Research map / Learning theory and algorithms / Kernels and regression theory
Learning and Memorization
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 |
|---|---|---|---|
| Increase LUT network depth (layers of 1024 LUTs, k=2, 20 to 25 hidden layers) (ablation) | Memorization LUT network · Pairwise-MNIST | test accuracy: test accuracy continues to improve for deeper networks (16 or 32 hidden layers) | Helped |
| Increase LUT size k (fan-in) from 2 to 12 on real data (ablation) | Memorization LUT network · Binary-MNIST | test accuracy: 0.66 at k=2 to 0.90 at k=12, then falls to 0.66 at k=16 | Mixed |
| Increase depth by adding hidden layers (5 layers of 1024 LUTs) versus layer-wise accuracy (ablation) | Memorization LUT network · Binary-MNIST | per-layer training accuracy: layer mean accuracy rises from 0.6055 (layer 1) to 0.8808 (layer 5) | Helped |
| Train on random labels instead of real labels with k=12 (ablation) | Memorization LUT network · Binary-MNIST | test accuracy: 0.90 on real labels vs 0.51 on random labels | Hurt |
| Memorization with LUT network vs. lookup table of full input size | Memorization · Binary-MNIST | test accuracy: network reaches 0.87 test accuracy; a full 784-input lookup table would not generalize | Helped |
| Use LUT memorization with best k per task rather than CIFAR-10 networks; on Pairwise-CIFAR-10 average test accuracy | Memorization LUT network · Pairwise-CIFAR-10 | average test accuracy: average 0.76 with range 0.61 to 0.85 | No clear effect |
| Augment LUT lookup with 1-NN search at test time (ablation) | Memorization LUT network · Binary-CIFAR-10 | test accuracy | No clear effect |
Closest papers by meaning
- The Pitfalls of Memorization: When Memorization Hurts Generalization (ICLR 2025 · similarity 0.91)
- Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise (NeurIPS 2026 · similarity 0.90)
- Generalizablity of Memorization Neural Network (NeurIPS 2024 · similarity 0.90)
- A mathematical theory of balancing relational generalization and memorization (NeurIPS 2026 · similarity 0.89)
- To Grok or not to Grok: Disentangling Generalization and Memorization on Corrupted Algorithmic Datasets (ICLR 2024 · similarity 0.88)
- Fisher information and the geometry of memorization in neural networks (NeurIPS 2026 · 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.