Increasingly 'human' algorithms…


Researchers at ULiège have developed a highly innovative method that allows algorithms to adapt to new situations in the same way as a human brain.

This discovery is a reward for more than two years of research based on a biological mechanism called neuromodulation, which is the modification of the neurones themselves. The researchers managed to develop an artificial neural architecture that has two sub-networks working together. The first sub-network takes account of the context, while the second determines the actions the agent must take depending on the "neuromodulation" of the first.

This new algorithm can create agents capable of carrying out tasks that have not yet been encountered during training. "What we have managed to do is make our algorithms adapt much faster, which means that if we teach an algorithm to play the piano, it will then be capable of using some of the concepts it learnt while playing the piano to play the guitar," said Nicolas Vecoven, a doctoral student in engineering at the University of Liège.

Taking inspiration from the human brain to meet the needs of artificial intelligence is a promising area for the future, but one in which a great deal remains to be done. Despite the obvious progress made during the last few years, the potential of artificial intelligence still remains well below what the human brain can achieve.