[Welcome to IPAL] Mathias Quoy

Mathias Quoy is a full time Professor at CY Cergy-Paris University (CYU) and a member of the ETIS Lab.

Graduated in computer science from the ENSEEIHT engineering school in 1990, Mathias Quoy got a PhD in computer science from the National School for Aeronautics and Space (SupAéro) in 1994. In 1995 he spent his post-doc at the Philips Air-Force Lab in Albuquerque (NM) and was then recruited at the Cergy-Pontoise University. In 2009 he spent a sabbatical at the Brain Lab in Reno (NV). From 2014 to 2019 he was director of the ETIS Lab.

From February 2022, he will spend 6 months at IPAL where he will conduct research on the learning of multimodal sequences by artificial neural networks. This research work is part of a joined collaboration between CYU, IPAL, A*STAR and NUS.

Mathias Quoy and Alexandre Pitti from CYU are also co-supervising a PhD student with Nancy Chen from A*STAR-I2R.

Project Motivation: Observations in the Broca area and in the pre-Supplementary Motor Area (pre-SMA) confirmed the existence of neurons sensitive (i.e. responding) to structures only. For instance, some of them were found sensitive to the temporal order in audio sequences and to proto-grammars but not to the particular sound emitted [2, 3]. Other neurons were found active to the syntax of actions performed in motor sequences but not to the particular motor units within [1, 6, 7] and different ones were observed salient to the temporal coherence in visual scenes [8].

Some similar results were found with neurons sensitive to orders, schemata in spatial contexts [9], and to geometrical rules in the recognition of shapes [10] or in visual sequences [11]. Surprisingly, these neurons were all found insensitive to the particular sound, action or visual information composing the sequence presented per se (i.e., the low-level percept), but only to the specific patterns or schemata that they were encoding such as AABB or ABAB or AAAA, or to a relative order in a temporal sequence (e.g., the beginning, the second place or at the end) or to a relative location in space.

These observations are in line with the recent idea that the Broca area plays a more general role than implementing language for perception and production of semantic and rule-based behaviors [12], as being a supra-modal “Syntax Engine” in the broaden sense, to abstract rules in other core domains and modalities such as music and action representation, as well as in visual scene understanding [2, 8, 13, 14, 15, 16]. In line with this, theoretical models describe how the Broca area is implicated in the hierarchical organization of human’s behavior in order to generate sequence of single acts and in the hierarchical control on posterior brain regions [17].

Project Method: On the basis of these hypotheses, Alexandre Pitti and Mathias Quoy, along with their team, created a neurocomputational model named INFERNO GATE (Gated Iterative Free-Energy Optimization for Recurrent Neural Networks [4,5]). INFERNO GATE realizes a working memory based on predictive coding by finding the most accurate spike-based sequence (effect) that minimizes prediction error (or surprise) corresponding to the most probable discrete rank-order code (cause). It implements models of the Broca area (Broadman area 44/45) and of Dorso-Lateral Prefrontal Cortex for chunking spike sequences into rank-order codes. It models as well the Anterior Cingulate and the Orbito-Frontal Cortices to process prediction error optimization for sequence retrieval. The idea is to use this model for sound, visual and motor processing.

In particular the team will investigate the structures emerging from speech processing and compare them to infant and children’s learning capabilities [18].