home.social

#textclassification — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #textclassification, aggregated by home.social.

  1. PEFT with AL selects more valuable data, preserves stable early‑layer representations, reduces forgetting, and outperforms FFT in low‑resource settings. hackernoon.com/why-peft-beats- #textclassification

  2. Experiments show PEFT, especially Prefix‑tuning and UniPELT, outperform FFT in low‑resource text tasks and remain strong in AL setups, boosted further by TAPT. hackernoon.com/who-learns-fast #textclassification

  3. This paper studies active learning with parameter‑efficient fine‑tuning (adapters), showing AL+PEFT improves PLMs in low‑resource text classification. hackernoon.com/teaching-big-mo #textclassification

  4. Just compared Claude-Sonnet-3.5 with OpenAI's o1 on a CLS task – classifying text inputs from US short stories with regard to focalization. Turns out, Sonnet doesn't recognize zero focalization and achieved an F1-score of 0.47, while o1 performed better with 0.69. Not bad - but problematic, as the hidden tokens of the optimizer (?) from o1 would be of particular interest.

    #CLS #AI #ClaudeSonnet #OpenAI's_o1 #TextClassification #Focalization

  5. I've started making a text classification library in Go. It's VERY WIP, but I think that there could be some interesting results once I add things to it. (and maybe redesign the entirety of it if needed in the future)

    Source code: codeberg.org/autodni/silt

    #Go #GoLang #Language #NLP #Classification #TextClassification

  6. Anyone use a machine learning api in their app? I had trouble finding a dead simple, cost effective one and ended up creating my own for #textclassification . Curious if other people would get value out of it.

    #ai #serverless #buildinpublic