International Teaching | GENERATIVE AI
International Teaching GENERATIVE AI
Back
Lessons Timetable
cod. 0622700124
GENERATIVE AI
| 0622700124 | |
| DEPARTMENT OF INFORMATION AND ELECTRICAL ENGINEERING AND APPLIED MATHEMATICS | |
| EQF7 | |
| COMPUTER ENGINEERING | |
| 2025/2026 |
| YEAR OF COURSE 2 | |
| YEAR OF DIDACTIC SYSTEM 2022 | |
| AUTUMN SEMESTER |
| SSD | CFU | HOURS | ACTIVITY | |
|---|---|---|---|---|
| ING-INF/05 | 3 | 24 | LESSONS | |
| ING-INF/05 | 3 | 24 | LAB |
| Objectives | |
|---|---|
| THE COURSE INTRODUCES THE FUNDAMENTAL CONCEPTS OF GENERATIVE MODELS, AND PRESENTS STATE-OF-THE-ART ARCHITECTURES BASED ON DEEP LEARNING FOR THE GENERATION OF COMPLEX INFORMATION. KNOWLEDGE AND UNDERSTANDING FUNDAMENTAL CONCEPTS ON GENERATIVE MODELS. CONDITIONAL GENERATIVE MODELS. AUTO-ENCODERS AND VARIATIONAL AUTO-ENCODERS (VAE). GENERATIVE ADVERSARIAL NETWORKS (GAN). DIFFUSION MODELS. APPLIED KNOWLEDGE AND UNDERSTANDING DESIGN AND IMPLEMENT SYSTEMS BASED ON MACHINE LEARNING FOR THE GENERATION OF COMPLEX DATA (E.G. IMAGES) STARTING FROM EXAMPLES, USING STATE-OF-THE-ART ARCHITECTURES. |
| Prerequisites | |
|---|---|
| PROPAEDEUTIC EXAMS: MACHINE LEARNING THE STUDENT MUST BE PROFICIENT WITH THE FUNDAMENTAL CONCEPTS OF PROBABILITY THEORY. |
| Contents | |
|---|---|
| DIDACTIC UNIT 1: INTRODUCTORY CONCEPTS (LECTURE/PRACTICE/LABORATORY 6/0/2) - 1 (2 LECTURE HOURS): RECALL OF PROBABILITY THEORY CONCEPTS - 2 (2 LECTURE HOURS): OUTLINE OF INFORMATION THEORY - 3 (2 LECTURE HOURS): DISCRIMINATIVE MODELS AND GENERATIVE MODELS. CONDITIONAL GENERATIVE MODELS. SAMPLING OF A RANDOM VARIABLE. - 4 (2 LABORATORY HOURS): EXAMPLES OF THE USE OF SAMPLING TECHNIQUES. KNOWLEDGE AND UNDERSTANDING: FUNDAMENTAL CONCEPTS ON GENERATIVE MODELS. CONDITIONAL GENERATIVE MODELS. APPLIED KNOWLEDGE AND UNDERSTANDING: APPLY BASIC GENERATIVE MODELS TO GENERATE SAMPLES OF SIMPLE DISTRIBUTIONS. DIDACTIC UNIT 2: TRANSFORMERS (4/0/4) - 5 (2 LECTURE HOURS): DOT-PRODUCT ATTENTION. SELF-ATTENTION. ENCODERS. POSITIONAL ENCODING. - 6 (2 LABORATORY HOURS): USE OF ENCODERS FOR ANALYZING SEQUENCES. - 7 (2 LECTURE HOURS): MASKED ATTENTION. DECODERS. GENERATIVE MODELS FOR TEXT. MULTIMODAL TRANSFORMERS. - 8 (2 LABORATORY HOURS): USE OF TRANSFORMERS FOR THE GENERATION OF SEQUENCES. KNOWLEDGE AND UNDERSTANDING: ATTENTION AND TRANSFORMERS. APPLIED KNOWLEDGE AND UNDERSTANDING: DESIGN AND IMPLEMENT TRANSFORMER-BASED SYSTEMS FOR COMPLEX DATA GENERATION AND OTHER DATA ANALYSIS AND PROCESSING PROBLEMS. DIDACTIC