/publications
On Data-Driven Discovery Of Symbolic Differential Equations From Unsuitable Coordinates Using SINDy-Autoencoders
2022"Machine Learning Methods have evolved to a powerful tool in many fields of science, including physics. While methods have become fast and sophisticated enough to learn a wide variety of tasks, they often come at the cost of interpretability. This problem can be partially overcome by making use of symbolic regression as shown in [1] and [2] , but it requires data in suitable coordinates. In my work, I assess and improve the SINDY-Autoencoder proposed by K. Champion et al. [3] , which is designed to learn not only symbolic equations of motion from high-dimensional data, but also the coordinates in which the equations would be most conveniently formulated. I conduct a verification and replication of the proposed method before creating and evaluating variants which eventually lead to better and more reliable results."
