I work on applying machine learning to products. Currently @ Uber | previously at Facebook AI and Yahoo Research
Joined November 2008
- Tweets 57
- Following 311
- Followers 93
- Likes 104
5 Photos and videos
May 7
Tyler Cohen called it years ago and understood LLMs real capabilities better than AI folks. But this is a better exposition with graphs and all.
38
Badri retweeted
22 Jun 2019
when your team gets assigned a "10x engineer"
148
6,697
22,150
Badri retweeted
13 May 2019
No, you put the armor on the part of the dragon with FEWER crossbow holes in it
13
74
554
Badri retweeted
24 Apr 2018
An Outsider’s Tour of Reinforcement Learning, Part 11: The surprising power of PID for model-free RL tasks. argmin.net/2018/04/24/ilc/
3
12
80
23 Feb 2018
Required reading: Follow up to @alirahimi0 NIPS test of time award talk openreview.net/forum?id=rJWF…
1
7
Badri retweeted
11 Nov 2017
A friend sent me this as his snarky answer to a question. Belly laugh.
21
929
2,470
Badri retweeted
20 Feb 2018
An Outsider’s Tour of Reinforcement Learning, Part 6: Policy Gradient Doesn’t Use Gradients. argmin.net/2018/02/20/reinfo…
9
43
156
Badri retweeted
27 Jan 2018
Choosing features: regularization. Choosing an algorithm: regularization. Using float32 instead of float64: regularization. A bug in your code: regularization.
4
23
101
Badri retweeted
14 Feb 2018
Tensor Comprehensions: einstein-notation like language transpiles to CUDA, and autotuned via evolutionary search to maximize perf.
Know nothing about GPU programming? Still write high-performance deep learning.
@PyTorch integration coming in <3 weeks.
research.fb.com/announcing-t…
11
411
1,033
Badri retweeted
28 Sep 2017
grad student: “For successful RL, it is important to carefully tune all hyperparameters, including the random seed.” arxiv.org/abs/1709.06560
4
46
151
12 Dec 2017
Thank you @alirahimi19 for your earnest call to demystify NNs to make real progress, and boldly express what many of us felt! Video: youtube.com/watch?v=ORHFOnaE… - see @beenwrekt clarifications just in case if you were offended! #spellbound x.com/beenwrekt/status/94024…
11 Dec 2017
Some clarifications and ruminations on our NIPS 2017 presentation by me and @alirahimi19 : argmin.net/2017/12/11/alchem…
Badri retweeted
3 Apr 2017
An optimizer’s perspective on a brilliant reinforcement learning paper by Salimans et al (with Roy Frostig) argmin.net/2017/04/03/evolut…
35
78
Badri retweeted
21 Jul 2016
Sculpture of Housing Prices Ripping San Francisco Apart
L: dougmccune.com/blog/2016/07/…
C: news.ycombinator.com/item?id…
4
2
18 Jun 2016
Taming non-convexity using "approximate" convexity... x.com/offconvex/status/72975…
9 May 2016
A framework for analyzing non-convex optimization: offconvex.org/2016/05/08/alm…
Badri retweeted
18 May 2016
Exploring the dual view of backpropagation: argmin.net/2016/05/18/mates-…
1
13
28
18 Jun 2016
Our talk on scaling machine learning at Yahoo is now online youtu.be/l_1S7W_l2cI #SparkSummit #MachineLearning #ParameterServer
2
2