Robust Intelligence

Robust Intelligence is a Cambridge, Massachussets-based company which is developing software infrastructure for building robust machine learning models and systems optimized to work under conditions of high statistical noise. Founded in 2019, the company aims to commercialize technologies and techniques developed by its co-founder, Yaron Singer, a professor in Harvard's computer science department.

The company is venture-backed and has received investment from Engineering Capital and Sequoia Capital. Engineering Capital led the company's seed round in April 2019.

Robust Intelligence was founded to confront real-world problems of data-driven optimization of functions subject to noise. Singer's work, as well as the work of his cofounders, involves applying adversarial methods to create machine learning models which are robust and resiliant in the face of noise and other perturbations in input data which may otherwise cause a model to fail.

Singer writes:

Our work on noise robust optimization continued to developing machine learning models that are robust to adversarial examples. There has recently been a great deal of work showing that small, adversarial, pertubations in data can lead to failure of state-of-the-art machine learning models. Machine learning models are trained by optimizing a loss function defined by the data, and the reason the models fail when the data is pertubed is because the optimization algorithms are not robust to noise.

Singer's academic work has shown that boosting can be used to design robust classifiers and can be used to design adversarial attacks on machine learning models.

Real-world use cases for robust machine learning models include classification of financial risk, among other application areas.