UNIT 3: AUTO-ENCODERS (LECTURE/PRACTICE/LABORATORY 4/0/4) - 9 (2 LECTURE HOURS): DETERMINISTIC AUTO-ENCODERS. - 10 (2 LABORATORY HOURS): USE OF DETERMINISTIC AUTO-ENCODERS FOR DENOISING AND DIMENSIONALITY REDUCTION. - 11 (2 LECTURE HOURS): VARIATIONAL AUTO-ENCODERS (VAE). - 12 (2 LABORATORY HOURS): USE OF AUTO-ENCODERS FOR IMAGE GENERATION AND DEEP-FAKES. KNOWLEDGE AND UNDERSTANDING: AUTO-ENCODERS AND VARIATIONAL AUTO-ENCODERS (VAE). APPLIED KNOWLEDGE AND UNDERSTANDING: DESIGN AND IMPLEMENT AUTO-ENCODER-BASED SYSTEMS FOR COMPLEX DATA GENERATION AND OTHER DATA ANALYSIS AND PROCESSING PROBLEMS. DIDACTIC UNIT 4: GENERATIVE ADVERSARIAL NETWORKS (LECTURE/PRACTICE/LABORATORY 4/0/6) - 13 (2 LECTURE HOURS) ADVERSARIAL MACHINE LEARNING. GENERATIVE ADVERSARIAL NETWORKS (GAN) AND CONDITIONAL GAN. - 14 (2 LECTURE HOURS) GAN TRAINING PROBLEMS. WGANS. CYCLE GANS. - 15 (2 LABORATORY HOURS) USE OF GAN FOR IMAGE GENERATION AND IMAGE-TO-IMAGE TRANSLATION. - 16 (2 LABORATORY HOURS) USE OF GAN FOR IMAGE ALTERATION (E.G. FACE AGING). - 17 (2 LABORATORY HOURS) USE OF GAN FOR AUDIO GENERATION. KNOWLEDGE AND UNDERSTANDING: GENERATIVE ADVERSARIAL NETWORKS (GANS). APPLIED KNOWLEDGE AND UNDERSTANDING: DESIGN AND IMPLEMENT GAN-BASED SYSTEMS FOR COMPLEX DATA GENERATION AND OTHER DATA ANALYSIS AND PROCESSING PROBLEMS. DIDACTIC UNIT 5: DIFFUSION MODELS (LECTURE/PRACTICE/LABORATORY 6/0/8) - 18 (2 LECTURE HOURS) DIFFUSION MODELS. DENOISING DIFFUSION PROBABILISTIC MODELS (DDPM). DENOISING DIFFUSION PROBABILISTIC MODEL (DDPM). - 19 (2 LECTURE HOURS) TRAINING OF DIFFUSION MODELS. DENOISING DIFFUSION IMPLICIT MODELS (DDIM) - 20 (2 LABORATORY HOURS) USE OF DIFFUSION MODELS TO GENERATE IMAGES. - 21 (2 LECTURE HOURS) GUIDED DIFFUSION FOR THE CREATION OF CONDITIONAL GENERATIVE MODELS. LATENT DIFFUSION MODELS. - 22 (2 LABORATORY HOURS) USING DIFFUSION MODELS FOR CONDITIONAL IMAGE GENERATION. - 23 (2 LABORATORY HOURS) USE OF DIFFUSION MODELS FOR VIDEO GENERATION. - 24 (2 LABORATORY HOURS) USE OF DIFFUSION MODELS FOR GENERATION OF OTHER TYPES OF DATA. KNOWLEDGE AND UNDERSTANDING: DIFFUSION MODELS. APPLIED KNOWLEDGE AND UNDERSTANDING: DESIGN AND IMPLEMENT SYSTEMS BASED ON DIFFUSION MODELS FOR COMPLEX DATA GENERATION AND OTHER DATA ANALYSIS AND PROCESSING PROBLEMS. TOTAL LECTURE/PRACTICE/LABORATORY HOURS: 24/0/24 |
| Teaching Methods | |
|---|---|
