Martín Abadi

Security and programming languages

Abadi's publication record goes back to 1985 and spans algorithms and theory, security, and privacy. With Michael Burrows and Roger Needham, Abadi wrote A Logic of Authentication (1990), which gave systems designers a logic-based notation for describing and reasoning about authentication protocols; ACM SIGOPS added it to its Hall of Fame in 2007.¹⁶

Abadi's work on programming languages for security includes the applied pi calculus, an extension of the pi calculus for modeling cryptographic protocols, first introduced in a 2001 conference paper and later presented in full with Bruno Blanchet and Cedric Fournet in the Journal of the ACM (2017).¹

Machine learning systems and privacy

After joining Google in November 2014, Abadi was first author of the March 2016 TensorFlow white paper and of the May 2016 paper TensorFlow: A system for large-scale machine learning. Abadi later co-authored work on dynamic control flow in TensorFlow and co-wrote an introduction to a computational model of TensorFlow that describes an operational semantics for its programs.¹³⁴⁷

On privacy in machine learning, Abadi was first author of Deep Learning with Differential Privacy (2016) and co-authored the PATE approach, Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data, at ICLR 2017.⁵⁸¹

Work and contributions

• TensorFlow Open-source code

  First author of the TensorFlow white paper and of the 2016 paper TensorFlow: A system for large-scale machine learning; the white paper describes the interface and implementation built at Google and their open-source release under the Apache 2.0 license.³⁴¹

• Deep learning with differential privacy Published research

  First author of the 2016 paper that develops algorithmic techniques and a refined analysis of privacy costs for training deep neural networks within the framework of differential privacy.⁵

• A Logic of Authentication Published research

  Co-author, with Michael Burrows and Roger Needham, of the 1990 paper that introduced a logic for reasoning about authentication protocols, inducted into the ACM SIGOPS Hall of Fame in 2007.⁶

Timeline

1. Feb 1990Published A Logic of Authentication with Michael Burrows and Roger Needham in ACM Transactions on Computer Systems.⁶
2. 2007A Logic of Authentication was inducted into the ACM SIGOPS Hall of Fame.⁶
3. Nov 2014Joined Google.¹
4. Mar 14, 2016Posted the TensorFlow white paper on large-scale machine learning on heterogeneous distributed systems.³
5. May 27, 2016Posted TensorFlow, a system for large-scale machine learning, as first author.⁴
6. Jul 1, 2016Posted Deep Learning with Differential Privacy as first author.⁵
7. Oct 18, 2016Co-authored the PATE paper on knowledge transfer from private training data.⁸
8. May 4, 2018Co-authored Dynamic Control Flow in Large-Scale Machine Learning, on control flow in TensorFlow.⁷

Honors and recognition

• ACM SIGOPS Hall of Fame Award (A Logic of Authentication) · ACM Special Interest Group on Operating Systems, 2007⁶

Where they work

• Google 121 people in the wiki
• UC Santa Cruz 1 person in the wiki

Related people

• Jeff DeanAlso at Google
• Fabian PedregosaAlso at Google
• Adam PaszkeAlso at Google
• Alex SmolaAlso at Google
• Cho-Jui HsiehAlso at Google
• Luca AntigaML frameworks and libraries · Training systems
• Roy FrostigML frameworks and libraries · Machine-learning theory
• Pieter NoordhuisTraining systems · ML frameworks and libraries
• Edward YangML frameworks and libraries · Training systems
• Clément FarabetTraining systems · ML frameworks and libraries