Machine | Ben Ahlbrand CV

Tutorial: RNN Fantasy Character Name Generator Pytorch

Tue, 07 Jan 2020 00:00:00 +0000

Advances in sequential modeling via deep learning allows procedurally generated names for RPG characters that belong in Tolkienesque worlds. I used pretty much the simplest sequential model - an RNN, more specifically a variation of CHARNN, to train a network capable of producing interesting names when sampled with various characters and category conditions (the language class).

The categories include:

dwarf
man
elf
ainur

Take a closer look at the dataset here - it’s a dump of names from the Tolkien universe w/a class label.

after just a few minutes training on my nvidia 2070 super, the network is capable of reconstructing the following names:

Dwarf

Romi
Ulmor
Salin

Elf

Galdir
Elenel
Rimegell

Man

Serimin
Parandil
Arandil

Ainur

Miner
Orome
Nain

Pytorch implementation source code available here

Procedural Music with markov chains

Tue, 31 May 2016 00:00:00 +0000

For our final project for the software engineering track at Purdue (department recognized as one of the best projects of the year!), we combined convolutional neural networks and markov chains to generate music based on the “emotions” detected in an image. Taking that and the pixel palette, we constructed a mapping to various music theory concepts and generated music with a music theory oracle that I wrote + statistics to choose among the various rules. Interesting results ensue!

Here’s a sad track :(

Here’s a happy one!

Query By Humming engine

Wed, 17 Dec 2014 00:00:00 +0000

For a course on Information Search and Management, my final research project was developing a query by humming system to detect hummed melodies and find likely matched songs, using auto correlation, damerau levenshtein, and wuManber edit distance algorithms, very simple approach tended to work well until it needed to scale to thousands (i.e. an interesting dataset).

Check out the source here:

https://github.com/bmahlbrand/queryByHumming