Predict the predictable, measure the rest: active learning for efficient biological experiments by Dr. Jason Hartford
Dates: | 26 February 2025 |
Times: | 13:00 - 14:00 |
What is it: | Seminar |
Organiser: | Department of Computer Science |
How much: | Free |
Who is it for: | University staff, Current University students |
Speaker: | Dr. Jason Hartford |
|
Dr. Hartford will discuss how active learning can be used to build hybrid experimental systems that combine predictive modelling with targeted experimentation in large scale experimental settings with high dimensional action and outcome spaces. By learning when to trust model predictions and when to acquire new measurements, we reduce experimental cost while ensuring we collect data where it is most informative. He will illustrate this idea through two applications: efficiently screening compounds in drug discovery and detecting pairwise interactions between biological perturbations. In each setting, active learning guides us to sample the most uncertain or challenging experimental designs. This approach not only limits the overall number of experiments but also uncovers deeper insights into the underlying phenomena. The talk will focus on two recent papers, https://arxiv.org/abs/2410.19631 and https://arxiv.org/abs/2409.07594."
Speaker
Dr. Jason Hartford
Role: Dame Kathleen Ollerenshaw Fellow in Computer Science
Organisation: University of Manchester
Biography: Dr. Jason Hartford is a Dame Kathleen Ollerenshaw Fellow in Computer Science at the University of Manchester and holds a joint appointment as a Research Unit Lead and Staff Research Scientist at Valence Labs. He was a post-doctoral fellow with Yoshua Bengio at Université de Montréal / Mila, and before joining Mila, he completed his PhD at the University of British Columbia with Kevin Leyton-Brown. He works on developing techniques for causal inference from high dimensional / unstructured data and active learning.
Travel and Contact Information
Find event
Kilburn Building
Manchester