LEARN, CONNECT, BUILD

Microsoft Reactor

Microsoft Reactor에 가입하고 개발자와 라이브 참여

AI 및 최신 기술을 시작할 준비가 되셨나요? Microsoft Reactor는 개발자, 기업가 및 신생 기업이 AI 기술 등을 기반으로 구축하는 데 도움이 되는 이벤트, 교육 및 커뮤니티 리소스를 제공합니다. 참여하세요.

LEARN, CONNECT, BUILD

Microsoft Reactor

Microsoft Reactor에 가입하고 개발자와 라이브 참여

돌아가기

Improving Large Language Model by Systematically Improving its Data

26 2월, 2024 | 12:00 오후 - 1:00 오후 (UTC) Koordinierte Weltzeit (UTC)

서식:
Livestream

항목: Data Science und maschinelles Lernen

언어: Englisch

Labeled data powers AI/ML in the enterprise, but real-world datasets have been found to contain between 7-50% annotation errors. Imperfectly labelled text data hampers ML models' training (and evaluation) across tasks like intent recognition, entity recognition, and sequence generation. Although pretrained LLMs are equipped with a lot of world knowledge, their performance is adversely affected by noisy training data (as noted by OpenAI).

In this talk, we illustrate data-centric techniques to mitigate the effect of label noise without changing any code related to model architecture, hyperparameters, or training. These data quality improvement techniques should thus remain applicable even for future advanced LLMs like GPT-10.

스피커