Deep Neural Networks: Optimization, application and explaining decisions

Macedonian Data Scientists, let’s have our second Data Science meetup on the 18th of June at 18:00.

During this meeting we'll have a great talk given by Sasho Nedelkoski Research Associate at Technische Universität Berlin. The talk will focus on Deep Neural Networks their optimization, application and explaining decisions.

Short summary of the talk:
In order to efficiently train deep neural network we use tricks like: proper initialization, normalization and non-linearities. Nevertheless, these tricks have strong theoretical background and intuition which must be part of our knowledge when designing complex solutions and deep network architectures.
Deep networks increasingly are used in various radiological imaging tasks. Due to the complexity of the problem and the computational requirements, a good solution for these kind of tasks requires both strong theoretical background and heavy engineering. In collaboration with the Charité, which is the largest university clinic in Europe, we put our effort into detection of brain aneurysms with novel deep learning methods.
Finally, although these methods perform impressively well, they have a significant disadvantage, the limited interpretability of the solution and thus the scope of application in practice. Therefore, explaining decisions in deep neural networks receives increasing interest in the ML community which opens new perspectives and application of these techniques.

Free snacks and drinks! Provided by Elevate Global!