Research map / Label-efficient and robust learning / OOD and anomaly detection
Functional Indirection Neural Estimator for Better Out-of-distribution Generalization
Kha Pham, Hung Le, Man Ngo, Truyen Tran
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 |
|---|---|---|---|
| Use FINE functional memory instead of data-space baselines (Transformer) | FINE-NICE · Omniglot Rotation | test accuracy (%): 64.8 -> 77.6 | Improved |
| Use MLP backbone instead of NICE backbone for non-invertible reflection (ablation) | FINE · Omniglot Reflection | test accuracy (%): 57.7 -> 95.1 | Improved |
| Use MLP backbone instead of NICE backbone for reflection (ablation) | FINE · CIFAR100 Reflection | test accuracy (%): 51.1 -> 80.6 | Improved |
| Add analogy-making and indirection memory to hypernetwork-style weight generation | FINE-MLP vs HyperNet · Omniglot multi affine | test accuracy (%): 54.0 -> 72.0 | Improved |
| Increase training set from 1000 to 10000 points (ablation) | FINE · Omniglot multi-affine | test accuracy (%): 64.4 -> 81.1 | Improved |
| Increase training set from 10,000 to 100,000 points (ablation) | FINE · Omniglot multi-affine | test accuracy (%): 81.1 -> 85.3 | Improved |
| Increase number of memories (from 0 limit case) and NICE backbone layers (ablation) | FINE · Omniglot multi-affine | test accuracy (%) | Improved |
| Replace Group CNN (p4-CNN) encoder with MLP encoder (ablation) | FINE-NICE · CIFAR100 multi-affine | test accuracy (%): 81.6 -> 70.0 | Hurt |
| Test on unseen translation rule vs ESBN on CIFAR100 | FINE · Extreme OOD CIFAR100 Translation | test accuracy (%): 16.4 -> 62 | Improved |
| Train on CIFAR100, test on Omniglot translation (unseen dataset and rule) | FINE · Extreme OOD Translation | test accuracy (%): 30 -> 22.1 | Hurt |
| FINE-NICE vs ESBN on CIFAR100 Swap | FINE-NICE · CIFAR100 Swap | test accuracy (%): 50.5 -> 51.6 | No clear effect |
| FINE vs Transformer on Omniglot Black-white | FINE-NICE · Omniglot B&W | test accuracy (%): 97.6 -> 98.4 | No clear effect |
Closest papers by meaning
- LEVI: Generalizable Fine-tuning via Layer-wise Ensemble of Different Views (ICML 2024 · similarity 0.85)
- Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise (NeurIPS 2026 · similarity 0.84)
- Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning (NeurIPS 2026 · similarity 0.84)
- Function Vectors in Large Language Models (ICLR 2024 · similarity 0.84)
- Data Distributional Properties As Inductive Bias for Systematic Generalization (CVPR 2025 · similarity 0.84)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 2024 · similarity 0.84)
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.
