Algomancer

This is very well written.

I hope work diffusion models develops such that the sample efficiency is improved. There is something really satisfying about how progressive sampling works in these models.

Will do, in hindsight that is a super obvious thing I didn't think about when trying to get this quickly out.

I've been enjoying learning model based image enhancement, decided to wrap it up as a little service.

This works by predicting the parameters to a pixel wise polynomial over r,g,b,r * g,r * b,g * b,r^2,g^2,b^2,r * g * b.

Trying to build a habit of pushing projects live rather than just finishing a neat algorithm and letting it gather dust.

There is a url to a demo store in the github about section.

Genuine Question, and not meant to be snarky. Who is getting value from a post like this? Considering it is on the front page.

I would like to second this.

If I didn't know better, these could be my words. Extremely relatable.

I'd love an community of people fighting against this feeling, especially one that isn't filled with 'gurus'. I am not sure that can exist.

This is great,

With personalization, can I see a preview before making a purchase?

Brilliant potential home brew for a DND campaign.

#0000FF has really seen a resurgence as of late, it is brilliant to see this medium executed by a master.

However, where does one go from here? Does it become old hat? I hear some Avant-garde types are beginning to explore post-blue ideas. #333333 has a romantic bleakness to it that I would love to see explored.

Clever, you could use branches to add sessions.

  Looking for cool problems, recently I have been doing a lot of deep learning stuff - but am just interested in hard engineering problems. 

  Location: Brisbane, Australia
  Remote:Yes 
  Willing to relocate: Maybe
  Technologies: Python (Web and Scientific), Pytorch, C++, JS/React, Linux

Designed architecture for automatic music mastering. Designed and Developed state of the art speech synthesis technology using deep learning including novel phase estimation approaches. Designed and Developed sequence generation models, worked with flow based likelihood models. Developed various real time and faster than real time audio pipelines with C, C++ and Python. Assisted in developing developer workflows for large scale machine learning training on AWS and GCP. Optimized performance of machine learning models for production by rewriting specific components in high performance C++ and weight pruning and porting models a set of sparse and quantized matrix operations for massive performance increases. Developed a real time audio engine for both Android and IOS.

I also have some experience with computer vision systems, recommendation systems and fraud detection. I have been coding since I was a kid, and can hack on most of the stack.

  Résumé/CV: On request.
  Email: [email protected]

This is pretty interesting, what would be the best code base to dig through to see it in action?

At the end of the month I am going to be in silicon valley touring facebook, google, twitter & 500 startups and a few other places - I am very Keen to meetup for beer and just generally network with anyone building cool stuff while I am in the valley.

Anyone down?