IT-8008: Adaptive Neural Networks
PhD students can no longer get PhD credit for IT-3708, or the "parallel" version of it that we used to run. Now, they need to take this separate course that
uses some of the techniques learned in IT-3708, but on a significantly harder problem and/or to a deeper level of conceptual understanding.
The course delves deeply into artificial neural network (ANNs) that can alter their own performance, either via learning or evolutionary methods, or a mixture of the two. The main application platform for these ANNs is intelligent agents. Students will learn to implement adaptive ANNs and to use them as controllers for such agents.
This course will only be given when:
- there are PhD students with the appropriate combination of competence and interest, and
- the main teacher (Keith Downing) is currently at IDI (i.e., not on sabbatical) or on sabbatical but replaced by another professor willing to run it.
Materials and Meetings
The book "Evolutionary Robotics" (Nolfi and Floreano, 2000) must be purchased by each student. Other research articles may be distributed during the semester. There are no special course meetings, but
students are advised to attend the lectures for IT-3708 (Sub-symbolic AI Methods) if they are not familiar with the detailed mechanisms of evolutionary algorithms and artificial
Students will read the book "Evolutionary Robotics" and other related articles. They will then have the opportunity to implement their own adaptive-ANN control system for a simulated robot (using Webots or a similar package). Students will write a report of approximately 10 pages describing their system, which they will also demonstrate at the end of the course. Students can choose to forego the implementation in exchange for a longer report (of approximately 20 pages) giving extensive coverage of the state-of-the-art in adaptive-ANN-controlled intelligent agents. The 20-page report or 10-page report plus demo will be due at the end of the semester: late April or early May.
Webots is a software package for building and running controllers of both simulated and real robots. It is quite popular, being used in over 600 universities world-wide. IDI has
a floating license for Webots, which IT-3708 and IT-8008 students are free to use.
For more information on using Webots with artificial neural networks, see this , and for information concerning the
use of Webots here at IDI, see this