Resumo: The goal of computer vision is to build machines that “interpret” scenes, in the sense of recognizing objects or other structures in a scene, and provide succinct descriptions of context, actions, and intentions. The goal in computer speech is analogous – “interpret” acoustic (voice) signals. While important advances have been made in specific applications (such as, speech: airline reservations, vision: industrial inspection, medical imagery, remote sensing, animation, etc), there is no machine that can read handwritten theorems on the blackboard, recognize faces in the New York subway, or find and track suspicious people in a soccer game at Maracanã; similarly for speech: we still cannot talk to our laptops instead of typing in.  

While specific questions and specific applications have motivated a great deal of mathematical innovation (e.g. statistical learning theory, modern Monte Carlo optimization and simulation algorithms, differential geometry in infinite dimensions, degenerate elliptic equations, etc), an underpinning mathematical framework for high-level (recognition) problems, is missing. It is quite clear (both from computational and biological studies) that the mathematical frameworks for vision and speech recognition are similar (basically the same) – I believe, as others do, that nature developed this framework first in optimizing the functioning of the cell, and then adapted it to vision, speech, and others cognition tasks. This belief is supported by part (c) below as well as by other studies in signal transduction pathways.

Though we do not have the framework, we know quite a bit of some of the problems that the framework needs to articulate and some of the properties it needs to have. The main purpose of the talk will be to (a) to identify the main sources that make the information processing in vision, speech and biological processes a difficult mathematical problem, (b) argue that the framework needs to exhibit a type of probabilistic hierarchical, syntactic, grammatical structure reminiscent to Chomsky’s systems for linguistics, and (c) demonstrate the overall methodology with a mathematical representation of genes and the finding of genes in a DNA sequence.

Local: Sala da Pós-Graduação do IM/UFAL (prédio antigo)

Data: 09/07/14 (Quarta-Feira)

Hora: 14:00