• non-parametric copula algorithm
• probability theory
• information theory
• neural coding
• TensorFlow
• high-dimensional dependency
• Copula
• non-linear dependency

Studies in neuroscience tell us that there are brain areas whose activity involves the integration of motor, sensory and cognitive information. State-of-art works in the same field have investigated the relationship between neural activity and behavioral and task variables using different methods, such as the Generalized Linear Model (GLM). The main shortcoming of this approach is the fact that GLM is able to describe only linear dependencies. Recent advances in this field came out with a novel algorithm, which has been developed to model whatever dependency between variables. This algorithm is based on a non-parametric Copula method. Although this algorithm proved to accurately model linear and non-linear dependency, it is still characterized by an important limitation, that is its computational time.
In this work, in order to overcome this issue, a Tensorflow implementation of an already existing algorithm has been provided which exploits the power of GPUs. In addition, the method has been also provided with the possibility to use parametric copula families and the bivariate case has been extended to model high-order dependencies through the pair Copula construction using vine Copulas. A systematic validation of the developed software has been performed by means of simulations analyzing its accuracy, speed and consistency.
Single-neuron resolution data from a mouse performing visual tasks in a virtual reality environment have been analyzed using the developed algorithm. Dependencies between neural activity and predictors, as well as dependencies between pairs of neurons, were investigated.