Idea for VAE #4
Description
Hi . I have the following idea.Imagine a image consisting of to inputs and outputs of a symbolic function concatenated together (something like this [input1,input2,input3,output1,output2]). And we feed this image as a input and the output or target of some kind of VAE. The idea is to encode the symbolic functions used to transform the inputs to the outputs in the latent space and use this encoding like a guided search for program synthesis.I know that something similar was already done for continuous functions ( https://arxiv.org/pdf/2201.05136 ) .Also do you have a discord were we can talk.