PhD in business analytics. Ensuring that Deep Learning (AI) systems are manageable and responsible at scale, while "wandering in latent space".
Boulder, Colorado USA
Joined July 2007
- Tweets5.7K
- Following470
- Followers4.7K
- Likes821
Richard Hackathorn retweeted
Simple (but fascinating) question! Obvious answer is your (cosine-sim) app for your vector database. Harder answer is how you understand structural relationships (manifolds?) among object embeddings, especially within local neighborhoods. Smashing 1000s dim into 2? 👎
Richard Hackathorn retweeted
3 open-source frameworks for SSL:
1. s3prl, SSL for Speech
2. SimCLR, SSL as contrastive learning
3. Lightly, SSL on images
Richard Hackathorn retweeted
Only 4 more days until my new book, Machine Learning with PyTorch and Scikit-Learn, is going to be released! Packt invited me on to their Expert Insight series for a live launch event on Wednesday morning. Please feel free to join and AMA about the book 😊linkedin.com/video/event/urn…
Richard Hackathorn retweeted
Rebellious ML students who took time to study causal inference might wish to look into this company causalens.ai that is looking to hire research scientists to lead their expanding operations.
Thrilled to announce our latest $45m funding round!
Thank you so much to everyone who made this possible including @DoriltonC, @MoltenVentures, @GPBullhound, @IQ_Capital_Fund, @GenVenVC, @fomcap and all others!
For more information see:
lnkd.in/dRH8Fst2
Richard Hackathorn retweeted
Interviewed on episode 6 of The Gradient podcast.
On episode 6 of The Gradient podcast, we are excited to have interviewed Deep Learning pioneer @ylecun!🎙️🔊
We touch on his start on #AI research and some exciting recent work on Self-Supervised learning for vision from @facebookai.
Take a listen:
thegradientpub.substack.com/…
Richard Hackathorn retweeted
Because deep learning is so empirical, success in it is to a large extent proportional to raw experimental throughput - the ability to babysit a large number of experiments at once, staring at plots and tweaking/re-launching what works. This is necessary, but not sufficient.
Richard Hackathorn retweeted
Just reading through the recent "[A] Survey on Neural Network Interpretability" (arxiv.org/abs/2012.14261) and there is this really cool interactive figure summarizing the respective literature in a 3D scatterplot (link to the interactive version: yzhang-gh.github.io/tmp-data……)
Richard Hackathorn retweeted
Room-scale Puzzle 'Eye of the Temple' Aiming for Spring 2021 Release on PC VR Headsets ow.ly/zwMq50CZuEM @runevision
Richard Hackathorn retweeted
Geoff Hinton says that deep networks are different from symbolic AI because they reason by analogy. Now we know how: arxiv.org/abs/2012.00152
Richard Hackathorn retweeted
New levels of creativity are required for leaders to correctly develop business use cases for analytics – I agree with @hackathorn in his latest article for @EckersonGroup: ow.ly/VOlU50x3kfe #ROI #actiondistance
Richard Hackathorn retweeted
I wonder how many programmers even know what a "Stack Overflow" is? Yeh, its a website that took their name from microprocessor instructions and functions... I miss assembler code sometimes.
Richard Hackathorn retweeted
Neural nets are the core machinery that make deep learning so powerful. This radical new design scraps the layers entirely to overcome a major shortcoming in AI. trib.al/ROUK6SX
