Snippets of the Second NeuroMat Workshop

Nov 30, 2016

The Research, Innovation and Dissemination Center for Neuromathematics has hosted this month its second workshop, gathering members of the three teams --research, technology transfer and scientific dissemination-- and guests. The event took place at the University of São Paulo with support from the São Paulo Research Foundation and combined short and long presentations, roundtables and working groups. This workshop provided a sense of the evolution of activities and was an opportunity to exchange interdisciplinary perspectives on ongoing and future lines of action. This piece provides snippets of the action at this event, officially called "New Frontiers in NeuroMathematics” and ran from November 22 to 25.

On day 1

* Eva Löcherbach presented NeuroMat’s model for neural nets as a leaky integrate-and-fire model with random threshold, thus incorporating the contribution by Brochini et al. (2016).

* Löcherbach’s presentation indicated ongoing work on the Galves-Löcherbach model, for instance, on testing interaction graphs, estimating weights of interaction, comparing the model to other stochastic models and developing a notion of criticality associated to a dynamical phase transition. The presentation is available here.

* Brochini et al. (2016) was mentioned at least in six presentations, thus indicating the relevance of this specific article in the development of a conceptual framework for neurobiological data.

* There were 27 presentations; 20 slide presentations are available here.

* Participants included NeuroMat PI and associate investigators, as well as guests. A poster with participants is available here.

* NeuroMat mathematical modeling of brain functioning is done in a context of high level of uncertainty. According to NeuroMat PI Jorge Stolfi, the behavior of neurons is mostly unknown, the topology is mostly unknown, the synaptic weights are unknown, the spiking history of neurons is mostly unknown, the meaning of spikes is mostly unknown.

* This level of uncertainty in modeling brain functioning does not only have negative aspects. “At the end,” said Stolfi, “we do not need to take into account many details."

* In his presentation, Stolfi discussed the challenge of including plasticity in the GL model.

* New possible uses of the GL model were the topic of NeuroMat investigator Osame Kinouchi, who opened a discussion on twenty lines of research that could rely on this specific leaky integrate-and-fire model.

* Kinouchi’s research agenda, available here, included proposals for neuroscience, plant neurobiology, sociophysics, geophysics, colonization processes, epidemiology, and econophysics.

* “Is there anything the GL model is unable to explain?”, NeuroMat coordinator Antonio Galves joked about Kinouchi’s twenty-project presentation.

* During the afternoon, a working group discussed the communications strategy of NeuroMat’s network on brachial-plexus injuries (BPI), called Abraço. A key aspect of the group was to consider a strategy of presenting to a broad audience why this type of injuries has become a public-health emergency situation in Brazil.

On day 2

* Day 2 was dedicated to NeuroMat’s technological-transfer and therapeutic initiatives. This day was opened by NeuroMat PI Claudia D. Vargas and associate investigator André Helene. The focus was on activities at the Deolindo Couto Neurology Institute, pertaining to BPI, and on the ongoing NeuroMat Parkinson Network, called Amparo.

* Claudia D. Vargas showed how BPI may be seen as a general framework of activities, combining research on plasticity and interventions on public policies. This is the ambition of Abraço, that Vargas presented here.

* José Vicente Martins, Gabriel Freire Miranda and Fernanda Torres, from the Federal University of Rio de Janeiro, presented clinical, experimental protocols associated to BPI. Vicente showed videos of patients, which signaled how important research advances could be to improve the condition of patients after surgery.

* André Helene talked on NeuroMat’s Goalkeeper Game indicating potential for therapeutic innovation.

* Both initiative of BPI and GK are associated to challenges of data collection and visualization. This has been the object of activity of NeuroMat’s technology-transfer team, that has developed the Neuroscience Experiments System (NES).

* NES is a piece of an ongoing effort to generate a NeuroMat public database for neurobiological data. This effort was presented by NeuroMat investigator Kelly Braghetto.

* During this activity, a discussion arose on technical challenges to incorporate in the development of NeuroMat tools a means of storing and proving resources for visualization and eventually analysis of data produced in the GK game.

On day 3

* NeuroMat researcher Bruno Monte also discussed on the GK game, now indicating how this game may be associated to an innovative research protocol in Neuromathematics.

