[vc_row][vc_column][vc_tta_accordion shape=»square» c_icon=»chevron» c_position=»right» active_section=»» no_fill=»true» collapsible_all=»true»][vc_tta_section title=»Resumen» tab_id=»resumen»][vc_column_text]
New technologies related to optical communications, sensors, the Internet of Things, and artificial intelligence are generating opportunities with the potential to improve our quality of life and provide new services for society and the economy. However, to manage and process the growing amount of data in optical communications and sensors, it is necessary to improve our ability to handle this data at high speed, with suitable hardware and high energy efficiency. The objective of this project is to develop a novel method for optically decoding data for optical communication networks, capable of meeting the aforementioned requirements, based on neuro-inspired information processing systems. Currently, there are techniques that offer excellent results for data processing, especially in the area of machine learning, but their energy demands and lack of speed hinder their use in many current and future applications. Furthermore, these techniques are not applicable to fiber optic communication networks, particularly in the optical domain. In this project, we follow a multidisciplinary approach, encompassing aspects of engineering, physics, neuroscience, and computer science, based on our experience in the design and development of information processing systems in photonic hardware. Unlike traditional machine learning, which typically consists of a network of many interconnected nodes, we use a simple dynamic system. This system comprises a nonlinear node with a delay and is capable of exploiting the dynamic richness of systems with delays to process information. Our goal is to use this multidisciplinary approach to improve our system by introducing novel preprocessing techniques, multiple levels, and new learning techniques tailored to the data requirements. In addition to implementation in the optical domain, to mitigate risk, our technique could be applied in the electrical domain after signal detection. The guiding principle of this project will be the development and implementation of data decoding techniques that combine conceptual simplicity, high-speed hardware, flexibility, energy efficiency, and high performance. To achieve this objective, three complementary and highly competitive groups have been assembled to carry out the following tasks: – Systematic exploration of data preprocessing, essential for making data more manageable and improving subsequent machine learning, without losing the essential characteristics necessary for high-performance decoding. – Expansion and adaptation of neuro-inspired data processing methods to data decoding in fiber optic communication systems, respecting the constraints imposed by the hardware. – Study of fast and efficient photonic devices with intensive numerical simulations to aid in system design and optimization. This project is a step towards the development of an ultra-fast and energy-efficient decoding technique, complementary to standard methods, using minimal resources and with high performance, serving a digital society in which technology is employed to improve data processing and provide new services.
[/vc_column_text][/vc_tta_section][vc_tta_section title=»Abstract» tab_id=»abstract»][vc_column_text]
Novel technologies related to optical communications, sensing, the Internet of Things (IoT) and artificial intelligence have been generating unique opportunities and potential to enhance our quality-of-life, and to provide new services for our society and economy. However, the perspective to manage and process the dramatically increasing amount of data relies on our ability to handle this data with high-speed, suitable hardware and much improved energy efficiency. In this project, it is our aim to develop novel all-optical decoding schemes for optical communication networks, that are based on neuroinspired concepts and are able to fulfill the previous requirements. Excellently performing neuro-inspired concepts and algorithms, in particular related to machine learning, have been developed, but their energy requirements and lack of speed hinder their implementation in a significant number of current and future applications. In particular, this approach faces severe challenges, when trying to apply it in all-optical communication networks. Therefore, in this proposal we follow a different approach, building upon our experience of designing and realizing neuro-inspired information processing systems, mainly in photonic hardware. In contrast to traditional machine learning, we replace the usual structure of a network composed of multiple connected nodes by a simple dynamical system. The latter comprises a nonlinear node subject to delayed feedback, exploiting the dynamical richness of the delay systems for computational purposes. We aim at extending these concepts by introducing novel pre-processing techniques, taking advantage of multilevel systems and applying novel learning concepts adapted to the particular data and processing requirements. To mitigate the risk, our approach could also be applied in the electronic domain after the signal detection. The guiding principle will be the realization and implementation of data decoding techniques that combine conceptual and hardware simplicity, high-speed, flexibility, energy efficiency and high performance. To achieve this goal we set up a multidisciplinary collaborative project between complementary and highly competitive groups, planning to carry out the following tasks: – To perform a systematic exploration of data-reduction and preprocessing. This task is essential in order to make the amounts of data tractable, shaping them to benefit most from the subsequent machine learning steps, all without losing the essential features required for a high-performance data decoding. – To extend and to adapt neuro-inspired processing methods for improving data decoding in the all-optical domain of optical communication networks, keeping hardware implementability, efficiency and performance high. – To explore fast, energy-efficient, photonic hardware via intensive numerical simulations helping in system design and optimization and via their experimental implementation. Altogether, this project represents an important step towards ultra-fast, energy-efficient data decoding techniques, complementary to standard approaches. It promises the identification of minimum requirements and the implementation of the concept with high performance. Ultimately, it serves a digital society, in which technology is harnessed to improve data handling and processing and to provide new service
[/vc_column_text][/vc_tta_section][/vc_tta_accordion][/vc_column][/vc_row]