David Arbour
Senior Research Scientist, Adobe Research
Causal inference, experimentation, AI evaluation
Google Scholar
·
arXiv
·
LinkedIn
Research
I am a Senior Research Scientist at Adobe Research working on rigorous evaluation
for complex systems, including generative AI and agentic systems. My work sits at
the intersection of causal inference and experimentation, and I use AI to design
more nuanced experimental designs, strengthen decision-making, and support
policy evaluation in real-world settings.
Causal inference for complex systems
I build methods for evaluating complex AI systems, especially where feedback
loops, dependence, and adaptive behavior are unavoidable. That includes work on
treatment assignment, adjustment, and anytime-valid inference under interference
and nonstationarity.
AI for experimentation
I use machine learning to design and analyze experiments that go beyond standard
A/B testing, including adaptive designs, surrogate outcomes, and decision-focused
objectives. The aim is to take advantage of generative AI without losing
statistical or causal validity.
Policy evaluation
I study how to evaluate policies when randomized experiments are limited or
infeasible, combining modern causal modeling with careful uncertainty
quantification. Recent work includes policy impact estimation in public safety
settings and other high-stakes domains.
