Characterization of low dimensional embeddings for the generation of closed-form mathematical expressions.

Identifier:  TFG:9481

Handle: https://hdl.handle.net/20.500.11797/TFG9481

Authors:  Svetoslavov Yordanov, David

Abstract:
This bachelor’s degree project aims to present and discuss a number of quantitative experiments carried out with the objective of better understanding and evaluating the performance of a Hierarchical Variational Autoencoder (HVAE) in the context of symbolic regression. HVAE is a recently proposed machine learning model that encodes expression trees into a continuous latent space from which it can then decode new ones. The experiments conducted in this project are split into two groups: latent space characterization and symbolic regression. The first group consists of random sampling, random walks and encode-decode reconstruction experiments. Their aim is to gain insight into the structure of the latent space and how encoded expressions are distributed within it. The symbolic regression experiments will be carried out using simple error minimization algorithms, which will test the effectiveness of HVAE as a plausible expression generator.