| THE TEACHING COMBINES THEORETICAL LESSONS WITH LABORATORY EXERCISES, WHICH INCLUDE BOTH THE DEVELOPMENT AND TRAINING OF SYSTEMS BASED ON THE MACHINE LEARNING TECHNIQUES PRESENTED IN THE COURSE, AND THE USE OF PRE-TRAINED GENERATIVE MODELS. EDUCATIONAL ACTIVITIES WILL BE SUPPORTED BY THE USE OF THE UNIVERSITY E-LEARNING PLATFORM TO FACILITATE AND STIMULATE DISCUSSION AND DEBATE AMONG STUDENTS, AS WELL AS FOR THE NOTIFICATION AND DISTRIBUTION OF EDUCATIONAL MATERIALS. ATTENDANCE OF LESSONS IS MANDATORY. TO ACCESS THE FINAL EXAM, THE STUDENT MUST HAVE ATTENDED AT LEAST 70% OF THE HOURS OF LECTURED TEACHING. |
| Verification of learning | |
|---|---|
| THE EXAM IS AIMED AT ASSESSING THE OVERALL KNOWLEDGE AND UNDERSTANDING OF THE CONCEPTS PRESENTED IN THE COURSE, THE ABILITY TO APPLY SUCH KNOWLEDGE TO CREATE AND DOCUMENT SOFTWARE APPLICATIONS AND THE ABILITY TO COMMUNICATE AND PRESENT THE WORK CARRIED OUT. THE TEST CONSISTS OF THE DISCUSSION OF A PRACTICAL PROJECT, CARRIED OUT IN PART DURING THE COURSE, THE PURPOSE OF WHICH IS TO ASSESS THE ABILITY TO APPLY KNOWLEDGE, TO COMMUNICATE THROUGH THE PRESENTATION OF THE RESULTS ACHIEVED, THE INDEPENDENT JUDGMENT, AND FROM AN ORAL INTERVIEW, THE WHOSE PURPOSE IS TO EVALUATE THE KNOWLEDGE AND UNDERSTANDING SKILLS ACQUIRED, THE ABILITY TO LEARN DEMONSTRATED, AND THE ORAL EXPOSURE. THE PRACTICAL PROJECT CONSISTS OF A SMALL SOFTWARE PROJECT THAT APPLIES ONE OF THE TECHNIQUES PRESENTED IN THE COURSE TO A PROBLEM ASSIGNED BY THE TEACHER. THE ORAL INTERVIEW WILL FOCUS ON THE THEORETICAL TOPICS OF THE COURSE AND THE DESIGN AND IMPLEMENTATION CHOICES ADOPTED FOR THE PRACTICAL PROJECT, AND THE ASSESSMENT WILL TAKE ACCOUNT OF THE KNOWLEDGE DEMONSTRATED BY THE STUDENT AND THE DEGREE OF THEIR IN-DEPTH, THE ABILITY TO LEARN, THE QUALITY OF THE EXAMINATION LOCATION. IN THE FINAL EVALUATION, EXPRESSED IN THIRTIETHS, THE EVALUATION OF THE PROJECT AND THE ORAL INTERVIEW WILL WEIGH FOR 40% AND 60% RESPECTIVELY. A "CUM LAUDE" MARK MAY BE AWARDED TO STUDENTS WHO DEMONSTRATE THE ABILITY TO APPLY THE KNOWLEDGE ACQUIRED INDEPENDENTLY EVEN IN CONTEXTS DIFFERENT THAN THOSE PROPOSED IN THE COURSE. |
| Texts | |
|---|---|
| REFERENCE TEXT: CHRISTOPHER M. BISHOP WITH HUGH BISHOP: "DEEP LEARNING. FOUNDATIONS AND CONCEPTS.", SPRINGER, 2024. SUPPLEMENTARY EDUCATIONAL MATERIAL WILL BE AVAILABLE IN THE DEDICATED COURSE SECTION WITHIN THE UNIVERSITY E-LEARNING PLATFORM (HTTPS://ELEARNING.UNISA.IT) ACCESSIBLE TO STUDENTS OF THE COURSE THROUGH THE UNIQUE UNIVERSITY CREDENTIALS |
| More Information | |
|---|---|
| THE COURSE IS HELD IN ENGLISH. |
BETA VERSION Data source ESSE3
