Deep Learnig and applications (block C1-opt)

Deep Learning
2 YEAR II semester  6 CFU

Eugenio Martinelli
A.Y. 2024-25 new
Didatticaweb
Code:
SSD: ING-INF/01

PREREQUISITES

Basic knowledge of probability theory, signal theory, and pattern recognition.

FORMATIVE OBJECTIVES

LEARNING OUTCOMES:
Learning the basic concepts of deep learning algorithms. The main Machine Learning algorithms will be covered, followed by a focus on those related to deep learning, with particular emphasis on their application in the field of mechatronics.

KNOWLEDGE AND UNDERSTANDING:
The student acquires knowledge related to the field of Machine Learning, with particular reference to the ability to extract quantitative and qualitative information from images and videos and multivariate data and their subsequent processing for regression and classification tasks.

APPLYING KNOWLEDGE AND UNDERSTANDING:
The student acquires the capability to implement the algorithms in Matlab through dedicated lessons during the course to the aim of being able to autonomously develop new codes for the solution of specific problems in different application fields.

MAKING JUDGEMENTS:
The student must be able to integrate the basic knowledge provided with those deriving from the other courses, such as probability, signal theory, and pattern recognition.

COMMUNICATION SKILLS:
The student develops a project in Matlab that illustrates during the oral exam. The project can be done in groups to demonstrate working group capabilities.

LEARNING SKILLS:
Students will need to be able to read and understand scientific texts and articles in English for in-depth exploration of the topics covered. They should also independently expand their knowledge of the subject to include topics not directly addressed in the course, particularly those connected with the rapid technological developments in the field of Deep Learning and, more generally, in machine learning.

SYLLABUS

Today, deep neural networks surpass traditional hand-crafted algorithms and match human performance in various complex tasks, including image recognition, natural language processing, and prediction models. This course offers a comprehensive introduction to neural networks (NNs), covering traditional feedforward (FFNN) and recurrent (RNN) neural networks, as well as the most advanced deep-learning models like convolutional neural networks (CNN), Variational Autoencoders, and Diffusion models.

The primary objective of the course is to equip students with the theoretical knowledge and practical skills needed to understand and utilize neural networks (NN), while also familiarizing them with deep learning techniques for solving complex engineering challenges.
This goal is pursued in the course by:
• Describing the most important algorithms for NN training (e.g., backpropagation, adaptive gradient algorithms, etc.)
• Illustrating the best practices for successful training and using these models (e.g., dropout, data augmentation, etc.) in a practical session using a phyton environment.
• Providing an overview of the most successful Deep Learning architectures (e.g., convolutional networks, autoencoder for embedding, diffusion models, etc.)
• Providing an overview of the most successful applications with particular emphasis on models for solving visual recognition tasks.

TEXTS

Pattern recognition and machine learning, Christopher Bishop.

Deep Learning, Ian Goodfellow et al.

– slides of the professor

 

 

Digital Signal Processing (block C1-C2-opt)

Digital Signal Processing (block C1-C2-opt)
1 YEAR II semester  6 CFU
ICT and Internet Engineering
Marina RUGGIERI (5cfu)

Tommaso ROSSI (1cfu)

A.Y. 2023-24
A.Y. 2024-25
Code: 8039514
SSD: ING-INF/03

OBJECTIVES

LEARNING OUTCOMES: The course aims at providing to the students the theoretical and practical tools for the development of design capabilities and implementation awareness of Digital Signal Processing (DSP) systems and applications.

KNOWLEDGE AND UNDERSTANDING: Students are envisaged to understand the DSP theoretical, design and algorithm elements and to be able to apply them in design exercises.

APPLYING KNOWLEDGE AND UNDERSTANDING: Students are envisaged to apply broadly and to personalize the design techniques and algorithm approaches taught during the lessons.

MAKING JUDGEMENTS: Students are envisaged to provide a reasoned description of the design and algorithm techniques and tools, with proper integrations and links.

