Reservoir Computing is an appealing methodology to design deep neural networks for temporal data processing that is strikingly efficient in terms of required computational resources. The core working principle resides in imposing dynamical stability properties to the developed neural representations, and restricting the training algorithms to operate on a small set of connections. After an introduction to the topic, this seminar will delve into recent advances that enable state-of-the-art performance in problems on temporal and graph-structured data, at a fraction of the cost required by fully trainable models.
Reservoir Computing is an appealing methodology to design deep neural networks for temporal data processing that is strikingly efficient in terms of required computational resources. The core working principle resides in imposing dynamical stability properties to the developed neural representations, and restricting the training algorithms to operate on a small set of connections. After an introduction to the topic, this seminar will delve into recent advances that enable state-of-the-art performance in problems on temporal and graph-structured data, at a fraction of the cost required by fully trainable models.
