AnalysisAI ModelsNovember 11, 2026

Podcast: why AlexNet's 60M-parameter design shouldn't have worked

Read original source →youtube.com

Machine Learning Street Talk talks with Simon Prince about AlexNet's 60 million parameters — orders of magnitude more than contemporary vision models — trained from random weights on a 150,000-dimensional input space. Prince calls its ImageNet fit "jaw-dropping, really."

People · Simon Prince

1 source

More stories today

Open the live feed