Nikhil (@wiznick) profile picture

Data Science and ML leader, enthusiast, travel hacker. Moving energy forward through #AI #MachineLearning #DataScience #DataProducts

San Francisco Bay Area, CA
Joined May 2009
Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)Photo by Nikhil (@wiznick)
As an #Eagles fan who lived in Philly, I wish I was in #philly today and not in SF. #eagles fly high.
1
1
122
It’s all about building the right teams. Excited to share my perspective and learnings during a panel at DataOps. #Data #organizationalstrategy #Leadership
Almost time! @emilieschario @sarahmk125 Nikhil Gulati and Disha Ahuja join us for a #DataOps panel conversation on strategically composing your data team. Extremely important as we scale #bigdata priorities. See their virtual panel dataopsunleashed.com Feb 2nd at 12:40PM PST
Faster tokenization. Something to trie for everyone!
Preparing text for processing by an #ML model often involves tokenization, in which text is split into smaller units (e.g., words or word segments). Today we present a new approach that speeds up the process by up to 8x compared to standard methods. goo.gle/3dKHhVi
9
68
2
624
👇Our probabilistic lifted inference book is out! 🙏 to everyone who contributed ! #Symmetries play an important role not only in physics but also in #AI. Strong connections to #geometricdeeplearning preview 👉 tinyurl.com/46a3ee4b @mitpress url 👉mitpress.mit.edu/books/intro…
7
75
4
293
Transformer architecture reality check. Most modifications do not lead to improved performance, unless the # of parameters is increased dramatically. One exception is decoupling input and output embedding parameters. h/t @seb_ruder arxiv.org/abs/2102.11972
3
42
1
203
To train PyTorch models with 2x the parameters via sharded, watch the 4-minute video on what it takes to refactor your model into the Lightning format. youtube.com/watch?v=uHMG2Xng…
2
17
94
No place in the world hit >800 cases/million in the course of the global pandemic. Until now. North Dakota is at 920 And other states are not far behind. Czech Republic is at 800, highest in Europe.
40
281
74
490
Excited to speak at the #ThinkTech Virtual #ML meet up and learn more about the digital awareness initiative in the region. #MachineLearing #DigitalTransformation #AI
The importance of Machine Learning spans across our present & our future ترتبط تقنية تعلم الآلة بحاضرنا ومستقبلنا انضم إلينا في اللقاء القادم لتعرف أكثر عن هذه التقنية ⏰الأربعاء، 7:00م سجل الآن: @achakhari1 @wiznick @igorholas #THINKTECH #Digital_Summer

Very excited to introduce Humans of AI: Stories, Not Stats! In this series, I interview AI researchers to get to know them better as people. Starting next week, I will release two interviews every week as videos and podcast episodes. (Link 👇)
27
289
33
1,792
Some asked me about concept drift so here you go. A predictive ML model learns theta to output P(Y|X; theta). Data drift is when P(X) changes: different data distributions, different feature space. Ex: service launched in a new country, expected features becoming NaNs. 1/5
12
161
14
857
Do we need multiple heads and wider FFN layers in Transformers? No. Our work, DeLighT, effectively replaces these components with single-headed attention and light-weight FFN layers. Joint work with @gh_marjan, @sriniiyer88, @LukeZettlemoyer, and @HannaHajishirzi
3
112
12
511
The transformer architecture of GPT upper bounds its ability at memorization. It cannot learn many algorithms due to the functional form of its forward pass, and spends a fixed compute per token - i.e. it can't "think for a while". Progress here critical, likely but non-trivial.
32
229
13
1,537
The GPT-3 hype is way too much. It’s impressive (thanks for the nice compliments!) but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We have a lot still to figure out.
130
724
173
6,418
Today we describe a #MachineLearning approach combining #NaturalLanguageProcessing with #ComputerVision to automatically extract data from structured documents—invoices, receipts, etc.—with the potential to streamline many business workflows. Learn more at goo.gle/2B0kcOs
15
398
39
1,350
Providing COVID-19 data to the world, @JohnsHopkins Prof. Lauren Gardner is the creative powerhouse behind the COVID CSSE dashboard. It has been used by natl/interntl governments and no-gov. organizations, to set policy and goals. That's impact most academics can only dream of
30
388
43
1,780
As data scientists, we (with @math_rachel) have done our best to look at the data around covid-19, and what it means to you and your community. Our view: it is appropriate to be very concerned, and significantly change your lifestyle, right now. 1/13 fast.ai/2020/03/09/coronavir…
113
3,087
555
5,748
After 2+ years in stealth, we’re excited to launch today! covariant.ai Thank you to our team, customers, partners and investors, we couldn’t have done it w/o your support and trust. Exciting milestone, even more exciting journey ahead! medium.com/covariant-ai/brin…
16
115
41
667
I'm sharing some of my top book recommendations from 2019, along with my ranking 1-10. Some of them are new, others I have read multiple times over the years. medium.com/@vkhosla/2019-boo…
15
90
10
510
Here is a great explanation of how to combine @huggingface Transformers and fastai to get great results from your NLP models towardsdatascience.com/fasta…
5
197
2
710
You can now build and run ML pipelines in an interactive Colab notebook using TensorFlow Extended (TFX)! 📓 Develop in a notebook and then export to a production-ready ML pipeline. Try it in your browser today → goo.gle/TFXnotebook
6
186
12
616