Blaz Skrlj retweeted
The ChatGPT API was released yesterday and it costs 90% less than expected. Here’s five methods (and resources to learn about them) that are **probably** being used to enable this price reduction… 🧵[1/6]
Really nicely done!
github.com/dionhaefner/pyhpc…
Blaz Skrlj retweeted
Hippocampal spatial representations exhibit a hyperbolic geometry that expands with experience | Nature Neuroscience nature.com/articles/s41593-0…
Blaz Skrlj retweeted
New paper: Equivariant MuZero
arxiv.org/abs/2302.04798
A version of MuZero that incorporates environment symmetries by design.
Great internship work by @andreeadeac22 👏👏👏
Blaz Skrlj retweeted
We've made so much progress in the last 7-8 years on complex domains like text and images and basically none on simple sequences of numbers (i.e. time-series forecasting).
In fact, I think we've regressed since good information is so hard to find.
Blaz Skrlj retweeted
One day I'll get to studying Normalizing Flows. When I do, I'll start here github.com/janosh/awesome-no…
Blaz Skrlj retweeted
We've just released the first version of our Deep Learning Tuning Playbook! This is our attempt to distill our process for actually getting good results with deep learning. We emphasize hyperparameter tuning since it has been a large pain point. github.com/google-research/t…
Blaz Skrlj retweeted
TheAlgorithms / Rust: All Algorithms implemented in Rust ★13080 github.com/TheAlgorithms/Rus…
Blaz Skrlj retweeted
How to train very deep NNs without shortcuts, but still achieve competitive results on ImageNet?
Our ICLR paper gives a simple solution derived from kernel approx theory. We hope this could enable further research into deep models.
openreview.net/pdf?id=U0k7XN…
Blaz Skrlj retweeted
animation work in progress🤓 - went from sphere to actual 3D Whole-Cell model #unity
Blaz Skrlj retweeted
Today I saw the impact that AlphaFold is having on speeding up drug discovery firsthand:
Blaz Skrlj retweeted
Here's a few I like for different reasons (not exhaustive):
Careful perturbations:
aclanthology.org/P18-1079.pd…
aclanthology.org/2020.acl-ma…
Cool ML modeling:
arxiv.org/pdf/1908.07125.pdf
arxiv.org/pdf/1909.00986.pdf
Data collection:
arxiv.org/pdf/2104.14337.pdf
arxiv.org/pdf/2110.08514.pdf
Blaz Skrlj retweeted
Tonga's Hunga Tonga volcano just had one of the most violent volcano eruptions ever captured on satellite.
Blaz Skrlj retweeted
GradMax: Growing Neural Networks using Gradient Information
abs: arxiv.org/abs/2201.05125
a new approach for growing neural networks that aims to maximize the gradient norm when growing
Blaz Skrlj retweeted
ProBiS-Dock Database: A Web Server and Interactive Web Repository of #Small_Ligand_Protein Binding Sites for #DrugDesign
pubs.acs.org/doi/10.1021/acs…
@koncj @SkBlaz
#current_issue #Bioinformatics
Blaz Skrlj retweeted
Stanford's ~entire AI Department has just released a 200 page 100 author Neural Scaling Laws Manifesto.
They're pivoting to positioning themselves as #1 at academic ML Scaling (e.g. GPT-4) research.
"On the Opportunities and Risks of Foundation Models"
arxiv.org/abs/2108.07258
Blaz Skrlj retweeted
Today with @emblebi, we're launching the #AlphaFold Protein Structure Database, which offers the most complete and accurate picture of the human proteome, doubling humanity’s accumulated knowledge of high-accuracy human protein structures - for free: dpmd.ai/alphafolddb 1/