Our PhD student Niklas Heim will present his research on neural arithmetic at Czech NeurIPS Meetup.
Neural Arithmetic - Learning to Extrapolate Beyond the Training Range
The paper Neural Power Units (pdf) by Niklas Heim was accepted to this year's NeurIPS. We are so proud of this success with Niklas being just a first-year Ph.D. student in our AI Center! You will get a chance to hear his presentation as part of the Czech Republic virtual edition of the NeurIPS Meetup on Friday, December 11 from 10 a.m. This will be followed by industry lightning talks (Promethist.ai, Resistant.ai, Seznam.cz and Rossum).
Abstract
Neural networks are great at approximating functions but often fail to generalize beyond the training range. Neural Arithmetic aims to overcome this issue by assuming that the underlying function is composed of simple arithmetic operations. During the talk, we will review the current state of the art of Neural Arithmetic Units with their advantages and drawbacks, possible applications, and finally our Neural Power Unit (NPU) which is published in this year's NeurIPS.
Registration
Please, register at the following link to attend the virtual event (free of charge): https://www.meetup.com/Prague-Machine-Learning/events/274968849/