Quora Feeds
Active Member
François Chollet
I think the best way to study machine learning goes something like this:
Overall, it is much easier to grok the theoretical side of machine learning when you are already familiar with its practice, rather than the other way around. A sound understanding comes from confrontation with real-world problems, not confrontation with MOOCs and books.
See Questions On Quora
Continue reading...
I think the best way to study machine learning goes something like this:
- At first, to get a precise understanding of how key algorithms work, try reimplementing yourself toy examples in Numpy (a Numpy convnet, a Numpy multi-layer perceptron, a Numpy LSTM).
- To get familiar with practical applications, look at the Keras examples provided with the Keras repo. Try modifying them, adapt them to new data, and tune the model architectures until you get the best result you can on your problem of choice.
- Get a feel for research and real-world data science applications by entering Kaggle competitions. Team up with other people, maybe win a competition!
- At last, you can start reading theoretical books (e.g. the deep learning book from Goodfellow, Bengio and Courville) and research papers to develop a more abstract and mathematical understanding of the processes your have been practicing.
Overall, it is much easier to grok the theoretical side of machine learning when you are already familiar with its practice, rather than the other way around. A sound understanding comes from confrontation with real-world problems, not confrontation with MOOCs and books.
See Questions On Quora
Continue reading...
