PyTorch is relatively new and PyTorch 1.0 has been released recently in order to compete with TensorFlow in terms of deployment, integration with other frameworks, ....

Besides PyTorch is supported by Facebook and although the community is getting bigger, from what I can see (but i am biased), PyTorch is heavily used mainly in the computer science domain.

Google is addressing many domain including health (radiology, DNA analysis, biology...) so I guess that you will find more open source applications (and more research papers on your topic) using tensorflow out-there.

To conclude and to answer your question about benchmarking:

Usually PyTorch is faster than Tensorflow, but it depends also of what you want to do.

If you are a researcher, what will make the difference is more if you are familiar or not with deep learning theory and python? If in your practice you use python (numpy, scipy or Scikit-Learn - indeed it is really easy to go form numpy to tensors and vice versa) and you have a deep knowledge of deep learning (you want to design your own special loss function for instance), you will prefer PyTorch, otherwise go for tensorflow or even keras.

If you are an engineer with deployment of your solution in the cloud as an objective - and you are concerned about scalability - probably tensorflow is more suitable for you at least now.

Hope that help

Mahboobeh Parsapoor Thanks but all examples that available on internet are generalized , i want to know specific example in biological research where they have performed any benchmark, or there is any specific case in lifescience , so that out of these 2 we can get idea which is good.

PyTorch is relatively new and PyTorch 1.0 has been released recently in order to compete with TensorFlow in terms of deployment, integration with other frameworks, ....

Besides PyTorch is supported by Facebook and although the community is getting bigger, from what I can see (but i am biased), PyTorch is heavily used mainly in the computer science domain.

Google is addressing many domain including health (radiology, DNA analysis, biology...) so I guess that you will find more open source applications (and more research papers on your topic) using tensorflow out-there.

To conclude and to answer your question about benchmarking:

Usually PyTorch is faster than Tensorflow, but it depends also of what you want to do.

If you are a researcher, what will make the difference is more if you are familiar or not with deep learning theory and python? If in your practice you use python (numpy, scipy or Scikit-Learn - indeed it is really easy to go form numpy to tensors and vice versa) and you have a deep knowledge of deep learning (you want to design your own special loss function for instance), you will prefer PyTorch, otherwise go for tensorflow or even keras.

If you are an engineer with deployment of your solution in the cloud as an objective - and you are concerned about scalability - probably tensorflow is more suitable for you at least now.

Hope that help

Fabrice Noreils Thanks a lot.

Both of them are important, you should focus on both of them. When you conduct an experiment, it is good idea to compare between them.

Motaz Al-Hami Thanks

Hi,

PyTorch is better for research and TensorFlow is better for production.

So, as researcher PyTorch is better and shoud use.

Sumit Das Thanks

Take advantage of both for better analysis and comparison.

Tensorflow provides simple ideas of AI and Machine learning in terms of mathematics. I suggest to use tensorflow.

wish you good luck

In this regard, tensorflow is better but depends on your problem. i recommend you to use keras with tensorflow backend, also having no compatibility issues as we face in tensorflow or caffe, theono..

I hope it might help.