VIDEOS

Single-mode fiber optical dendritic unit for input correlation learning

EC H2020 ADOPD project 899265 Dissemination actions. We use a single-mode fiber (SMF) optical dendritic unit (ODU) to apply input correlation learning on a system with multiple sensory signals and a reference signal. Each SMF plays the role of a dendritic branch, where plasticity rules apply. At the end of the ODU, a weighted sum of the multiple signals is obtained. Copyright 2023 IFISC (CSIC-UIB) – Spain, ALL RIGHTS RESERVED.

Multi-mode fiber optical dendritic unit for multi-bit header recognition

EC H2020 ADOPD project 899265 Dissemination actions. We use a multi-mode fiber (MMF) optical dendritic unit (ODU) to perform multi-bit header classification at ultrafast speed. The different transient modes that propagate along the MMF have different group velocities and play the role of the dendritic branches. By measuring a single temporal sample, and multiple spatial points of the beam profile of the output pattern of the MMF, we can obtain the classification outcome every 36 picoseconds. Copyright 2023 IFISC (CSIC-UIB) – Spain, ALL RIGHTS RESERVED.

Single-mode fiber optical dendritic unit demonstrator

EC H2020 ADOPD project 899265 Dissemination actions. 1st experimental demonstrator of the ADOPD project. Implementation of an ICO learning rule with a single mode fiber (SMF) optical dendritic unit (ODU) for a temperature stabilization task. The objective is to stabilize a periodically perturbed system parameter, like the temperature. Control is achieved using the ICO learning rule (Hebbian learning), which dynamically updates the plastic weights of the SMF-ODU to stabilize the parameter of interest. Copyright 2024 IFISC (CSIC-UIB) – Spain, ALL RIGHTS RESERVED.

Multi-mode fiber optical dendritic unit demonstrator

EC H2020 ADOPD project 899265 Dissemination actions. 2nd experimental demonstrator of the ADOPD project. Implementation of a multi-mode fiber (MMF) optical dendritic unit (ODU) to perform up to 7-bit digital header classification at 28.5 Gb/s. The system runs in an autonomous mode, with an optimization algorithm that finds an appropriate operational point of the system, defining the input optical beam coupling conditions. The same algorithm is used to preserve a stable and acceptable performance of the system, despite the changes in the input/output signal transformation that occurs along the fiber. Copyright 2024 IFISC (CSIC-UIB) – Spain, ALL RIGHTS RESERVED.

Optical spiking neural network gesture recognition

Context dependent hand-gesture recognition by 200 leaky integrate and fire neurons based on an optical memory cell memory structure.

For a video of the ADSYNX method see the ADSYNX page.