Hello all,
I will speak at group meeting tomorrow. My title and abstract can be found below. We will
meet at the usual time and place, SS571 at 2:30pm. If you would like to join the meeting
via Skype, please let me know.
Best,
Riley
---------------------------------------------------------------------------------------------
Title: Gemini: Dynamic Bias Correction for Autonomous Experimentation
Abstract: I will present Gemini, a machine learning algorithm capable of predicting
expensive to evaluate experimental response surfaces using heterogeneous data. Here,
evaluation expense is measured with respect to any budgeted resources (ie. time consuming
calculations, synthesis of complicated intermediates, etc). Gemini is shown to be useful
in making accurate predictions of expensive response surfaces when supplemented by
evaluations of a cheaper to evaluate surface that shares a high degree of correlation with
its counterpart. I will detail analytic tests performed on an ensemble of 1D response
curves, before presenting preliminary results from two applications of Gemini in
chemistry: the design of photo-stable polymer blends for organic photovoltaics, and the
prediction of hydration free energies for small organic molecules. Lastly, I suggest how
Gemini might be used in tandem with the deep Bayesian optimizer Phoenics in a global
optimization setting.
Show replies by date