Aravind Putrevu
Devrel
SCOPRI, CONNETTI, ESPANDO
Pronti per iniziare a usare l''intelligenza artificiale? Microsoft Reactor fornisce eventi, formazione e risorse della community per aiutare startup, imprenditori e sviluppatori a costruire il loro prossimo business sulla tecnologia IA. Unisciti a noi!
SCOPRI, CONNETTI, ESPANDO
Pronti per iniziare a usare l''intelligenza artificiale? Microsoft Reactor fornisce eventi, formazione e risorse della community per aiutare startup, imprenditori e sviluppatori a costruire il loro prossimo business sulla tecnologia IA. Unisciti a noi!
26 febbraio, 2024 | 12:00 PM - 1:00 PM (UTC) Ora universale coordinata
Argomento: Data Science & Machine Learning
Lingua: italiano
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.
Altoparlanti
Per domande, contattaci all''indirizzo reactor@microsoft.com