CIM Seminar

Abstract: This presentation gives new insights for an obvious learning algorithm: let's take the good old logistic regression, but run it online on a single non-stochastic sequence. Which results can still be guaranteed formally? This problem is motivated by many industrial applications where guaranteed worst-case behaviour is much more relevant than average-case performance.

This seminar only takes place online via Zoom. The Zoom link will be provided here the morning of the day of the event. Registrate your attendance to the event here.