
| add MP-ISMoE on UniPT | UniPT · GLUE avg (T5-base) | average score: 84.3 -> 84.8 | Improved |
| add MP-ISMoE on SHERL | SHERL · GLUE avg (T5-large) | average score: 87.5 -> 88.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UniPT — UniPT: Universal Parallel Tuning for Transfer Learning with Efficient Parameter and Memory (CVPR 2024) · better on 5 of 5 shared results
- SHERL — SHERL: Synthesizing High Accuracy and Efficient Memory for Resource-Limited Transfer Learning (ECCV 2024) · better on 5 of 5 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
- MxMoE: Mixed-precision Quantization for MoE with Accuracy and Performance Co-Design (ICML 2025 · similarity 0.87)
- Quant Experts: Token-aware Adaptive Error Reconstruction with Mixture of Experts for Large Vision-Language Models Quantization (CVPR 2026 · similarity 0.87)
- Multilinear Mixture of Experts: Scalable Expert Specialization through Factorization (NeurIPS 2024 · similarity 0.87)
- DeRS: Towards Extremely Efficient Upcycled Mixture-of-Experts Models (CVPR 2025 · similarity 0.86)
- Mixture Compressor for Mixture-of-Experts LLMs Gains More (ICLR 2025 · similarity 0.85)
- MoEQuant: Enhancing Quantization for Mixture-of-Experts Large Language Models via Expert-Balanced Sampling and Affinity Guidance (ICML 2025 · 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.
