TensorFlow:
Pros: Easier to read and use for beginners, can use Google’s powerful machine learning technology, great for larger projects or deep learning
Cons: Not as flexible for modeling, can be memory intensive
Keras:
Pros: Easy to use for beginners, great for quickly building and training models, supports multiple backends like TensorFlow and Theano
Cons: Not as customisable for building complex models, not as fast as other options.
MXNet:
Pros: Offers great scalability,has efficient memory usage, very fast with GPU learning
Cons: Not as popular as other frameworks, can be harder to learn.