PIMS-SFU Applied & Computational Math Seminar: Laura Thesing

This talk will focus on two topics in the foundations of neural networks: computability and false structures. In the first part I will recall briefly the concept of recursive functions and then discuss the gap between existence results, the minimization problem and the construction of networks. In the second part I will talk about the problem of adversarial examples in classfication from the point of view of false structures. With different toy examples we will have a look at underlying structures that are easier to be captured by neural network than the one that humans would normally use for the decision making process. Both topics are ongoing research.

This talk will be informal. Therefore, questions and open discussions are very welcome.