24 July, 2024
Natal, Brazil
The following are invited and contributing speakers for the workshop.
Now Closed!
| Chair: Marilyn Gatica | |
| 14:00 - 14:45 | Patricio Orio University of Valparaiso "The emergence of high-order interdependences in learning neural networks" |
| 14:45 - 15:15 | Samy Castro Institut de Neurosciences des Systèmes "On the role of synergy in cortical communication and frequency diversity within canonical cortical circuits" |
| 15:15 - 15:50 | Coffee break |
| Chair: Samy Castro | |
| 15:50 - 16:30 | Marilyn Gatica Northeastern University London "High-order functional interactions in health and therapy: aging and transcranial ultrasound stimulation" |
| 16:30 - 16:50 | Tousif Jamal Radboud University "NETSCOPE: Information-Theory based Network Discovery and Analysis" |
| 16:50 - 17:10 | Diego Becerra University of Valparaiso "Finding lower bounds of brain complexity: integration of information and multivariate information metrics in Caenorhabditis elegans" |
| 17:10 - 17:30 | Vinicius Lima Aix-Marseille University "Self-organized emergence of multi-areal information processing in a non human primate connectome-based model" |
Patricio Orio - "The emergence of high-order interdependences in learning neural networks"
Marilyn Gatica - "High-order functional interactions in health and therapy: aging and transcranial ultrasound stimulation"
In this talk, I will discuss the relevance of understanding high-order interactions in clinical applications using fMRI data, highlighting their significance in healthy aging in humans and their potential in therapeutic applications, such as transcranial ultrasound stimulation, in both human and macaque brains.
Samy Castro - "On the role of synergy in cortical communication and frequency diversity within canonical cortical circuits"
Cognition has been correlated to coordination among cortical regions. Specifically, oscillatory activity coordinates in different and specific frequency bands to match specific tasks, such as bottom-up processing for saliency tasks (i.e. gamma band) and top-down processing for attention tasks (i.e. alpha-beta band). In cortical regions, it has been
proposed that there is a gradient of frequencies across the brain, with high frequencies showing an increasing gradient from deep to superficial layers and low frequencies showing a decreasing gradient. Hypothesis about the origin of this frequency diversity ranges from specific cellular types (e.g. fast-PV, SOM) being associated with each frequency band, to
their functions in cortical communication, where feedforward connections carrying high-frequency/bottom-up information and feedback connections carry low-frequency/top-down information. We address this hypothesis using rate models of coupled and homogeneous regions embedded in a realistic multi-layer cortical organization.
Our findings demonstrate that different layers exhibit specific and segregated frequencies without the necessity of interneuronal diversity. Moreover, this frequency segregation appears to be a by-product of chaotic, self-organized collective dynamics rather than the
hardwired anatomy of the canonical cortical circuit.
Considering that the information processing capabilities of the systems depend on the coordination of at least three elements, it is natural to consider high-order interactions in the analysis. Through high-order information analyses (in terms of S- and O-information), we find that the canonical circuit exhibits exceptional information processing capacities. It gives rise to collective dynamics that simultaneously maximize “complexity” (the coexistence of integration and segregation) and inter-regional “synergy” in areas of maximal frequency diversity. This synergy in frequency diversity was also observed in randomized models of the
connection patterns. Next, we evaluated the components across the cortical column that maximize synergy to identify those carrying distinctive information in these interactions. We found that the components driving synergy are a mix of superficial and deep layers within a single cortical column and a mix of low and high hierarchy regions between two columns. In conclusion, the emergence of frequency diversity is tightly linked to synergistic coordination in a structurally relevant way. Overall, this work suggests that synergy may be a necessary mechanism for frequency diversity, which underpins cortical communication.
Diego Becerra - "Finding lower bounds of brain complexity: integration of information and multivariate information metrics in Caenorhabditis elegans"
Multivariate informational metrics have been recently developed and applied to study complex aspects of brain functioning in vertebrates, specifically humans, other primates, and rodents. However, the application of such metrics has methodological constraints. For example, combinatorial explosion due to the size of the power set for searching bipartitions when we try to calculate integration of information using Phi; the O(n^2) time requirement of k-nearest-neighbor method when calculating multivariate mutual information; and the dependence of the timeseries length of all previous metrics. Those problems are hard to tackle in systems with numbers of neurons exceeding ten thousand millions, but almost no research has been done so far in simpler nervous systems. Here, we use calcium imaging recordings of C. elegans, a 302-neuron nematode, during sleep-wakefulness transitions, and under isoflurane anesthesia, to probe the behavior of Phi-ID (which combines partial information decomposition and Phi-2.0) and O-information (which measures statistical synergy and redundancy in systems with more than 3 parts) metrics.
