Claudia Perlich

Claudia Perlich is a data scientist whose career spans academia, digital advertising and quantitative finance, built on a reputation for rigorous, competition-tested machine learning. Born near Leipzig in what was then East Germany, she studied computer science at the Technical University of Darmstadt before earning master's degrees in the United States and Germany and a PhD in Information Systems from New York University's Stern School of Business.

Perlich rose to prominence in the applied machine learning community through her team's consecutive victories in the KDD Cup, the field's leading data mining competition, from 2007 to 2009 — an unusual run of dominance that established her as one of the sharpest practical modellers in the discipline. Her subsequent research on 'leakage', the ways in which machine learning models can inadvertently exploit information that would not be available at prediction time, won best paper honours at KDD and became required reading for practitioners seeking to avoid inflated, unreliable model performance.

She spent years as a research staff member at IBM's T.J. Watson Research Center before becoming Chief Scientist at Dstillery, an advertising technology company, where she applied large-scale machine learning to real-time bidding and audience targeting — work that produced a second KDD best paper award, this time on bid optimisation in online advertising. In 2014 she became the first woman to serve as general chair of the KDD Conference, a milestone for a field long dominated by men in its leadership roles.

Perlich later moved into quantitative finance as a senior data scientist at Two Sigma, applying predictive modelling techniques honed in advertising and web-scale data to investment research, while continuing to teach as an adjunct professor at NYU Stern. Her career reflects the portability of rigorous machine learning practice across industries, from research labs to ad-tech to systematic investing.

Widely respected within the applied machine learning community for her technical candour — she is known for arguing that real-world data science is inherently messy and judgement-driven rather than a matter of neatly applying textbook methods — Perlich remains an influential voice on the practical, non-idealised realities of building models that work in production.