(this is using the same code that I used in generating the original figure in my earlier Medium post https://medium.com/@karpathy/a-peek-at-trends-in-machine-learning-ab8a1085a106 … from Apr7, 2017)
Show this thread
Unique mentions of deep learning frameworks in arxiv papers (full text) over time, based on 43K ML papers over last 6 years. So far TF mentioned in 14.3% of all papers, PyTorch 4.7%, Keras 4.0%, Caffe 3.8%, Theano 2.3%, Torch 1.5%, mxnet/chainer/cntk <1%. (cc @fchollet)pic.twitter.com/YOYAvc33iN
-
-
-
(sorry, percentages in original tweet only refer to the last month. looking at the totals so far it's 5.9% of all papers in database mentioned TensorFlow, 5.4% Caffe, 3.2% Theano, 2.3% Keras, 1.6% Torch, 1% PyTorch, 0.5%- for others)
Show this thread
End of conversation
New conversation -
-
-
Great stuff! Is it all of arXiv or just CS?
-
it's based on arxiv-sanity data, so: cs.[CV|CL|LG|AI|NE]/stat.ML
-
Thanks Andrej After
@fchollet tweet the first thing I did was start running your project and trying to get mentions in your search api (nearest neighbor), and I have realized how amazing this work was (vectorized tfidf and svm is as simple as great!)https://github.com/karpathy/arxiv-sanity-preserver …
End of conversation
New conversation -
-
-
Looks like that during the past year
#PyTorch has become a leader in Deep Learning. Amazing!Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
What do you think about the idea that PyTorch should be copyleft? http://keithcu.com/wordpress/?p=3847 … Do you know how many would prefer that?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
2/2018 1-Tf 2-Caffe 3-Keras 4-pyTorch 5-Theano 6- Torch 7-mxnet 8-Cntk/chainer
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
PyTorch is my favorite DL lib and all my clients are starting to adopt it.
-
Hello Shlomo? Are you more inclined to Deep Probabilistic Learning? I ask b/c people who talk Bayesian & Neural Networks often tend to end up thinking a prior can be set on weights between compute nodes- this idea distorts in my mind the tidyness that comes with SGD optimizers
End of conversation
New conversation -
-
-
Great
@karpathy, but I would say that in your statement Keras is not a framework. It is just a high level API. I understand your whole point in the end, but just to make things clear. -
In guess only TF,CNTK,Theano ,caffe count as frameworks,the rest are APIs based on the valid frameworks.
End of conversation
New conversation -
-
-
Pytorch here is #1 in growth rate
-
Good gradient yes
End of conversation
New conversation -
-
-
It will be good to see how many useful applications coming out of these papers,and weather Deeplearning was really needed.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
chainer and pytorch both going up. Tensorflow way more!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Do all of these frameworks support CUDA?
-
yes!
End of conversation
New conversation -
-
-
I wish more papers would provide implementations! At least so claimed "state-of-the-results" should be 100% reproducable.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Please enlighten me. What is the purpose of this chart?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.