At the seminar, we will discuss feature selection task - selecting key parameters of model state in demonstration learning.
We will discuss a paper that compares different approaches to this task: numeric, based on statistic analysis, human-driven, based on additional train dataset.
Authors study the dependence between algorithm selection and accuracy in classification task and demonstrate supremacy of human-driven approach in the case of limited demonstrations.
Speaker: Nikolai Izyumov.
Presentation language: Russian.
Date and Time: October 29th, 18:30-20:00.
Place: Times, room 204.