Reservoir computing with multiple optical nodes

Reservoir computing (RC) is a machine-learning (ML) paradigm. Like other ML schemes, it uses a network to process information. Unlike other ML schemes, it does not require this network to be reconfigured. This makes it easier to implement RC in physical hardware. Consequently, many different types of systems have been used to realize RC, such as electronic, mechanical, spintronic, and photonic systems.

One approach that is especially successful in photonics is to use a single physical node with delayed feedback, creating a network consisting of multiple virtual nodes along the delay line. Using a semiconductor laser with delayed feedback, information processing without performance degradation has been demonstrated at up to 20 GigaSamples per second. Still, one could finally arrive at a speed limit due to the serial nature of the delay concept. To overcome this potential barrier, it is useful to study concepts for photonic RC with multiple physical nodes.

In this talk, I will present our approach to photonic RC using more than one physical node, which uses vertical-cavity surface-emitting lasers (VCSELs) that are coupled via diffraction in an external cavity. Following that, I will present two more experimental realizations of photonic RC with more than one physical node and compare all three introduced concepts regarding their future perspectives.