Teresa Yeo

My research has previously focused on closed-loop methods for efficient adaptation, using feedback, whether from the model's own performance or from its environment as signal for improvement. This has spanned targeted training-data generation and gradient-free test-time adaptation, and, more recently, self-improving models.

My PhD was at EPFL supervised by Amir Zamir, on making models more reliable under changing environments. I was also a postdoc at the Singapore-MIT research centre working on neurosymbolic methods for adaptation. In my past life, I was a quant in New York and London and worked on creating systematic investment strategies (or, a glorified coin flipper).