@datitran

Partner & CTO @DATANOMIQ. Ex-Co-Founder/CTO @priceloopai. Ex-Head of AI @axelspringerai, Ex-@idealo_de, Ex-@Pivotal. Tweets and opinions are my own!

 Berlin
dat-tran.com/
Joined October 2013
  • Tweets1.2K
  • Following828
  • Followers2.4K
307Media
Dat Tran retweeted
Chip Huyen
@chipro24 Jun 2021
Takeaway from @karpathy's CVPR talk: The most successful ML projects in prod (Tesla, iPhone, Amazon drones, Zipline) are where you own the entire stack. They iterate not just ML algorithms but also: - how to collect/label data - infrastructure - hardware ML models run on
19
253
27
1,739
I’ve recently struggled to understand Python concurrency: - How threads & processes work on your OS vs Python - The role of CPUs, hardware & the GIL - Understanding beyond rules of thumb After tons of research, I wrote this w/the answers plus more 👇🧵 python.hamel.dev/concurrency…

Python Concurrency: The Tricky Bits

An exploration of threads, processes, and coroutines in Python, with interesting examples that illuminate the differences between each.

python.hamel.dev
20
378
16
1,723
🚨Transformers is expanding to Speech!🚨 🤗Transformers v4.3.0 is out and we are excited to welcome @facebookai's Wav2Vec2 as the first Automatic Speech Recognition model to our library! 👉Now, you can transcribe your audio files directly on the hub: huggingface.co/facebook/wav2…
Taken from the official Wav2Vec2 paper: https://arxiv.org/abs/2006.11477

ALT Taken from the official Wav2Vec2 paper: https://arxiv.org/abs/2006.11477

17
303
21
1,349
Dat Tran retweeted
Chip Huyen
@chipro28 Dec 2020
Based on conversations with a dozen companies doing real-time ML in both US and China, I wrote about online inference and online learning -- their use cases, solutions, and challenges. Machine learning is going real-time, and most companies aren't ready. huyenchip.com/2020/12/27/rea…

Machine learning is going real-time

Real-time machine learning is the approach of using real-time data to generate more accurate predictions and adapt models to changing environments. There are two levels of real-time machine learning....

huyenchip.com
31
349
30
1,472
Dat Tran retweeted
Chip Huyen
@chipro27 Oct 2020
I love teaching and I’ve always wanted to come back to Stanford to teach. After 2 years with the help of many people, it's happening! I’ll be teaching Machine Learning Systems Design in Jan '21. It'll be online. Syllabus is below. Feedback appreciated! huyenchip.com/2020/10/27/ml-…

Course announcement - Machine Learning Systems Design at Stanford!

This project-based course covers the iterative process for designing, developing, and deploying machine learning systems. It focuses on systems that require massive datasets and compute resources,...

huyenchip.com
65
232
15
2,318
Our API now includes a brand new pipeline: zero-shot text classification This feature lets you classify sequences into the specified class names out-of-the-box w/o any additional training in a few lines of code! 🚀 Try it out (and share screenshots 📷): huggingface.co/facebook/bart…
Image of web widget. Text to classify: "I spent a year studying abroad in Paris with my friends during college. We didn't learn much French, but it was the most fun I've ever had." Possible class names: happy, sad, proud, nostalgic, angry Model predicts nostalgic with a score of .678

ALT Image of web widget. Text to classify: "I spent a year studying abroad in Paris with my friends during college. We didn't learn much French, but it was the most fun I've ever had." Possible class names: happy, sad, proud, nostalgic, angry Model predicts nostalgic with a score of .678

Image of web widget. Text to classify: "I have a problem with my iphone that needs to be resolved asap!!" Possible class names: urgent, not urgent, phone, tablet, computer Allow multiple true classes? True. Model predicts urgent and phone and does not predict computer, not urgent, or tablet

ALT Image of web widget. Text to classify: "I have a problem with my iphone that needs to be resolved asap!!" Possible class names: urgent, not urgent, phone, tablet, computer Allow multiple true classes? True. Model predicts urgent and phone and does not predict computer, not urgent, or tablet

Image of web widget. Text to classify if a snippet from a news article about a new model developed about Galilean satellites Possible class names: space & cosmos, scientific discovery, microbiology, robots, archeology Model predicts scientific discovery and space & cosmos and does not predict archeology, robots, or microbiology.

ALT Image of web widget. Text to classify if a snippet from a news article about a new model developed about Galilean satellites Possible class names: space & cosmos, scientific discovery, microbiology, robots, archeology Model predicts scientific discovery and space & cosmos and does not predict archeology, robots, or microbiology.

