Artificial intelligence systems based on photonic computation inspired by dendritic trees

Neuron dendrites are an essential part of the transmission of information in the brain. Their generally highly branched structure is designed to receive incoming signals from other neurons and to perform complex and efficient computation. Exploring the possibilities offered by dendritic computing when developing new concepts in artificial intelligence systems is the aim of the ADOPD project. ADOPD combines two apparently very distant fields, such as neuroscience and photonics, with the aim of developing new concepts and implementing advanced information processing devices.

ADOPD (Adaptive Optical Dendrites) is a project funded by the European Commission, coordinated by the University of Göttingen and with the academic participation of IFISC (UIB-CSIC, Spain) and Technische Universitaet Graz (Austria) and the company Leoni Fiber Optic GMBH (Germany). IFISC will also coordinate the participation of Yale University in the project. With a duration of 3 years and a budget of € 2,879,375, of which € 955,250 corresponds to IFISC, it is part of the Future and Emerging Technologies (FET) program, Open Domain, which aims to "support the initial stages of research and innovation in science and technology around new ideas for radically new future technologies". FET Open is a highly competitive program, with a success rate of 12% for this call, where only 50 of the 418 projects submitted have been funded. ADOPD obtained an overall score of 4.9 out of 5, with the third highest score, and the highest score in the scientific excellence section, being the most important in the final weighting.

The ADOPD project, whose initial meeting was held on 6 October, aims to contribute to the advancement of AI systems by developing ultra-fast computing units based on optical fiber technologies that imitate the information processing carried out by neurons in their dendritic trees. Its main objective is to achieve the development of computational systems with advanced capabilities compared to conventional artificial neural networks, while taking advantage of the benefits that neuro-inspired photonic implementations offer: high processing speed (hundreds of Gbits/s) and potentially low power consumption. IFISC participates via two project partners: IFISC-UIB, concentrating on theoretical and modeling aspects  of the photonic implementations (coordinated by the researchers Claudio Mirasso and Miguel C. Soriano) and IFISC-CSIC, focusing on experimental aspects and, in particular,  the implementation of neuro-inspired photonic devices capable of performing specific tasks ultra-quickly and efficiently (coordinated by the researchers Ingo Fischer and Apostolos Argyris). Moreover, the latter partner coordinates the participation of Professor Hui Cao (researcher at Yale University, USA).

More information on the project can be found on its website.

• Diario de Mallorca