A global team of scientists have developed an #AI system to detect 17 diseases from exhaled breath with 86% accuracy by @spelldotrun #artificialintelligence #machinelearning #datascience #bigdata #analytics #deeplearning #ml #medicine #health #healthcare

Today’s realities: 1. Non-neural net code written more than 5 years ago works just fine 2. Neural net code written a year ago on TensorFlow doesn’t work anymore

hamiltorch A PyTorch-based library for Riemannian Manifold Hamiltonian Monte Carlo (RMHMC). Work with Atılım Güneş Baydin in a Python package that can perform HMC over neural networks. Blog post: adamcobb.github.io/journal/hamilt… Repo: github.com/AdamCobb/hamil… Paper: arxiv.org/abs/1910.06243

"My analysis suggests that researchers are abandoning TensorFlow and flocking to PyTorch in droves." Was TF 2.0 too late? thegradient.pub/state-of-ml-fr…

Using machine learning & quantum physics, GTN LTD is overcoming challenges in drug discovery: pscp.tv/w/cHpH7DFXZ0Vn…

all I see is hummus

Only at a math conference do you hear phrases like, "In a randomly mating population..."

In ML, it's often simultaneously the case that (1) the announcement of a research result is accurate, informative, and measured, and (2) most of the excitement is from people misinterpreting the result as more profound than it really is.

Recalling and explaining the popular internet phenomenon of "The Dress" with the help of some graphical tools buff.ly/2prohRb [source of the gif: buff.ly/2qSxVgH]

That's awkward.... "Production of Brexit 50p coin paused amid exit uncertainty" theguardian.com/politics/2019/…

hbr.org/2019/10/the-ki…

techcrunch.com/2019/10/31/twi…

Have updated website to include recent work. Includes NeurIPS BDL workshop paper on HMC and Robin's (Bin Xin Ru) ICLR paper on BayesOpt. #bdl2019 #ICLR2020 adamcobb.github.io/menu/academic

We have two functions f and g. For any ε>0, some δ>0 satisfies ||g(x) - g(y)|| < δ => ||f(x) - f(y)|| < ε. Does this property have a name? d(x, y) := ||g(x)-g(y)|| is a pseudometric. Is "f is continuous wrt this pseudometric" the best way to say this? (NB g may not be invertible)

Looking forward to the ICML Uncertainty and Robustness in Deep Learning Workshop. Meet Vadera and I will be available at the poster session to discuss our benchmarking toolbox "URSA". This is work with Brian Jalaian and Ben Marlin (UMass). arxiv.org/abs/2007.04466

New work on applying HMC to BNNs! Our symmetric splitting approach preserves the entire Hamiltonian and can be run on a single GPU. All the code is in hamiltorch :) Paper: arxiv.org/pdf/2010.06772… Blog post: adamcobb.github.io/journal/bnn.ht… Work with Brian Jalaian!

Lovely to have all these new followers .. can we just check that you know that you’re following The World Wide Robin Hood Society in Nottingham and not the Robin Hood App .. if so .. a big welcome from Sherwood 🙌

Interested in watching the mean paper review scores rise while reviewers ignore your rebuttals for NeurIPS Dataset and Benchmarks? Check out this colab which connects to OpenReview's API: github.com/ikiskin/OpenAP… Highlights: Mean before rebuttal: 5.5 Mean after rebuttals: 5.8