Quoc Le

Le grew up in rural Vietnam in a home without electricity, but he lived near a library, where he read extensively about great inventions. Around age 14, he decided that humanity would be helped most by a machine smart enough to be an inventor in its own right. He graduated from Quoc Hoc Hue High School in Hue.⁠[1][1][2]

In 2004, Le began studying artificial intelligence and machine learning at the Australian National University (ANU), working under Alex Smola at ANU and NICTA, Australia’s national information and communications technology research center, and earned a Bachelor of Software Engineering (First Class Honors), having been named a Distinguished Scholar while there; during this period he was also a research visitor at Schölkopf’s department at the Max Planck Institute for Biological Cybernetics in Tübingen, Germany.⁠[3][4][2][4]

He then pursued a PhD at the AI Lab of Stanford University’s Computer Science Department, advised by Andrew Ng. Machine learning software at the time required people to annotate data and specify features by hand, which Le found deeply frustrating; deep learning using networks of simulated neurons was progressing slowly, and he found a way to speed it up significantly by building simulated neural networks 100 times larger that could process thousands of times more data, an approach that attracted Google’s attention.⁠[4][1][1][1][1][1]

In 2011, Le joined Google Brain, working with Andrew Ng and testing large-scale deep learning methods under his guidance.⁠[3][1]

On June 26, 2012, Google announced that its team had spread the computation of an artificial neural network across 16,000 CPU cores and trained models with more than 1 billion connections; one neuron in the network learned to detect cats from still frames of unlabeled YouTube videos alone. The experiments were led by Le and reported at the ICML conference that week, in the paper “Building High-level Features Using Large Scale Unsupervised Learning,” with Le as first author. After these results became public, they sparked a race at Facebook, Microsoft, and other companies to invest in deep learning; as of 2014, the technique was used in Google’s image search and speech recognition software. In 2022, the paper received an ICML Test of Time Honorable Mention.⁠[5][5][5][6][1][1][7]

In 2013, Le left Stanford and formally became a research scientist at Google. He helped build deep learning-based systems such as speech recognition on Android phones and automatic tagging of photos on the web.⁠[4][3][8]

In May 2014, Le and fellow Googler Tomas Mikolov published the paper “Distributed Representations of Sentences and Documents,” proposing “paragraph vectors,” a method for representing whole paragraphs of text as mathematical vectors that can be used for tasks such as sentiment analysis.⁠[9][9][9][8][8]

In August 2014, at 32, Le was named to MIT Technology Review’s “Innovators Under 35” list.⁠[1][1]

In 2014, Le and fellow Googlers Ilya Sutskever and Oriol Vinyals published “Sequence to Sequence Learning with Neural Networks,” which used a multilayered Long Short-Term Memory (LSTM) network, a type of recurrent neural network, to map an input sequence to an output sequence for machine translation, and presented it at the NIPS conference that December. Le said the method can remember the order of the words in a sequence and can deal with variable-sized inputs. In 2024, the paper received the NeurIPS Test of Time Award, with the award committee calling it the cornerstone work that set the encoder-decoder architecture, inspiring later attention-based improvements and today’s foundation model research.⁠[10][8][8][10][8][8][11][11]

On September 27, 2016, Le and Mike Schuster, as research scientists on the Google Brain team, published a post announcing the Google Neural Machine Translation system (GNMT), and Le was also one of the authors of the technical report; Google Translate had by then switched entirely to the system for Chinese-to-English translation, about 18 million translations per day.⁠[12][12][12]

On May 17, 2017, Le and Barret Zoph introduced the approach Google calls “AutoML”: a controller neural net proposes a “child” model architecture, which is trained and evaluated, with the results fed back to the controller to improve its next proposal, a process repeated thousands of times. Their paper “Neural Architecture Search with Reinforcement Learning” was published at ICLR 2017. The machine learning product Cloud AutoML, later launched by Google, is also based on his ideas.⁠[13][13][13][13][14][3]

In an interview with the Vietnamese newspaper Tuoi Tre in February 2019, Le was one of three project leaders at Google Brain, managing five to six projects related to facial recognition, speech recognition, and natural language processing, with a team of 25 people.⁠[3][3][3]

On May 29, 2019, Le, then a principal scientist at Google AI, and Mingxing Tan released EfficientNet, proposing a method that uses a compound coefficient to uniformly scale the width, depth, and resolution of convolutional neural networks.⁠[15][15]

On January 28, 2020, Google released Meena, a 2.6 billion parameter end-to-end neural conversational model, with the team thanking Le and others for their leadership support of the project.⁠[16][16][16]

On May 11, 2022, Google researchers introduced “chain of thought prompting,” a method for improving the reasoning abilities of language models; Le was one of the collaborators on the project.⁠[17][17][17]

On January 17, 2024, Google DeepMindGoogle DeepMindGoogle DeepMind is Google’s artificial intelligence research organization, formed in April 2023 by merging DeepMind, which Google acquired in 2014, with the Brain team from Google Research, and led by CEO Demis Hassabis. It developed AlphaGo, AlphaFold, and the Gemini family of models, and Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for AlphaFold’s protein structure prediction work.Open the full entry and New York University published AlphaGeometry in Nature, with Le among the authors. The system solved 25 of 30 Olympiad geometry problems within the standard time limit, compared with 10 for the previous best system and an average of 25.9 for human gold medalists.⁠[18][18][18]

On August 5, 2026, Le, Jeff Dean, Sanjay Ghemawat, and Oriol Vinyals announced that they were leaving Google to co-found Discovery Loop, a public benefit corporation that uses AI to automate complete experimental loops, initiating and iterating thousands of experiments simultaneously to accelerate engineering and scientific discovery, starting with the automation of machine learning research. The initial funding round was co-led by Radical VenturesRadical VenturesRadical Ventures is an artificial intelligence-focused venture capital firm founded by Jordan Jacobs and Tomi Poutanen shortly after the 2017 launch of the Vector Institute in Toronto. It invests from incubation and seed through growth and pre-IPO stages, and its portfolio includes companies such as Cohere, Waabi, and World Labs. In 2026, it co-led the investment in Discovery Loop, founded by Jeff Dean and others, and completed the first close of its late-stage Radical Breakouts Fund.Open the full entry and Khosla VenturesKhosla VenturesKhosla Ventures (KV) is an American venture capital firm founded in 2004 by Vinod Khosla, the founding CEO of Sun Microsystems, and headquartered in Menlo Park, California. It made heavy early bets on clean technology, became OpenAI’s first venture capital investor in 2019, and its portfolio also includes companies such as DoorDash, Block (formerly Square), Impossible Foods, and Rocket Lab.Open the full entry, with participation from Kleiner Perkins, Lightspeed, and Doerr Capital. Le said he was very excited about automating machine learning: “It might be that we will discover a different transformer architecture.”⁠[19][19][19][19][19][20][19][21]

Le serves as a trustee of Fulbright University Vietnam.⁠[2]

Discovery Loop: Official website: https://www.discoveryloop.com/ (opens in a new window).⁠[20]

Fulbright University Vietnam: Official website: https://fulbright.edu.vn/ (opens in a new window).⁠[2]

Personal website (Stanford): https://cs.stanford.edu/~quocle/ (opens in a new window)⁠[4]