RecSys veterans on repeated data overfitting: 'that's why we only train one epoch'
thomasahle · x · 2026-10-11
Joey Lee flagged that engram/n-gram approaches may markedly amplify overfitting under data repetition. RecSys practitioner YouJiacheng quipped in response: "that's why we only train one epoch" — pointing out the recommendation-systems community's long-standing single-epoch training habit reflects hard-won experience with data repetition issues.
More from Models
- Users spot unreleased 'Opus 5.5' model option on Anthropic's platform — koltregaskes · 2026-10-11
- Abliterated GLM-5.3-Flash EXL3 4bpw quant trends on Hugging Face — Mia-AiLab · 2026-10-11
- Google's Embedding Gemma 2 called the 'DeepSeek moment' for embedding models — HankYeomans · 2026-10-11
- Rumor: Anthropic's Fable 5.5 to land weeks ahead of its IPO, claims big leap — kimmonismus · 2026-10-11
- LLMs Reward Information Gain — Just Like the Best Humans Do — sanderssays · 2026-10-11
- Bilingual devs: OpenAI's models still crush rivals at code-switching mid-sentence — bigdickzillionaire · 2026-10-11