Duration
25h Th, 20h Pr, 20h Proj.
Number of credits
| Master of Science (MSc) in Electrical Engineering | 5 crédits |
Lecturer
Language(s) of instruction
English language
Organisation and examination
Teaching in the second semester
Schedule
Units courses prerequisite and corequisite
Prerequisite or corequisite units are presented within each program
Learning unit contents
Brains are very bad at counting (i.e., at digital computing) but are excellent at continuously receiving sensory stimuli, making decisions about the received sensory information, and putting those decisions into actions in an ever evolving, highly uncertain, environment. Each decision made by a nervous system is an event along its temporal evolution: a decision happens at a specific time and marks a before and an after it was made. Decision events happen in brains at all scales and levels of organization, from cellular spikes in response to electrical inputs and the emergence of percepts in response to sensory inputs, to the selection of motor commands.
In other words, at all levels and scales of organization, brains function as analog, yet event-based, and adaptive decision-making signal processors.
With the goal of providing the students with the tools to study the mechanisms of brain computing and translate the gained knowledge into engineered machines, this course introduces the basic mathematical and computational modeling principles to describe and analyze brain and brain-inspired computing architectures as event-based, analog signal processors and decision makers. We will particularly explore the importance of the mixed nature, that is, simultaneously analog (continuously evolving in time) and digital (event-based), of neuronal systems for understanding brain computing.
It is highly recommended to take this course together with the Neuromorphic Signal Processing course as the two courses are thought to complement each other.
The course will cover the following table of contents (some themes are optional and will be covered depending on time - optional themes can be developed by students in their final projects):
1. The organization of the central nervous system and the biology of cognition.
2. Early (and not so much) examples of brain-inspired computing architectures:
a. Braitenberg's reactive Vehicles (simple reactive embodiment is just enough);
b. Cybernetics (everything is feedback);
c. Turing machines and automata (the birth of the digital);
d. Perceptrons (learning maps through thresholds);
e. CNN (learning receptive fields);
f. RNN (learning through time and computing through dynamics).
3. Review of bifurcation theory; bifurcations as decision thresholds; bifurcations and feedback; bifurcations and analog/event-based computation.
4. Single neurons as event-based decision makers: when (and how) to spike (or not)? The analog/digital nature of spiking. A glimpse into the molecular nature of neurons: is there a computing network in each neuron?
5. Synapses: converting events back into analog dynamics.
6. Forms of plasticity. Optional: learning in biological and artificial neuronal networks.
7. Receptive fields: filtering incoming stimuli to make informed decisions. Optional: Receptive field adaptation and active sensing.
8. Zooming out: simplified models of neuronal circuits. Decision bifurcations at the circuit level. And zooming back in: the importance of tuneable spiking for adaptive computation in neuronal circuits.
9. Zooming outer: computation through neuronal population dynamics. Decision bifurcations at the population level. And zooming back in again: are spikes important at the population level?
10. Optional: Tools to process and analyze electrophysiological neuronal data from the computing mechanism viewpoint.
11. Putting all together and bridging scales. How do we put the theory into practice? The importance of embodiment. And... everything we did not cover in this course.
Learning outcomes of the learning unit
Through theoretical classes and computational exercises developed in the Julia environment, at the end of the course students will:
- Know the basic biological mechanisms underlying brain computing, including elements of learning.
- Have a panorama of historical and recent theoretical frameworks to model and understand brain-inspired computing architectures.
- Be able to design elementary brain-inspired computing modules using temporal dynamics and bifurcations.
- Optional: Know the tools to extract brain-computing principles from electrophysiological data.
Prerequisite knowledge and skills
Basic linear and nonlinear dynamical and control systems. Some previous knowledge in basic neuroscience and computational neuroscience is welcomed.
Planned learning activities and teaching methods
The course includes both face-to-face lectures, exercise sessions, and projects. For project development, students will be able to access hardware and experimental data from the Department Neuromorphic Lab.
Mode of delivery (face to face, distance learning, hybrid learning)
Face-to-face course
Recommended or required readings
Suggested preliminary readings:
- E. Izhikevich - Dynamical Systems in Neuroscience - MIT Press (Chapters 1-7)
- S.H. Strogatz - Nonlinear Dynamics and Chaos - Perseus Books (Chapters 2,3,5-8)
The course will borrow material from a number of references:
- Braitenberg - Vehicles Experiments - MIT Press
- Dayan & Abbott - Theoretical Neuroscience - MIT Press
- Koch - Biophysics of Computation - Oxford University Press
- Sterling & Laughlin - Principles of Neural Design - MIT Press
- Wiener - Cybernetics - MIT Press
- Wilson - Spikes, Decisions, and Actions - Oxford University Press
Exam(s) in session
Any session
- In-person
written exam AND oral exam
Written work / report
Additional information:
Two project-like homework; final project development and oral presentation/exam.
Work placement(s)
Organizational remarks
Contacts
Alessio Franci.
https://sites.google.com/site/francialessioac/