Recurse Center - Batch 2 - Cycle 20240922-20240924 - Reunion

Reunion

I'm back at Recurse Center :)

My first day at Recurse Center is caught in the middle of my cycle format, so this is a bit of a shorter first post.

So, to start, to those just arriving, here's a little about me and what I'm planning on doing at Recurse this time around (cross-posted in the Zulip #welcome channel)

Hi everyone! My name is Johann :) This is my SECOND batch at RC and I am beyond thrilled to be returning this fall/winter :)

I'm an artist, engineer, and educator based in NYC (so I'll be coming to the Hub most of the time). I'm focused on topics like machine listening, computational audio, and programmatic sound. I like to use technology for creative and artistic means, which sometimes looks like: making musical instruments like the Harvester, building environmental sound classifiers like Whisp, and writing software to generate multi-channel sound installations from large speech datasets. A bit more about me and my last batch was written about on RC's blog.

Right now I’m currently halfway through a two-year fellowship at the Social Science Research Council, called Just Tech.

During my time as a fellow I’ve been researching the origins of bias in automatic speech recognition (ASR) systems used widely in voice interface technology. In the pursuit to dive deep and push myself as a programmer, I wanted to return to RC and do another batch in order to dedicated time and space to learn how this technology works, through building an open-source ASR system from the ground up. Returning to RC for a batch feels like the perfect place to dedicate myself to the work of programming this system, alongside a community of talented and curious programmers with whom I can be in supportive community with.

My goals for this next batch include developing an open-source automatic speech recognition system, trained with public datasets on RC’s GPU cluster Heap. In the process of self-teaching, I plan on putting together a self-study guide on building these types of systems. I’d also be doing associated research on the history of speech technologies to help contextualize the technology I’m building.

Alongside these goals, I will be preparing for the job market post-fellowship, and will be spending time preparing to move into an AI/ML engineering career path. This will involve studying for technical interviews and learning fundamental concepts needed to pursue a career in this field.

I've been pretty active on Zulip in the past, so I plan on documenting this work regularly on Zulip and my blog (you are here!).

Really excited to meet all of you next week!!! Never graduate (I certainly didn't hah)

My to-do for my follow cycles:

Daily Ambition:

- 2 hours: DS+A

- 4 hours: ASR Deep work

- 2 hours: Self-Study Guide

Tasks:

- Get set up on Heap, RC's GPU cluster

- Download some datasets

- Put together a plan for DS&A study, ASR buildout, Self-Study Guide, and ML/AI professional development

Some things that I feel like will guide my self-study this fall / winter:

- https://medium.com/bitgrit-data-science-publication/a-roadmap-to-learn-ai-in-2024-cc30c6aa6e16

- https://github.com/iconix/openai/blob/master/syllabus.md

- Ilya’s list

- https://huyenchip.com/ml-interviews-book/