Today, Ekin Dogus Cubuk and I are excited to introduce Periodic Labs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is

Just registered and got the tickets for #NIPS2013, looking forward to meeting everyone!

I don't do graphical models. The cliques, DAGs, message passing, expectation-maximization, and belief propagation remind me of high school.

New blog post: #NIPS2013 reading list atpassos.me/post/675608315…

Optimization and Learning in Factory (scala) people.cs.umass.edu/~luke/nips-ws-… by Alexandre Passos

Congratulations to Dan Stubbs, *winner* of the Computing Research Association's Outstanding Undergraduate Researcher Award (Male) 2014!

I'm moving to SF and joining google in January 6th!

Top 5 Predictions for 2014: 5. + 4. -35.4 3. (0.12, 0.54, 0.34) 2. cow 1. discovery of the doge neuron (wow!)

Really cool ICLR submission: arxiv.org/pdf/1312.6199v… ; (1) it's as easy to interpret linear combinations of units as units in a deep net

(re arxiv.org/pdf/1312.6199v… ) (2) it's easy to generate adversarial examples which generalize across nets and look like normal data

@eat_umami I've never had either flank or venison that tender, it was awesome. How did you do it? Sous vide? Marinade?

@eat_umami Thanks! I'll give it a try when I get the chance.

PhD students: Find low hanging fruit, hurl it to the ground, and crush it. That’s how you make wine. #mixedmetaphors #badadvice Neil Lawrence

Stochastic Gradient Methods 2014 workshop nice summary writeup from Yaroslav Bulatov yaroslavvb.blogspot.com/2014/03/stocha…

New paper out: arxiv.org/abs/1403.1349 . Not only is dual decomposition with soft constraints easier but you can also learn the constraints

I finally made a point to learn how Johnson-Lindenstrauss Lemma projections work and wrote up a quick description: bit.ly/1iUHp23

This ML algorithm is rated R. It contains: Strong Assumptions, Frequent Sampling, and Matrix Inversions. Restricted to masochistic users.

@emmatosch I want summaries, or at least links :-)

ICYMI: The business models being adopted by the new crop of deep learning startups, and the challenges of each. npbay.es/deep-learning-…

"technical debt" of ML systems in the real world. Very interesting read from Google. (glad I am in academia ;) ) static.googleusercontent.com/media/research…