Research map / Learning theory and algorithms / Clustering
Clustered Influence Functions
Miklós Máté Badó, Kristian Fenech
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 |
|---|---|---|---|
| Cache clustered IHVP solves and reuse across folds instead of per-query IF | CiF, LeNet · CIFAR-10 K=50+100 | speedup: 32.1 h IF vs 2.1 h CiF, 15.3× | Improved |
| Cache clustered IHVP solves instead of per-query IF | CiF, AlexNet · CIFAR-10 K=50+100 | speedup: 63.3 h IF vs 5.4 h CiF, 11.7× | Improved |
| Increase cache size from C=10 to C=50 (ablation) | CiF, LeNet · CIFAR-10 K=50+100 | speedup: 15.3× -> 3.0× | Hurt |
| Increase cache size from C=10 to C=50 (ablation) | CiF, LeNet/AlexNet · CIFAR-10 K=50 | Pearson agreement: 0.943 -> 0.942; 0.994 -> 0.994 | No clear effect |
| Increase cache size from C=10 to C=50 (ablation) | CiF, ResNet-18 · CIFAR-10 K=50 | Pearson agreement: 0.740 -> 0.935 | Improved |
| Increase cache size from C=128 to C=512 (ablation) | CiF, DeiT-Tiny · ImageNet-1k K=5000 | rank correlation rho: 0.07 -> 0.19 | Improved |
| Increase cache size from C=128 to C=512 (ablation) | CiF, DeiT-Tiny · ImageNet-1k K=5000 | speedup: 32.9× -> 9.3× | Hurt |
| Add CiF cache on top of LoGra backend | LoGra · CIFAR-10 ResNet-18 online queries | mean time per query: 8.68 ms -> 3.45 ms | Improved |
| Aggregate gradients by clusters for backdoor detection with fixed-location trigger | CiF (C=80) · MNIST backdoor, fixed trigger | poison recall | Improved |
| Aggregate gradients by clusters for backdoor detection with random-location trigger | CiF (C=80) · MNIST backdoor, random trigger | poison recall | Hurt |
Closest papers by meaning
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.86)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.85)
- Rescaled Influence Functions: Accurate Data Attribution in High Dimension (NeurIPS 2025 · similarity 0.84)
- DataInf: Efficiently Estimating Data Influence in LoRA-tuned LLMs and Diffusion Models (ICLR 2024 · similarity 0.84)
- Data-Efficient Learning via Clustering-Based Sensitivity Sampling: Foundation Models and Beyond (ICML 2024 · similarity 0.84)
- LoRIF: Low-Rank Influence Functions for Scalable Training Data Attribution (NeurIPS 2026 · similarity 0.83)
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.