Higher average values of Φ-R (a Phi-ID based metric) during wakefulness were found, meaning that pairwise synergy is higher while awake. Yet, O-information shows no difference for both conditions when considering triplets of neurons, and shows higher synergy while awake only when we sample a subset of 10 or more neurons. The apparent discrepancy between Φ-ID and O-information might be due to the different way in which they treat the time-dependency of the data. Different hubs of highly synergistic neurons are identified for sleep and wakefulness using both methods.
Tousif Jamal - "NETSCOPE: Information-Theory based Network Discovery and Analysis"
The brain encompasses molecular, cellular, structural, and functional networks, each
representing different levels of organization and complexity. These networks are interconnected,
facilitating the brain's capabilities in information processing, bodily regulation, and behavioral control. Despite the critical need to understand how higher-level networks (such as structural and functional) emerge from lower-level interactions (like molecular and cellular), a universally applicable analytical method for brain networks is lacking. To bridge this gap, we introduce NETSCOPE, an open-source toolbox that employs mutual information (MI) and variation of information (VI) to delineate weighted network architectures. We validated NETSCOPE's accuracy using synthetic data and through the reconstruction of five molecular networks in Saccharomyces cerevisiae. Demonstrating its versatility, we used NETSCOPE to identify cell type-specific transcriptional networks and to reconstruct brain-wide neural networks that process tactile information in mice. NETSCOPE is compatible with multiple platforms including Python, Google Colab, Jupyter Notebook, MATLAB, and Octave. Its potential extends beyond its core functionality and include, but not limited to, the development of bioinspired sparse artificial networks.
Vinicius Lima - "Self-organized emergence of multi-areal information processing in a non human primate connectome-based model"
Brain function requires the maintenance, exchange and merging of information conveyed by multiple brain regions, which is achieved by coordinated patterns of distributed neural activity. Brain structure constrain pathways through which information can flow, yet doesn’t determine them in a fully hardwired manner. Indeed flexibility in information processing is essential to behavior, since it allows for rapidly adapting to changes in the environment. It is the system’s level collective dynamics which provides a rich intrinsic repertoire of possible states that can be exploited to enable alternative modalities of emergent information processing. Brain functional computations thus stem from both structure and dynamics. Here we focus on working memory (WM), as a key cognitive function involving rich, multi-faceted operations of network-level information processing. How to quantify which type of primitive computing operations is performed based on the observed dynamics of a given brain area? And, how do these different operations localize in space and time at a system-wide level? We quantify the occurrence of the three primitive operations of active storage, transfer and synergistic modification of information based on in silico experiments about the perception, maintenance in working memory and attentional gating of visual stimuli. Specifically, we perform simulations of a large-scale connectome-based model of non human primate (NHP) cortical networks and study the propagation of the activity evoked by presented visual stimuli through the cortical hierarchy. We then compute the amount of storage (for each region), the information transfer (between pairs of regions) and information modification (between pairs of areas and the stimulus) as the stimulus propagates and is processed. Remarkably, while the feed-forward connectivity mediates information transfer, we find that information modification (indicative of top-down gating) is established backwards, with the highest order frontal regions (source of top-down gating) being the first to modify primary visual cortex response, followed only later by lower order regions. Such non-trivial phenomena emerge spontaneously from the interplay between system’s structure and dynamics. We also find that the model's activity, even in the absence of stimulus, gives rise to self-organized patterns of coordinated fluctuation across distributed networks. Groups of areas display packets of spatially and temporally uninterrupted transient activity, that we define as a novel type of coordinated activity events called “co-crackle assemblies” (CrAs). If the system has the intrinsic capacity to produce transient spatio-temporally structured activity, are these transients exploited for functional computations? To explore this question, we turn to the analysis of local field potentials, recorded from dozens of cortical areas simultaneously in NHPs performing a visual WM task. We first find that CrAs predicted by the model are detected as well in actual in vivo recordings. Second, they are better predictors of which areas hold WM information than local features (like absolute oscillatory power). In summary, those results suggest that neural information processing can rely on intrinsic multi-regional coordinated activity patterns, recruited “on demand”. Thus, analyzes going beyond pairwise correlations can be used to better track computations emerging from the system's collective dynamics.
This workshop has been run at CNS for over two decades now -- links to the websites for the previous workshops in this series are below:
Image modified from an original credited to dow_at_uoregon.edu, obtained here (distributed without restrictions); modified image available here under CC-BY-3.0