Reservoir SMILES Workshop 17th Nov 2022 @Inria Bordeaux
Reservoir SMILES workshop Thursday November 17th @Inria Bordeaux from 9.15am to 5.15pm.
Room "Grace Hopper 2" (B405), building B, 4th floor.
Free registration & Zoom link (for food please register by 7th of November):
https://docs.google.com/forms/d/e/1FAIpQLSfV7_S5LKCD3UmY00fadhWQ61dKriSbNH-MY2xZ_fTV8zs-0g/viewform?vc=0&c=0&w=1&flr=0&usp=mail_form_link
Zoom link: https://u-bordeaux-fr.zoom.us/j/3080809449?pwd=cTBDUG1lV1RpWU14MHhUQlg1VTV4dz09
Following my Habilitation Defense the day before, I am please to announce the following schedule:
## Morning Session 9.15am-12pm (shifted by +30min)
- 9.15am 9.45 Introduction (Xavier Hinaut)
- 9.30am 10am Christophe Pallier (Neurospin, INSERM-CEA, Saclay)
Using Natural Language Processing Models to model fMRI data during naturalistic story listening
- 10.45am 11am Joanna Rączaszek-Leonard (University of Warsaw, Poland)
Models for reconciling symbols with dynamics in early interactions
## Lunch @Inria 12pm-1.30pm (probably shifted a bit)
(please register if you'd like to join for food)
## Afternoon Session 2pm-4.15pm 5.15pm (we'll keep the same schedule)
- 2pm Alex Pitti (CY Cergy Paris University)
Language as a technological tool for the brain, from sensorimotor learning to digital computation
- (TALK CANCELLED) 2.45pm Julien Diard (CNRS, Université Grenoble Alpes)
COSMO: a family of perceptuo-motor Bayesian models of speech perception and production
- 3pm 4pm Clément Moulin-Frier (Inria Bordeaux)
The dynamics of cultural evolution: from the formation of communication systems to their role in open-ended skill acquisition
Location: Inria Bordeaux, 200 avenue de la vieille tour, 33400 Talence. Room "Grace Hopper 2" (B405), building B, 4th floor. (Please come with an ID card to have a visitor badge)
If you are not able to come please register in case an online option is available, we will let you know.
Julien Diard (talk cancelled)
### Title
COSMO: a family of perceptuo-motor Bayesian models of speech perception and production
### Abstract
Abstract: We have developed over the last 15 years a computational framework for speech communication modeling, called COSMO (for "Communicating Objects by Sensory-Motor Operations"). COSMO allows simultaneously modeling speech perception and speech production, accounts for major theories of speech communication – the auditory and motor theories –, and also provides means to integrate them naturally into perceptuo-motor theories. COSMO models have been applied to study the observed regularities in phonological systems, the complementarity and dynamics of auditory and motor learning, of phonemic and syllabic learning, and how production and perception idiosyncrasies jointly emerge during learning. The most recent COSMO variant, called COSMO-Onset, deals with the issue of speech segmentation, and more precisely, the mechanisms for temporal event detection, such as syllabic onset detection, in the acoustic signal. In this context, the main originality of COSMO-Onset is that it features "top-down" lexical knowledge to complement "bottom-up" envelope processing. We show that their fusion improves temporal segmentation, and thus word recognition, in particular in adverse conditions. The model also aims at accounting for experimental data suggesting that recognition is more efficient when natural speech is isochronous, than for speech artificially made isochronous, or than for natural speech in general.
Joanna Rączaszek-Leonard
### Title
Models for reconciling symbols with dynamics in early interactions
### Abstract
Early interactions have to reconcile symbols with dynamics: out of basically continuous and co-constructed interactive flow children arrive at skills of controlling this flow by various means, including symbols. The extended question that this talk aims to ask is: which dynamics have to be accounted for in such early interactions and which models are available to model them?
Alex Pitti
### Title
Language as a technological tool for the brain, from sensorimotor learning to digital computation
### Abstract
I propose that two complementary neural mechanisms are at work to enable the language-ready brain. I suggest that they can be generalized into one framework, to potentially shed light on information acquisition, memory retention, and the development of higher-order cognitive skills. First, predictive coding links the causes to the effects of sensorimotor coordination during vocal learning. Second, serial order codes permit the recognition of hierachical patterns in sentences to abstract syntactic rules. The coupling between two neural architectures relying on these mechanisms, resp. the cortico-basal system and the fronto-striatal system, may serve for information acquisition, memory retention, and scaffolding. As a surprising extension of this idea, I found that neural networks with serial order codes can reach the limit in terms of information capacity based on Shannon Theory of Information. Although the brain is not a computer, it may use digital computing as a technological tool for cognition and energy efficiency.
[1] Pitti, A. Weidmann, C. Quoy, M. (2022) Digital computing through randomness and order in neural networks, PNAS, 119 (33) e21153351
[2] Pitti, A. et al. (2022) In Search of a Neural Model for Serial Order: A Brain Theory for Memory Development and Higher Level Cognition, IEEE Transactions on Cognitive and Developmental Systems, 14, 2, 279-2
[3] Pitti, A. et al. (2021) Brain-inspired model for early vocal learning and correspondence matching using free-energy optimization. PloS Computational Biology. 17(2) e1008566.
[4] Pitti, A. Quoy, M. and Boucenna, S. and Lavandier, C. (2020) Gated Spiking Neural Network using Iterative Free-Energy Optimization and rank-order coding for structure learning in memory sequences (INFERNO GATE). Neural Networks, 121, 242-258