Greg Corrado

Gregory Corrado is an American computer scientist and neuroscientist, known for co-founding the Google Brain in 2011. A Distinguished Scientist at Google, he led the company's Health AI division and oversaw the application of artificial intelligence in healthcare. His work spans AI, computational neuroscience, and large-scale machine learning.

Early life and education Corrado received a Bachelor of Arts in physics from Princeton University in 1999. He later attended Stanford University, where he earned a Master of Science in computer science and a Ph.D. in neuroscience. His doctoral research focused on systems and computational neuroscience, specializing in the mathematical modeling of value-based decision-making, with concentration in artificial intelligence. A major early achievement came in 2012 when the team built a neural network using 16,000 processors that learned to autonomously identify cats from unlabeled YouTube video stills.

At Google, Corrado has been instrumental in integrating AI into consumer products. He helped lead the development of RankBrain, a machine learning system launched in 2015 to improve Google Search results, which Corrado identified as the third most important ranking signal. He also guided the creation of Smart Reply, which uses neural networks to generate automated email responses in Gmail.

Corrado was also involved in the development and open-sourcing of key Google technologies, including the TensorFlow machine learning framework and word2vec, an influential algorithm for creating word embeddings.

In the late 2010s, Corrado shifted his focus to the application of artificial intelligence in medicine, eventually becoming Head of Health AI at Google Research in 2016. In this role, he led research into using AI for medical diagnostics, particularly in medical imaging, predicting clinical outcomes, and analyzing genomic and sensor data. Under his leadership, the Health AI team developed several notable projects. These include the Automated Retinal Disease Assessment (ARDA), a tool designed for detecting diabetic retinopathy, and the Med-PaLM series of medical large language models. while its successor, Med-PaLM 2, achieved 86.5% accuracy on the MedQA benchmark. Another multimodal model, Med-Gemini, reached 91.1% accuracy on the same benchmark.

Selected publications

Awards and recognition In 2023, Corrado was a recipient of the NeurIPS Test of Time Award for his co-authorship of the 2013 paper that introduced word2vec.