Seminar of Theoretical Computer Science - Věra Kurková (29.3.2019)

Friday 29.3.2019 at 11:00, room M/213

---

Věra Kurková (Czech Academy of Sciences):
Limitations of Shallow Networks

Abstract:
Although originally biologically inspired neural networks were introduced as multilayer computational models, later shallow architectures (with one hidden layer of computational units) became dominant in applications.
Recently, interest in architectures with several layers was renewed due to successes of deep convolutional networks.
Experimental results motivated theoretical research aiming to give some answers to the question: When and why are deep networks better than shallow ones? This lecture will present some recent theoretical results on complexity of shallow networks performing high-dimensional tasks. In particular, it will focus on probabilistic and constructive lower bounds on approximate measures of network sparsity. Connections with the No Free Lunch Theorem, the central paradox of coding theory, and pseudo-noise sequences will be discussed.

web: https://beda.dcs.fmph.uniba.sk/seminar

---