Linear Regression Can Also Experience Grokking
RichmanRonald · x · 2026-07-14
This referenced paper and thread discuss an interesting conclusion: grokking is not unique to deep neural networks.
The author points out that classic grokking can be observed even in simple, linear, and over-parameterized ridge regression:
- Perfect fitting is achieved on the training set very early on
- But generalization performance suddenly improves dramatically much later
This suggests that grokking might reflect more general optimization and generalization dynamics, rather than just being a quirk of specific neural networks.
More from Research
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26
- The paper behind the EMNLP 2026 keynote on bigram 'simulated learners' — najoungkim · 2026-10-26