COMMUNICATION SKILLS: Students are envisaged to describe analytically the theoretical elements and to provide a description of the design techniques and the algorithm steps, also providing eventual examples.

LEARNING SKILLS: Students are envisaged to deal with design tools and manuals. The correlation of topics is important, particularly when design trade-offs are concerned.

BACKGROUND

A good mathematical background (in particular on complex numbers, series, functions of complex variable) is strongly recommended.

PROGRAMME

(Prof. M.RUGUERI)

PART I – Discrete-time signals and systems; sampling process; Discrete-time Fourier transform (DTFT); Z-transform; Discrete Fourier Series (DFS).
PART II – Processing algorithms: introduction to processing; Discrete Fourier Transform (DFT); finite and long processing; DFT-based Processing; Fast Fourier Transform (FFT); processing with FFT.
PART III – Filter Design: introduction to digital filters: FIR and IIR classification; structures, design and implementation of IIR and FIR filters; analysis of finite word length effects; DSP system design and applications;
PART IV – Random sequences; processing of random sequences with digital filters; introduction to random sequence estimation; estimators of mean, variance and auto-covariance of random sequences with performance analysis; power spectrum estimation; periodogram and performance analysis; smoothed estimators of the power spectrum and performance analysis; use of FFT in power spectrum estimation.

(Dott. Tommaso ROSSI)

PART V – VLAB: applications with design examples and applications of IIR and FIR filters, Matlab-based lab and exercises; use of Matlab in the power spectrum estimation.

 

VERIFICATION CRITERIA

a) Combination of: design test (written); deepening on DSP System development (written); oral.
The design test is propedeutic to the oral one.
The course offers a verify in progress (with a related recovery date) that if passed exempts from the design test of the exam session.

b) The written exam includes design exercises.
The oral part envisages questions on the whole program and a discussion on the design test.

c) The written exam is scored from FAIL to EXCELLENT. The design test and the oral concur almost evenly to the final score (x/30).

d) The final score is based on the level of knowledge of the theoretical, design and algorithm elements and tools as well as on their effective use in design exercises; in particular, the final evaluation refers to 70% of the student’s knowledge level and for 30% to her/his capability of expressing the knowledge and providing an autonomous judgment in the design and oral exam phases.

The detailed final evaluation criteria are as follows:

Failed exam: deep lack and/or inaccuracy of knowledge and comprehension of topics and design techniques; limited capabilities in analysis and synthesis, critical ability and judgment; designs and topics are presented with a non-coherent and technically inadequate approach.
18-20: sufficient knowledge and comprehension of topics and design techniques with possible imperfections; sufficient capabilities in analysis, synthesis and autonomous judgment; designs and topics are presented with a not too much coherent and technically appropriate approach.
21-23: flat knowledge and comprehension of topics and design techniques; appropriate capabilities in analysis and synthesis with fair autonomous judgment; designs and topics are presented with sufficient coherency and technically appropriate approach.
24-26: more than fair knowledge and comprehension of topics and design techniques; good capabilities in analysis and synthesis with good autonomous judgment; designs and topics are presented with coherency and technically appropriate approach.
27-29: complete knowledge and very good comprehension of topics and design techniques; remarkable capabilities in analysis and synthesis with very good autonomous judgment; designs and topics are presented with a rigorous and technically very appropriate approach.
30-30L: excellent knowledge and complete comprehension of topics and design techniques; excellent capabilities in analysis and synthesis with excellent autonomous judgment and originality; designs and topics are presented with a rigorous and technically excellent approach.

TEXTBOOKS

[1] “Digital Signal Processing Exercises and Applications”, Marina Ruggieri, Michele Luglio, Marco Pratesi. Aracne Editrice, ISBN: 88-7999-907-9.
[2] The River Publishers’ Series in Signal, Image & Speech Processing, “An Introduction to Digital Signal Processing: A Focus on Implementation”, Stanley Henry Mneney. River Publishers, ISBN: 978-87-92329-12-7.
[3] Slides (exercises are also included therein) published on the teaching website.