Politecnico di Torino (logo)

Neural networks in optical domain

Enrica Racca

Neural networks in optical domain.

Rel. Emiliano Descrovi, Alfredo Braunstein, Luca Dall'Asta. Politecnico di Torino, Corso di laurea magistrale in Nanotechnologies For Icts (Nanotecnologie Per Le Ict), 2019

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (5MB) | Preview

Objects classification is one of the applications which most efficaciously have been improved by deep learning, in the same way as other innumerable functions deeply pervading the modern society. In the present work we implement a deep learning framework with diffractive layers that collectively perform digit recognition. Neural networks constitute the computing system performing deep learning, which differs from any application of artificial intelligence – enabling machines to automatically learn from experience without being explicitly programmed – in that it creates the representations essential for classification organizing them into multiple levels. Our neural network is inspired by a framework recently introduced in the literature, termed as Diffractive Deep Neural Network (D2NN). It is physically formed by multiple layers of diffractive surfaces that collaboratively perform optical diffraction when an input image is exposed to electromagnetic radiation. During the training phase implemented on a computer through deep learning’s methods, trainable parameters represented by the diffractive layers’ transmission coefficients are properly adjusted. Once completed the numerical stage, the design of the framework is established: the obtained transmission coefficients provide in fact information about the thickness of the diffractive layers that, if 3D-printed and settled in their well-suited laboratory setup, would give rise to a powerful device performing digit classification at the speed of light. Optical machine learning’s earlier works concerned realizations of programmable devices performing typical machine learning applications and equipped with optical components, which satisfy sought-after requirements like speed and power efficiency. In this line of work, our framework represents a pioneering innovation since, once physically fabricated, it may execute the specific task for which it is trained, exploiting no power but the effectiveness of optical diffraction through passive optical layers.

Relators: Emiliano Descrovi, Alfredo Braunstein, Luca Dall'Asta
Academic year: 2019/20
Publication type: Electronic
Number of Pages: 87
Corso di laurea: Corso di laurea magistrale in Nanotechnologies For Icts (Nanotecnologie Per Le Ict)
Classe di laurea: New organization > Master science > LM-29 - ELECTRONIC ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/12608
Modify record (reserved for operators) Modify record (reserved for operators)