Application of parametric activation function A string in the task of multimodal data analysis

Data analysis is a dynamically developing field, currently. One of the actual tasks of data analysis is the task of classification. The problem of dividing a specific group of objects into a predetermined number of groups united in various ways is also important. On the other hand, computational performance increases and the volume of observed data increases, therefore, assigning them to certain subgroups becomes more complicated. In this paper, the binary classification problem is solved and a new parametric activation function for the machine learning model under consideration is analyzed. An important difference between the proposed classifier, for example, from standard classifiers based on logistic regression, is the connection with infinitely differentiable splines, the so-called atomic functions. At the same time, it is of interest to study the dependence of the classifier quality on the value of the variable parameter of the activation function. By changing the parameter from the activation function, you can make dependencies on the quality of the presented classifier. The comparison of quality indicators with various parameters of the activation function is considered. As data for model training, cross-validation and testing, an open multimodal data set MELD was used, consisting of a parallel set of videos and their textual interpretation. It is worth noting that MELD is a pre– labeled data corpus. The data were divided into two categories of sentiment analysis: positive and negative. A comparison of the work of a classifier based on parametric AString and logistic regression is given.