I want to AI to help with literature review for scientific research. Between PyTorch or TensorFlow or something else, how can I know what is right for me?
I am hoping to do things like literature searches. I also want to use it to help gain clarification on scientific disputes. I would like to be able to load a key science paper into the knowledge base and then search the papers that cite the key paper to find which papers affirm the conclusions of the key paper or build off the conclusions and which papers dispute the conclusions.
I am not even sure if TensorFlow and PyTorch are different applications or different datasets. What exactly is a dataset? Does a dataset include a bunch of science papers? How do I know what data is included in a dataset? I’m trying to take the Beginner’s Course but basic terminology is not being defined.
Hi Ron, nice to meet you. I’m in a similar situation. I still am not sure, which framework I should learn and I currently don’t have the time to just learn them both. After doing some research for my NLP and chatbot project, I decided to start with PyTorch first. If you like to see a short comparison of the two frameworks, there is a YouTube video from Patrick Loeber that gives an overview. I will go with PyTorch from here, but always keeping an eye on TensorFlow too. I heard that TensorFlow might still have some advantages in regard to a later deployment, but the recent usage numbers seem to show a strong tendency towards PyTorch. Another important point for me is, that if you consult the documentation for the transformers on the HuggingFace website, then you notice that the support for PyTorch is much wider than for TensorFlow.
I found a blog post from Ryan O’Connor who is with AssemblyAI about PyTorch vs TensorFlow which made me pretty confident with my choice of going with PyTorch. You might want to check it out.