* The GK game is associated to a major line of research within NeuroMat around the conjecture that the brain functions as a statistician in the way it processes information. This conjecture was presented by Antonio Galves, to which followed small presentations pertaining to mathematical and therapeutical advances associated to this line of research.

* Cissa Soares presented on the devising of an experimental protocol relying on the GK game to compare how individuals with BPI and individuals without the condition respond to stimuli in the game and generate a game strategy.

* "We propose a stochastic modeling to explore how our brain represents sequential sensor motor structures,” Cissa Soares said.

* A challenge, someone reacted in the audience, is to consider a context tree for sensory and motor dynamics.

* Predicting upcoming sensorimotor events was the topic of Maria Luiza Rangel’s presentation, here, where she claimed that the ability to predict an upcoming action is an intrinsic property of the motor system.

* Does BPI affect prediction ability? BPI leads to severe impairment of upper limb function, so an experimental question is to be considered, Maria Luiz Rangel claimed.

* As the session on the statistician brain ended, Antonio Carlos Roque and Nilton Kamiji presented on computational challenges and perspectives associated to the GL model for stochastic models for neural nets.

* NeuroMat PI Antonio Roque presented on neuronal spontaneous activity, that occurs in the absence of external stimuli. A classical hypothesis is that the cortex operates at a balanced state in which excitatory and inhibitory input currents to a neuron mutually cancel. Neuronal spikes are caused by fluctuations around average net input. This would explain irregular spiking activity of neurons. The classical model is based on an Erdös-Renyi graph (80% excitatory neurons) with sparse connectivity and integrate-and-fire dynamics. In this scenario, it is assumed that inhibitory synapses are stronger than excitatory synapses and external stimulus.

* According to Roque and Kamiji, the challenge is to move beyond classical models, for instance, adopting networks with more realistic architectures and stochastic neuron models, considering large-scale models (i.e., a local cortical network), adding multi scale models, area models, with hierarchy of large-scale network models, and including high rate of variability in neuronal activity.

* A special challenge to use the GL model for computational simulation is to answer: when does a spike occur?, said Roque.

* Kamiji presented on a project he has worked on in order to incorporate elements and assumptions from computational neuroscientists onto the GL model and then have an efficient strategy to model neural activity. An example of how this would look like is the creation of a GL(Izhikevich) RS set of neurons.

On day 4

* Day 4 was set as open panels on stochastic models uses in analyzing data and neural networks. Participants of a first panel included NeuroMat investigators Christophe Pouzat, Daniel Yasumasa Takahashi and Sergio Neuenschwander.

* During the first panel, Sergio Neuenschwander presented his work on neuronal rhythms and how to explain complex processes in the brain through synchronization. His studies mainly focused on the analysis of gamma waves, and how their oscillations can affect processes such as vision, for instance. Presentation slides are here.

* A sequence of short presentations followed, indicating new lines of research in modeling neural dynamics.

* NeuroMat investigator Ludmila Brochini talked about using a new model of the Galves-Löcherbach class without leakage to infer anatomical or functional connectivity of neurons with real spike train data of neurobiological interest.

* Anatoly Yambartsev, a NeuroMat investigator, presented a study on the strengthening of the synapses between neurons and a model for a network of binary point neurons based on the Ising model, combining stochastic dynamics with coupling constants. Presentation slides are here.

* Miguel Abadi presented a study on phase transition in an integrate and fire model. He relied on a finite number of neurons and discrete time to calculate the decision probability of only one neuron firing in the system with a leakage rate.

* “A stochastic model for neural nets” was the last panel presented on day 4, chaired by Claudia D. Vargas.

* Pierre Hodara, a newly-arrived researcher at NeuroMat, presented a study on estimating the spiking rate for a model of interacting neurons through a deterministic Markov process, building an estimator for the intensity function. Presentation slides are here.

* Guillem Via presented a study with computational modeling of a network of excitatory stochastic spiking neurons with a number tending to infinity, together with an analytic, via methor of characteristics, and a numerical solution, via Lax-Wendroff method. Presentation slides are here.

* The second workshop was recorded, and access to videos will be provided shortly.

This piece is part of NeuroMat's Newsletter #34. Read more here

Share on Twitter Share on Facebook