Direct-fit to nature an evolutionary perspective on biological (and artificial) neural networks, bioRxiv, 2019-09-10

AbstractEvolution is a blind fitting process by which organisms, over generations, adapt to the niches of an ever-changing environment. Does the mammalian brain use similar brute-force fitting processes to learn how to perceive and act upon the world? Recent advances in training deep neural networks has exposed the power of optimizing millions of synaptic weights to map millions of observations along ecologically relevant objective functions. This class of models has dramatically outstripped simpler, more intuitive models, operating robustly in real-life contexts spanning perception, language, and action coordination. These models do not learn an explicit, human-interpretable representation of the underlying structure of the data; rather, they use local computations to interpolate over task-relevant manifolds in a high-dimensional parameter space. Furthermore, counterintuitively, over-parameterized models, similarly to evolutionary processes, can be simple and parsimonious as they provide a versatile, robust solution for learning a diverse set of functions. In contrast to traditional scientific models, where the ultimate goal is interpretability, over-parameterized models eschew interpretability in favor of solving real-life problems or tasks. We contend that over-parameterized blind fitting presents a radical challenge to many of the underlying assumptions and practices in computational neuroscience and cognitive psychology. At the same time, this shift in perspective informs longstanding debates and establishes unexpected links with evolution, ecological psychology, and artificial life.

biorxiv neuroscience 100-200-users 2019

BrainSpace a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets, bioRxiv, 2019-09-09

AbstractUnderstanding how higher order cognitive function emerges from the underlying brain structure depends on quantifying how the behaviour of discrete regions are integrated within the broader cortical landscape. Recent work has established that this macroscale brain organization and function can be quantified in a compact manner through the use of multivariate machine learning approaches that identify manifolds often described as cortical gradients. By quantifying topographic principles of macroscale organization, cortical gradients lend an analytical framework to study structural and functional brain organization across species, throughout development and aging, and its perturbations in disease. More generally, its macroscale perspective on brain organization offers novel possibilities to investigate the complex relationships between brain structure, function, and cognition in a quantified manner. Here, we present a compact workflow and open-access toolbox that allows for (i) the identification of gradients (from structural or functional imaging data), (ii) their alignment (across subjects or modalities), and (iii) their visualization (in embedding or cortical space). Our toolbox also allows for controlled association studies between gradients with other brain-level features, adjusted with respect to several null models that account for spatial autocorrelation. The toolbox is implemented in both Python and Matlab, programming languages widely used by the neuroimaging and network neuroscience communities. Several use-case examples and validation experiments demonstrate the usage and consistency of our tools for the analysis of functional and microstructural gradients across different spatial scales.

biorxiv neuroscience 0-100-users 2019

 

Created with the audiences framework by Jedidiah Carlson

Powered by Hugo