Validating χ-separation using phantom simulations

How can we validate χ-separation algorithm? In the absence of ground truth to validate χ-separation, my project aims to validate the χ-separation results using realistic in-silico head phantom simulations. Simulations offer a valuable advantage by providing a controlled environment where we can define and manipulate various parameters with known ground truth values. By choosing specific values for the simulation, we can create a ground truth against which we can compare the results obtained through the χ-separation algorithm.

Continue reading

Experimenting with Occlusion methods to visualize the features learned by a CNN from audio or visual inputs

This project has for goal to explore, understand and learn how to create comprehensive visualizations of the features learned by a convolution neural network, whether the model is specialized in auditory or visual input.

Continue reading