Adaptive Experimentation: Accelerating & Automating Practical Design & Scientific Discovery. Areas include Chemistry & Materials Science, Business, Social Good (Education, Mental & Physical Health), Social-Behavioural Sciences, Economics, Human-Computer Interaction
Postdoctoral positions are available for investigating tools and methods for adaptive experimentation. Faculty & Grad Students who might be interested in adaptive experimentation – applying it in their own work, or collaborating on open questions – can also express interest, even if they are not seeking postdocs.
What are Adaptive Experiments and why could they transform life for 8 billion people? Adaptive experiments accelerate and automate discovery in science, by more rapidly and efficiently analyzing data during an experiment, in order to adapt data collection. One goal is to more quickly benefit participants in the experiment (e.g. users of a website or app). A second goal is to enable rigorous yet expedited statistical inference. Everyday billions of people are taking actions to achieve goals, and adaptive experimentation techniques could be used to help them change their behaviour, to more effectively do so.
You will be working on Adaptive Experimentation with Joseph Jay Williams at the University of Toronto, to develop Intelligent Adaptive Interventions for changing people's behaviour through technology. He is based in Computer Science (Human Computer Interaction, Applied AI, Applied ML), Psychology (Educational, Health), Statistics, as well as Economics & Industrial Engineering. You can read more about his research program at tiny.cc/williamsresearch. The adaptive experimentation work is specifically at the url tiny.cc/iadaptiveexperiments.
You will have a chance to collaborate with researchers such as Jeff Watchorn, Alan Aspuru Guzik, John Stamper, Anastasia Kuzminykh, Meredith Franklin.
Applications will span areas in Behaviour Change through Technology (e.g. Education, Mental & Physical Health, Marketing). Applications will also include Materials Science & Chemistry, with UToronto's $200 M award for the Acceleration Consortium. This concerns everything from engineering and adoption of consumer products (from soaps to creams to perfumes) to advancing materials science and chemistry – the 'central science'.
The postdoc can come from multiple disciplines, although preference is given for those who have statistical knowledge of randomized experiments.
Statistics:
Evaluating existing and developing new algorithms for adaptive experimentation. Prior experience in designing algorithms is not as important as rigorous understanding of applied statistics, and being able to evaluate Frequentist Hypothesis Tests (& Bayesian analyses)
Evaluating existing and developing new analysis techniques (hypothesis tests, Bayesian analyses) for analyzing data from adaptively randomized experiments
HCI Human-Computer Interaction & Social-Behavioural Sciences (Psychology):
Design of tools and explanations for how adaptive experimentation algorithms and analysis techniques work.
A user-centered design approach to getting practitioners and scientists to understand, adopt, and modify these techniques.
Machine Learning/Artificial Intelligence:
You will have an opportunity to apply algorithms to cutting edge scientific and engineering problems in the social and natural sciences.
Priority will be given to candidates who have experience in the following or clearly demonstrate interest and capacity in learning:
Multi-armed bandit algorithms, reinforcement learning, Bayesian Optimization
Statistics, Biostatistics, Applications of Statistics of the kind typically done in statistics departments
Experience working in applied settings with stakeholders who have requirements for the kinds of analyses they will use.
Experience working with domain scientists, eliciting their needs, and communicating which algorithms are relevant for which problems. Experience understanding how to work with domain scientists and other stakeholders to parameterize algorithms appropriately, choose parameters, and explain how these algorithms operate.
There are several potential funding sources for the postdoc:
Carnegie Mellon NSF grant (position will be between CMU & UToronto: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2209819&HistoricalAwards=false).
Bringing your own funding source: From Canada, US, Europe, or other sources.
Other options - we may make available access to other sources of funding, once attempts are first made to secure the above.
Actions you can take:
1. Request access here to a Google Document with more information about the research program, and current projects and papers: tiny.cc/learnadaptiveexpt Learn about Adaptive Experimentation as a Research Area
2. To learn about the postdoctoral positions available: Provide your email at https://forms.gle/k4DcipX2zFsiWAPcA
3. Faculty, Grad Students who don't need a postdoc: To learn more about using our adaptive experimentation techniques, collaborating, or about this area in general: Provide your email at https://forms.gle/SRPjNgohxynTqBP39