Abstract of the paper:
How can we endow artificial neural networks with the capacity for life-long (i.e., continual) learning? Recently, algorithmic insights into the olfactory (neural) circuit of the Drosophila Melanogaster have been discovered that endows it with the ability to acquire novel information, while preserving old memories. In this work, we leverage these algorithmic insights, notably sparse coding via a random projection, a
$k$ -winner-take-all computation, (partial) synaptic freezing, and associative learning, to build an artificial system with the ability of continual learning.