Faster PyTorch Training by Reducing Peak Memory. Update weights immediately after computing the gradients, instead of waiting for the entire backward pass to finish. This can significantly reduce peak memory usage and speed up training. Fuse backwards and optimizer steps into one operation.

Tim Dettmers's Which GPU(s) to Get for Deep Learning. Not sure how up-to-date this is for this year.

Optimizing models using the PyTorch JIT, likely out of date

Maxwell Forbes, The PhD Metagame. Gamification of attention slowly bleeds into the ML research.

Google deep learning tuning Playbook -- maximize the performance of deep learning models. Read ~50% of this.

Umar Jamil's machine learning videos. Deep, formal and intuitive explanations with practical implementations.

Michael Nielsen's Long-Term Memory - Augmenting Cognition. Didn't read yet.

Processing video frames can be slow. This blog helped me to optimize some parts of the TensorPix video inference engine.

Jordi Cenzano's Frame accurate video client in the browser. This was a super useful resource when we tried to figure out how to build backend-based video player in TensorPix. The idea was ditched but I would still use this approach.

Evan Hahn's "Scripts I wrote that I use all the time"

Dr. PI - Kompleksna analiza. Didn't read yet.

Art of Problem Solving's (Richard Rusczyk) Curriculum Map

A distributed lock pattern with Redis. This was quite relevant for me when I was implementing per-user distributed queueing in TensorPix.

Julia Evans's blog. Goldmine for everything. Thanks Šaja.

Jeronim Matijević's blog. Biggest CLIPpers fan I know

Marijan Smetko's blog

21 Facts About Throwing Good Parties. Some of these scream 'USA culture' and 'millennial' but relevant nonetheless.