Context:

Currently, machine learning plays a central role in many applications, and the adoption of increasingly decentralized solutions requires learning tasks to be carried out in a decentralized fashion. Such decentralization implies that nodes in the system can assume multiple roles: providing local data to feed the learning process, performing the learning, and aggregating the knowledge of other nodes.

In this context, given the decentralized nature of the problem, it is of paramount importance to have tools to simulate, evaluate, improve, and validate decentralized learning solutions.

Objective:

The aim of this internship is to extend an existing simulation tool, gossipy (https://github.com/makgyver/gossipy), by designing, implementing, and evaluating at least one distributed learning solution within that tool.

The successful candidate will join the Laboratory for Trustworthy, Smart, and Self-Organizing Information Systems (LICIA) at CEA LIST in a multicultural environment with the possibility to collaborate with other researchers in the field.

Methodology:

The intern will have the following responsibilities:

1.      Prepare a state-of-the-art review of decentralized learning.

2.      Become familiar with gossipy.

3.      Choose or design a decentralized learning solution

4.      Implement, in the simulation environment, the chosen decentralized learning solution.

5.      Assess the solution's performance.

Competences:

·        Being Master 2 in Computer Science/Engineering.

·        Knowledge about distributed systems in general.

·        Good experience in programming in any language.