Taejong Joo is a Research Scientist at Google.
How media typically covers Taejong Joo
Referenced in coverage
Random masking of parameter updates in adaptive optimizers like RMSProp outperforms state-of-the-art optimizers; a new method called Magma reduces perplexity by 19% for 1B models compared to Adam.
“Taejong Joo is first author on Google research showing random masking can outperform adaptive optimizers for LLM training.”