Neural Networks in Neuroscience
When: From Monday 30th November to Monday 14th December 2020
Where: Online Talk offered through Zoom Platform
- to undergraduate students of the programme in Science and Methods of Cognitive Psychology (STPC)
- to students of the Master’s degree in Psychology Neuroscience
- to students of the Master’s degree in Data Science
- to students of the Doctoral course in Cognitive Science
- to DIPSCO lecturers
The Department of Psychology and Cognitive Science is honored to present the online seminar series on “Neural Networks in Neuroscience”.
Topic of the seminars
The seminar series is an introduction to the basis of deep neural networks and their applications to neuroimaging data. The seminars will consist of combined lectures (45 minutes) and exercises. During seminars attendees will be assigned specific data analysis tasks to be solved using Python. The final seminar will consist of an open discussion of previous assignments. It is advantageous to have some programming skills, not necessarily in Python.
The online seminars will be held through Zoom platform. All seminars will be recorded and made available upon request.
- Prof. Christoph Braun PhD, University of Tübingen
- Antonino Greco MSc, University of Trento
30 November, 13:00-15:00
- Python for Data Science
General introduction to the seminars and introduction to fundamental concepts of Python programming for Data Science with a focus on the object-oriented framework. Attendees will be required to preinstall the Anaconda package on their computers in order to do some preliminary exercises.
01 December, 13:00-15:00
- Introduction to Machine Learning
Introduction to the general field of Machine Learning with a focus on supervised learning. Discussing fundamental concepts like regression and classification, training and testing, overfitting and k-fold cross validation. Hands-on with some simple machine learning algorithm applied to neuroimaging data classification using scikit-learn library.
02 December, 13:00-15:00
- Introduction to Neuronal Networks
Introducing the history of neural networks with a focus on McCulloch–Pitts neuronal model neuron and the Rosenblatt's Perceptron. Hands-on implementing these models in an object-oriented framework.
03 December, 13:00-15:00
- Activation and Loss Function
Introducing the concepts of activation and loss functions and how to select them based on the problem to solve. Hands-on implementing new activation and loss functions as methods for the previously created Perceptron class.
04 December, 13:00-15:00
- Backpropagation Algorithm
Introducing the core part of the learning algorithm modern neural networks use to accomplish tasks. A note on differential calculus, partial derivatives and how these concepts are related to the network's weights optimization. Hands-on implementing a simple neural network optimized by the backpropagation algorithm.
07 December, 13:00-15:00
- From Shallow to Deep Neuronal Networks
Discussing traditional models like linear and logistic regression under the framework of Deep Learning. Discussing the reasons for the transitions from shallow to deep neural networks and how to deal with some pitfalls. Hands-on implementing a fully-connected multi-layer deep neural network.
09 December, 13:00-15:00
- Advanced Topics and Applications
Discussing advanced techniques to reduce the problem of overfitting and the vanishing and exploding gradient problem. Introducing the Convolutional Neural Network architecture and some applications to neuroimaging data. Hands-on implementing deep neural networks using the Keras and Tensorflow libraries.
14 December, 13:00-15:00
- Report Presentation and Discussion
Presentation of the reports made by groups of attendees about some data analysis problems proposed during the seminars.
Prof. Andrea Caria
Dipartimento di Psicologia e Scienze Cognitive
andrea.caria [at] unitn.it
A cura dello Staff per la Comunicazione - Polo Rovereto