Semester Projects

 

List of Projects – Fall 2018

 

Exploration of deep networks

Based on the discussion in the class on Artificial Neural Networks, the following topics can be proposed for further exploration

(i) a variant of batch update

(ii) visualization of saddle points and online descent algorithms

This project can be for one or maximum two students. Profile of candidate: EPFL master student in Computer Science (or related subject) who has followed the class ‘Artificial Neural Networks’ in spring 2018.

 
Please send your CV and an up-to-date grade sheet of results to Bernd Illing, Johanni Brea or Florian Colombo.
 

List of Projects – Spring 2018:

Characterization of simulated plasticity-induction protocols

In the experimental literature, a large amount of different plasticity-induction protocols can be found. Building a single modeling framework that incorporates all different experimental techniques is challenging and often requires arbitrary modeling choices. The aim of this semester project is to identify the plasticity effects among protocols that are model-independent. In order to achieve this, the student will be required first to become familiar with the experimental literature and secondly, to use existing implementations of neural and plasticity models (or, possibly, to implement new ones). Finally, the student will analyze the simulations results in order to extract invariant features.

The preferred programming language is Python. The project is open to all students.

Interested students should send grades and CV to Chiara Gastaldi or Samuel Muscinelli.

Adaptive learning through surprise minimisation over extended time

Whenever there is a mismatch between our expectation and our actual experience we get surprised, and often need to update our belief about the world. In a framework recently developed in LCN, we mathematically defined a novel measure of surprise and we proposed a surprise-minimisation learning algorithm. The goal of this semester project is to develop an extension of this framework that will allow: 1) the contribution of the history of data points in learning, and 2) the adjustment of the subjective propensity to be surprised in a principled way.

The project requires a substantial background in mathematics and good programming skills in Python or Matlab.

Interested students should send grades and CV to Dane Corneil and Vasiliki Liakoni