ANNOUNCEMENT FOR PROSPECTIVE PHD STUDENTS
Prof. Francisco Alegria, Instituto Superior Técnico, University of Lisbon
Workplan Proposal
The most amazing result of evolution in the entire known Universe is the Human brain. The enormous sophistication of processes that go on inside our brains is almost unfathomable. How can we do so much with what amounts to a collection of living cells?
Much is now known about the elements of the Human brain, the neurons, their connections, their organization, and their emergent behavior. One of the ultimate achievements of science and engineering would be for our brains, in collaboration, to create an artificial brain that would be able to perceive and reason like our own brains. The study goal proposed here is to take a small step in that direction.
The Human brain is mostly made up of neurons and their interconnections through axons and dendrites [F. Alegria, "Measurement challenges in trying to understand our brain", Measurement 46 (2013), pp. 2950-2962, Elsevier]. The majority of the 100 billion neurons are in the neocortex where sensing and reasoning take place. The neocortex has the size and thickness of a dinner napkin and is crumpled up inside the skull. The neurons are organized in six layers with specific types of cells and connections. Layer 4, for instance, receives connections from our sensors like the eyes, ears, and skin. Different regions are connected in a hierarchical fashion where regions receiving signals from the sensors and sending signals to the muscles are lower down in the hierarchy and where those regions connect to other regions which subsequently connect to other regions which carry out what is usually called higher cognitive functions.
It is believed that what all those neurons are doing is essentially the same thing but on different levels of abstraction. They learn sensorimotor sequences which are context-sensitive. That context comes from neurons from the rest of the neocortex. Different functions arise in different parts of the brain essentially due to the diversity of locations where those signals come from. In this way, the neocortex learns to predict what comes next. The confirmation or not of those continually created hypotheses modulates further learning.
The proposed work plan consists in creating, on a computer, a "brain" that would be the first stepping stone to one day achieving the goal of mimicking our own brains. It necessarily must be done in a small scale with limited ambition but that, nonetheless, follows the same principles as the brains that evolution has produced.
A starting line of action is described in what follows which may inevitably be subject to redirections and reframing as the work progresses. It is divided into stages of increasing complexity which make use of the achievement of previous stages.
Workplan stages:
1) World of Letters: Create a prototypical neocortex whose "world" is made up of characters (not the image of characters, yet) and where the sensorimotor sequences are made up of letters that can be "read" from left to right and from right to left depending on the movement of a 1-dimensional actuator. The goal of this first simulation is for the application to learn what words the sense that valid words are the ones that it encounters often, and invalid words are those that are seldom encountered, presumably arising from sensor errors. The goal of the stage is to learn a dictionary of words through mere exposure to human written texts. Obviously, the intent is not to replicate what existing software can very easily do but to start implementing the framework of neurons, connection, and learning procedures that will be used in the next stages of development.
2) World of Sounds: Create a prototypical neocortex whose "world" is made up of auditory signals which is a continuous sequence of phonemes that arise from a text that is "heard" by the application. A simple "ear" is implemented which will be able to learn sounds that come from words that are heard.
3) World of Letters and Sounds: Create a prototypical neocortex whose "world" is made up of auditory signals like the ones used in stage 2 and words like the ones in stage 1. This stage will permit the development of an association of different sensory modalities. The goal would be for the artificial neocortex to predict the letters from the sounds heard and predict the sounds heard from the letters presented to the "letter reading sensor" from stage 1.
4) World of 2D Images of Letters: Create a prototypical neocortex whose "2D world" is made up of actual drawings of letters and words to implement a rudimentary visual sensor. This stage would focus on the prediction of visual sensory information given the 2D movement of an "eye sensor". This would then be associated with the "ear sensor" from stage 2. Naturally, these association capabilities would mimic the ones developed in stage 3.
5) World of Moving Images: Create a prototypical neocortex whose "2D world" is made of moving images. This stage would focus on learning sensorimotor visual sequences through the action of the "eye", scanning the movie, and, at the same time having the content of the image change regardless of the "eye" movement.
This work plan is meant to be part of a scientific project proposal to be submitted to the Foundation for Science and Technology (FCT) in 2023. It constitutes the first step in a decade-long endeavor that would involve other scientific projects, researchers, and Ph.D. students. The long-term program would focus on this and other parts of the brain which are tentatively envisioned as:
A) Neocortex and sensorimotor perception (which includes the current work plan proposal).
B) Memory and the hippocampus.
C) Embodiment and the cerebellum.
D) Higher cognitive functions.
The work proposed here would take place at Instituto de Telecomunicações and would be carried out in the context of a doctorate degree at Instituto Superior Técnico, specifically in the Ph.D. Programme in Electrical and Computer Engineering.
Please send your applications to falegria@lx.it.pt.
December 6th, 2022
Prof. Francisco André Corrêa Alegria
Associate Professor at Instituto Superior Técnico, University of Lisbon, Portugal