Vahab Mirrokni

Google Research introduces Nested Learning, a new ML paradigm that treats models as interconnected multi-level optimization problems to mitigate catastrophic forgetting in continual learning tasks.

“VP and Google Fellow at Google Research, co-author of the Nested Learning paper presented at NeurIPS 2025”

“Co-authored research on the sleep framework for continual learning in large language models.”

“Google researcher who co-authored paper on Sleep paradigm for continual learning in language models.”

“VP and Google Fellow at Google Research who introduced TurboQuant and related vector quantization algorithms.”

“VP and Google Fellow at Google Research who co-introduced TurboQuant quantization algorithms enabling massive compression for large language models.”

“Co-authored research on TurboQuant vector compression and extreme AI efficiency.”

Google Research introduces Titans architecture and MIRAS framework that enable AI models to process extremely long contexts with linear computational scaling by combining RNN speed with Transformer accuracy through test-time memorization.

“Co-introduced the Titans architecture and MIRAS framework for helping AI models have long-term memory.”