Image of web widget. Text to classify is the opening line from Anna Karenina, written in Russian. Possible class names: English, Russian, Arabic, Urdu, Mongolian Model predicts Russian with a score of .775

ALT Image of web widget. Text to classify is the opening line from Anna Karenina, written in Russian. Possible class names: English, Russian, Arabic, Urdu, Mongolian Model predicts Russian with a score of .775

13
102
13
430
Check out Performer, a generalized attention framework based on the Transformer architecture, which implements the novel FAVOR+ algorithm to provide linearly scalable, low-variance and unbiased estimation of attention mechanisms. Read more at goo.gle/3mhiCu4
5
290
33
1,176
Just learned about this "network dissection framework" to understand CNN predictions, i.e., individual, single units being mainly responsible for recognizing particular objects. Also, that's an impressive info-per-space maximization in their figures :P. arxiv.org/abs/2009.05041
5
38
1
225
Dat Tran retweeted
comma
@comma_ai24 Sep 2020
Laneless is getting better! Watch #openpilot drive on a dirt road with no lane lines. Source: youtu.be/Re6O9cve9NU
12
37
9
302
RigL is a new algorithm for training sparse neural networks. Instead of pruning a pre-existing dense network, it dynamically builds one during training without sacrificing accuracy relative to traditional approaches. Learn how it’s done at goo.gle/2ZJryiU
15
351
29
1,153
Updated GPU recommendations for the new Ampere RTX 30 series are live! Performance benchmarks, architecture details, Q&A of frequently asked questions, and detailed explanations of how GPUs and Tensor Cores work for those that want to learn more: timdettmers.com/2020/09/07/w…
31
241
34
860
Zum Mäusemelken. Echt. Wir gucken seit Jahren dabei zu wie zahlreiche Länder die Digitalisierung als Chance begreifen. Und wir? Sehr gemütliche Beobachter.
Dietmar Neuerer@dneuerer7 Sep 2020
Studie zur digitalen #Wettbewerbsfähigkeit: Deutschland fällt um 52 Ränge zurück. handelsblatt.com/politik/int…

Digitalisierung: Frankreich läuft Deutschland den Rang ab

Nach einer neuen Studie verliert Deutschland massiv an digitaler Wettbewerbsfähigkeit. Frankreich holt dagegen dank Präsident Macron mächtig auf.

handelsblatt.com
8
12
1
79
Exciting results from our partnership with @googlemaps! Using advanced Graph Neural Networks we were able to improve the accuracy of ETAs in major cities across the world by up to 50% bit.ly/2Z0IMYC
24
295
63
1,527
Replying to @karpathy
@karpathy, love your minGPT demo! I removed about 100+ lines of boilerplate by adapting it to use @PyTorchLightnin . also gave you half precision, multi gpu and tpu training for free ;) github.com/karpathy/minGPT/p… Here's the demo: youtube.com/watch?v=2aJFRQ-v…

Adapted this awesome minGPT to use PyTorch lightning! by williamFalcon · Pull Request #14 ·...

Fixes #13 Fixes #11 Fixes #12 Fixes #8 Fixes #7 Huge fan, and amazing code to teach GPT! I think the code can be more readable and easier to digest and understand what's happening by removi...

github.com
4
40
2
282
Dat Tran retweeted
Justin Johnson
@jcjohnss10 Aug 2020
Videos for my Fall 2019 course "Deep Learning for Computer Vision" are now on YouTube! This is an evolution of @cs231n that I used to teach at Stanford: - All content refreshed - New topics: Transformers, Video, 3D, etc - HW in @PyTorch + @GoogleColab youtube.com/playlist?list=PL…

Deep Learning for Computer Vision

Computer Vision has become ubiquitous in our society, with applications in search, image understanding, apps, mapping, medicine, drones, and self-driving car...

youtube.com
49
707
39
2,917
Have you ever wondered what is so special about convolution? In a new blog post, I show how to derive #convolution from translational symmetry principles: medium.com/p/deriving-convol… This is key to extending #DeepLearning to #graphs
12
310
23
1,269
GPT-3 Does The Work™️ on generating SVG charts, with a quick web app I built with @billyjeanbillyj. With a short sentence describing what you want to plot, its able to generate charts with titles, labels and legends from about a dozen primed examples. cc @gdb
18
164
65
712
Machine Learning Summer School 2020 #MLSS2020 streaming live Monday 12:30 CEST on youtube.com/watch?v=8staJlMb… @MLSS_Tuebingen @MPI_IS @ELLISforEurope
3
241
15
704
Interested in learning how to build OCR models? Here's a great code example from @A_K_Nain showing a model that can break Captchas. keras.io/examples/vision/cap…
9
183
13
816
Dat Tran retweeted
Google AI
@GoogleAI8 Jun 2020
Recent updates to the technology behind Google Translate—including changes to model architecture, training, treatment of noise in datasets, and multilingual transfer— have led to improved translation quality for over 100 languages. Learn more at goo.gle/3dKEWs5.
19
330
37
1,306
Load more