Enrico Picco (ULB)

The research of Enrico Picco (ESR4) at the Université Libre de Bruxelles focussed on Optoelectronic Reservoir Computing (RC). RC is a powerful, highly versatile set of methods for designing and training artificial neural networks, which has been applied in unconventional computing in attempts to overcome the Von Neumann bottleneck.

His project started by building an experimental setup, comprising optical hardware interfaced by an FPGA board: in this way I realized an optoelectronic reservoir computer based on time multiplexing. The setup has been tested on the complex task of Human Action Recognition, i.e. the classification of human actions in videos, showing remarkable results compared to machine learning approaches implemented in silicon digital platforms. Furthermore, the system was capable of inferring real-time classification, thus making a step forward in RC for real-world applications [1-4].

The same system has been modified to implement different RC architectures. First it has been used to implement a Deep Reservoir Computer, demonstrating that a deep architecture is better than a shallow one [7]. The deep architecture has also been studied in a collaboration with another ESR, Alessandro Lupo, using a frequency multiplexed system [5,6]. Second the use of delayed inputs as a way to improve the performance of time delay reservoir computers has been studied in [8]. Finally, being on a 4-year PhD track, Enrico Picco is currently working  with scientists from other universities to investigate new potentially strategic topics such as asynchronized delay-based RC, or edge computing by inferring classification without the need of complex pre-processing stages.

By being part of the POST-DIGITAL project Enrico had the opportunity to meet and collaborate with many scientists across Europe. Many workshops and training events ensured high quality complementary activities to Enrico’s journey as an Early Stage Researcher, broadening my scientific and cooperative skills. The 3-months secondment at IBM-Zurich proved to be a very valuable experience: it made me interface with a different reality than academic world, and it strengthened some of my expertise by making me interface with scientific experts in a top-level industry sector.

Key publications by Enrico Picco:

[1] E. Picco, P. Antonik, S. Massar, “High speed human action recognition using a photonic reservoir computer”, Neural Networks, Volume 165, 2023, Pages 662-675, ISSN 0893-6080, https://doi.org/10.1016/j.neunet.2023.06.014

[2] E. Picco, P. Antonik, S. Massar. “Time-Multiplexed Photonic Reservoir Computer for Recognition of Human Actions in Videos.” (CLEO/Europe-EQEC), Munich, Germany (2023)

[3] E. Picco, P. Antonik, S. Massar , “Time-multiplexed photonic reservoir computer for recognition of filmed human actions”, OPTIQUE NICE 2022, NICE (FR)

[4] E. Picco, S. Massar. “Real-Time Photonic Deep Reservoir Computing for Speech Recognition.” The 2023 International Joint Conference on Neural Networks (IJCNN). IEEE (2023). DOI: 10.1109/IJCNN54540.2023.10191786

[5] Lupo, A., Picco, E., Zajnulina, M., & Massar, S. (2023). Deep photonic reservoir computer based on frequency multiplexing with fully analog connection between layers. Optica10(11), 1478-1485.[6] A. Lupo, E. Picco, M. Zajnulina, S. Massar. “Time and frequency multiplexed implementation of a deep Reservoir Computer” IEEE Benelux Photonic Chapter 2022, pp. 168-171 (2022)

[7] Picco, E., Lupo, A., & Massar, S. (2023). Deep Photonic Reservoir Computer for Speech Recognition. arXiv preprint arXiv:2312.06558, in press in IEEE Transactions on Neural Networks and Learning Systems

[8] Picco, E., Jaurigue, L., Lüdge, K., & Massar, S. (2024). Efficient Optimisation of Physical Reservoir Computers using only a Delayed Input. arXiv preprint arXiv:2401.14371.