Skip to content

Latest commit

 

History

History
13 lines (11 loc) · 1.45 KB

README.md

File metadata and controls

13 lines (11 loc) · 1.45 KB

This repo

This repo is a work in progress, if you stumble across this repo I'm by no means a tensorflow expert so if you spot something amiss I would appreciate any feedback! Thank you

Analog RNN paper

Cool interpretation of the wave equation reformulated as the RNN update equations. This is an attempt to replicate the work thought about in https://arxiv.org/pdf/1904.12831.pdf. (This isn't my original work! Just trying to have a play from a cool paper and their code found at - https://github.com/fancompute/wavetorch).

Summary

The general idea of the paper is by reformulating the wave equation (written out using finite difference methods) as the RNN update equations, a physical wave system can be trained to do a similar task to that of the computations in a traditional RNN. Here the trainable parameter is taken to be the wave speed and the non-linearity provided by activation functions in the RNN are replaced by the wave speed depending on the field as well as non-linearities introduced when taking intensity measurements to give an output. The outcome of this is that you could have a physical manifold custom designed (using an inverse method) for different tasks such as the vowel classification described in the paper without needing a computer whatsoever! It would be a lens that would direct different vowels to different points! I wanted to see how effective this was on some data myself so this repository is my attempt to replicate the work