DBTB INT Shubho Sengupta r
summary
Argues that large-scale supervised learning wins when data scale matches model size and gives specifics: convolutional nets with hundreds of millions of parameters trained on what are claimed as the largest datasets at Baidu produce rapid gains and often reach or exceed human-level accuracy. Emphasizes an experimental workflow — train models every day, run many experiments, iterate on data and architectures — and credits sheer dataset size and repeated training runs for most progress.
AI-written summary · about
related talks
36mWhen the Only Way to Scale up, is to Scale Down
20mFoundation Models for Big Data and Prompt Tuning
10mIs Bigger Really Better? Predictive Analytics with Fine-grained Behavior Data
17mFocus on Data, on the Road to AI
42mTraining AI To Code Using The Largest Code Dataset (Project CodeNet)
21mDataEngBytes 2023 - SYD-T2-04 - Suneeta Mall
48mProcessing Terabytes of Data in Credit Scoring
41mChallenges of Machine Learning at Scale