Skip to main content
CODE 106738
ACADEMIC YEAR 2023/2024
CREDITS
SCIENTIFIC DISCIPLINARY SECTOR ING-INF/06
LANGUAGE English
TEACHING LOCATION
  • GENOVA
SEMESTER 1° Semester
TEACHING MATERIALS AULAWEB

OVERVIEW

Several real-world applications already rely upon modules based on neural paradigms as the computational engines of AI algorithms (e.g., deep/convolutional neural networks). The course aims at overcoming the formal framework of artificial neural networks and to relate more decisively to models derived from Neurosciences. A multidisciplinary approach, which links bidirectionally with the Brain Sciences is crucial: from one side, it fosters the transfer towards artificial systems of the knowledge gained from the study of biological systems (i.e., models specified in hardware, software and eventually in wetware that embody in an essential form their principles, architectures and functionalities), and, from the other side, it demonstrates the usefulness of the “artificial” approach as a method for the investigation of the nervous systems.

AIMS AND CONTENT

LEARNING OUTCOMES

Neuromorphic models for the representation and distributed processing of multidimensional signals. Computational primitives and architectural schemes. Applications to the development of perceptual engines to enable autonomous behaviors in complex systems and natural environments

 

AIMS AND LEARNING OUTCOMES

Through a reverse engineering of the brain, the course presents methods and techniques for the analysis, simulation, and synthesis of cortical-like perceptual engines. A specific emphasis is given to how information from the external world is coded, represented and eventually transformed in the cerebral cortex at network level. The neuromorphic solutions of perceptual problems that support visually-guided behaviour are taken as application domain examples and case-studies.

At the end of the course, the student will have the proper instruments to understand, describe, and design the functionalities of neuromorphic perceptual systems.

PREREQUISITES

Basic Linear Algebra: vector spaces, bases, scalar product, least squares problem, eigenvalue problems. Basic signal processing (corresponding to Linear Systems): sampling, convolution and Fourier transform of one-variable signals. Basic skills in MATLAB are recommended.

TEACHING METHODS

Traditional lectures. Journal club.  Seminars. Guided practical classes on spiking networks. 

Working students and students with certification of DSA, disability or other special educational needs are advised to contact the teacher at the beginning of the course to agree on teaching and exam methods which, in compliance with the teaching objectives, take into account individual learning methods.

SYLLABUS/CONTENT

Contents follow a dialectic narrative through AI, Computational Vision and Theoretical Neuroscience, touching the following topics:

  • Firing rates models and Neural circuital primitives
  • Retinocortical transformations and models of spatial and spatiotemporal receptive fields.
  • Multichannel harmonic representations for early vision
  • Steerable filters
  • Computational theory of early vision and regularization techniques
  • Population docing
  • Cortical architectures: discrete models and neural field models
  • Feed-forward lateral inhibition and predictive coding
  • Recurrent lateral inhibition, selective amplification and WTA networks
  • Divisive normalization
  • Motion perception
  • Depth perception from stereo vision
  • Neuromorphic solutions: motion and disparity detectors

RECOMMENDED READING/BIBLIOGRAPHY

Slides and other distributed material (available through Aulaweb).

Further reading, for consultation only:

  • P. Dayan and L.F. Abbott. Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems. The MIT Press, 2001.
  • H.A. Mallot. Computational Vision: Information Processing in Perception and Visual Behavior. The MIT Press, 2000.

TEACHERS AND EXAM BOARD

Exam Board

SILVIO PAOLO SABATINI (President)

FRANCESCA PEVERI

FABIO SOLARI

ANDREA CANESSA (President Substitute)

LESSONS

Class schedule

L'orario di tutti gli insegnamenti è consultabile all'indirizzo EasyAcademy.

EXAMS

EXAM DESCRIPTION

Discussion on an example/application/deepening of a topic seen in class proposed in the Journal Clubs (weight 30%). A 20 minute presentation to summarize the main take-home messages of a scientific paper chosen from a list that will be available in Aulaweb. The assignment is to analyze only those parts that mostly relate to the topics presented during classes. An open discussion will follow the presentation, and everyone is invited to participate. The discussion will be also a chance to introduce some supplementary topics, which will be part of the exam’s program.

Oral examination. The exam consists in appropriately framing and discussing two assigned topics (typically, a "theoretical model" and a "computational solution of a perceptual problem"). The oral exam is aimed at verifying the acquisition of the notions presented in the teaching, and at evaluating the ability to analyze and plan a specific problem (weight 70%).

ASSESSMENT METHODS

The student should eventually demonstrate the ability to analyze and synthesize neuromorphic processing paradigms at the cellular, circuit and system level.

In general, in addition to the correctness and completeness of the answer, the evaluation criteria comprise: the relevance to the question, the clarity of the answer, and the ability to synthesize.