B5799 - APPLIED MACHINE LEARNING-ADVANCED

Learning outcomes

The course aims to apply Machine Learning to complex real-world datasets and relies on the basic concepts introduced in the course "Applied Machine Learning – Basic" that is propedeutic to this one. At the end of the course the student has competences on how to exploit different hardwares for Machine Learning and Deep Learning solutions, both on-premise and via cloud. The student will be also introduced to most recent approaches and active areas of work in the Artificial Intelligence community worldwide.

Course contents

The course includes lectures and hands-on sessions (i.e., practical activities on topics covered in the course), focusing on the following themes.

Lectures on:

• Decision Trees
• Ensemble Learning Techniques
• Support Vector Machines (SVM)
• Introduction to Neural Networks (NN)
• Building a NN for an XNOR gate
• Getting familiar with TensorFlow Playground
• Overview of tools for hands-on work: libraries and frameworks for ML and DL
• TensorFlow and Keras
• Multi-Layer Perceptron (MLP)
• The basic building blocks of a NN: tensors and tensor operations
• Convolutional Neural Networks (CNN)
• Visualizing feature maps, filters, and class outputs
• A historical and evolutionary overview of CNN architectures
• Autoencoders (AE)
• Processing sequences: Recurrent Neural Networks (RNN)
• More on sequences and time series: memory cells, Long Short Term Memory (LSTM), peephole connections, GRU cells
• Training Deep NNs: vanishing/exploding gradients, non-saturating activation functions, batch normalization, and beyond
• Introduction to Natural Language Processing (NLP) problems
• Transformers and Large Language Models (LLMs)
• Machine Learning Operations (MLOps)
• Interpretability and Explainability in AI (XAI)
• A look into the (near) future: an introduction to Quantum Machine Learning

Hands-on sessions on:

• Working with basic benchmark datasets for DL, such as IRIS flowers, MNIST, etc.
• Decision Trees
• Tensor operations
• Ensemble Learning
• Support Vector Machines (SVM)
• Convolutional Neural Networks (CNN)
  
• Autoencoders (AE)
• Recurrent Neural Networks (RNN)
• Explainable AI (XAI)
  

Readings/Bibliography

Since the bibliography for this type of course evolves rather rapidly over time, and in order to avoid listing outdated references on the website, a list of texts and an updated bibliography will be provided to enrolled students during the first lecture of the course, and will be updated as the course evolves. As stated in another section, please note that all teaching materials used during the lectures are made fully available to students in a dedicated digital repository for the course.

Teaching methods

The teaching methodology of the course includes:

• Traditional frontal lectures;
• Interactive notebooks for the practical sessions that involve coding.

All of this is complemented by innovative teaching approaches, which may include the use of Microsoft Teams for remote access, online materials (e.g., videos), and the organization of "flipped classroom" experiences on specific topics.

The slides and notebooks used during the lessons, along with any other materials shown in class, are fully made available to students. In particular, the slides for the lectures are provided to students before the corresponding class, allowing them to take notes directly on the course material.

For the hands-on sessions, the availability of computer rooms is not guaranteed, so students are advised to bring their own laptops

Assessment methods

The exam consists of two parts, each contributing equally to the final grade (15 points each, for a total of 30).

1) WRITTEN TEST (15 points out of 30)

• Objective: Assess the students' understanding and conscious, accurate elaboration of the concepts taught during the lectures.
• Participation in the assessment: To take the written test, students must register via the Almaesami website (almaesami.unibo.it) by the indicated deadlines, typically a few days before the scheduled date.
• Delivery method: The written test is held in person. Exceptions that allow remote participation via Microsoft Teams are considered only if a justified and documented request is submitted to the instructors, and only if official certification verifying the impediment is received by the course coordinator no later than one week before the scheduled exam date.
• Description and evaluation of the assessment: The test consists of a multiple-choice questionnaire, administered either through EOL (eol.unibo.it) or on paper. The questions are of equivalent difficulty, and each one carries the same weight. Correct answers award positive points; incorrect answers may result in negative points, at the instructors’ discretion (who may decide to apply or not apply penalties for wrong answers, for a specific session or the entire academic year). The test duration may range from a minimum of 1 hour to a maximum of 2 hours, depending on the number and difficulty of the questions. To ensure fairness and inclusiveness, the test is never time-pressured: students are always given ample time to reflect and answer all questions calmly — on average, at least 5 minutes per question. During the test, the use of a calculator and any printed reference material (notes, textbooks, handouts, etc.) is allowed, including teaching materials shown and distributed during lectures. However, the use of any device or method allowing communication outside the exam room (whether with people or digital tools), as well as any form of communication between students during the exam, is strictly prohibited. Violation of these rules will result in immediate annulment of the test. The written test can be taken multiple times by registering for future sessions on Almaesami, with the goal of improving the score. The highest score obtained is always retained as the valid one. Each score remains valid for one calendar year; if it expires, the student must retake the written test.

2) PROJECT WORK (15 points out of 30)

• Objective: Assess the student’s ability to apply the course content to a coding-based project.
• Participation in the assessment: No registration via Almaesami is required.
• Delivery method: The project may be chosen by the student, either by submitting their own proposal or by selecting one from a list provided by the instructors. Students must inform the instructors of the project they intend to work on, including a brief description, and must receive instructor approval before proceeding.
• Description and evaluation of the assessment: Project submission guidelines and timelines are presented during the first lecture of the course. Evaluation is based primarily on: clarity in explaining the goals and the approach used to achieve them; correctnes in the ML methodology and quality of the implemented code; documentation and overall presentation of the project. Each project may be submitted up to two times. After the first submission, the student will receive feedback and a proposed score. The student can either accept the score or decide to improve the project based on the feedback and submit it once more. If the score of the second submission is not accepted, the same project cannot be submitted a third time: the student must propose a new project from scratch. In the case of special needs (e.g., deadlines for scholarships or graduation sessions), the submission modalities and deadlines may be discussed on an individual basis with the instructors, who will make every effort to accommodate the student within the bounds of fair and equal treatment for all.

The maximum score a student can obtain is 15+15=30. Honors (“30L”) may be awarded to students who obtain the maximum score in both components of the exam and whose overall performance is unanimously considered excellent by the instructors.

 

Students with Specific Learning Disorders (SLD) or Temporary/Permanent Disabilities are strongly encouraged to contact the University’s dedicated support office in advance (https://site.unibo.it/studenti-con-disabilita-e-dsa/en). This office will propose any necessary accommodations, which must be submitted to the course instructor at least 15 days in advance for approval. The instructor will assess the appropriateness of the accommodations in relation to the course’s learning objectives.

Teaching tools

As explained in the previous sections, lecture content is delivered using:

• Slides;
• Audio-video materials;
• Relevant scientific articles;
• Handouts.

All materials used during the lectures are made available to students in a shared digital repository. Furthermore, as part of the innovative teaching methods adopted, tools such as Teams for videoconferencing, as well as various digital platforms are used to deliver quizzes and assess the assimilation of the course content.

Office hours

See the website of Daniele Bonacorsi

See the website of Luca Clissa