Recommended machine learning reading by Antoine Savine
Many financial quants are on a journey to brush up machine learning skills. A vast amount of material of uneven quality is available. Here are the recommendations I wish were made to me when I started my own journey:
- Andrew Ng’s (Stanford) machine learning course on Coursera offers an excellent overview of all the major machine learning algorithms, with cristal clear explanations, exercises and matLab programming assignments. Andrew Ng is a leading expert in the field and the clarity of his explanations makes me somewhat jealous. Coursera’s material takes deliberate shortcuts on maths and must be complemented with deeper mathematical material. The course is feely available. Andrew Ng also offers a (paid) specialization on deep learning, which explores the specific field of neural networks in deep detail.
- Stanford’s CS229 lecture notes are freely available online and nicely complete Coursera’s cursus by discussing and demonstrating all the maths behind the algorithms. The notes are also (mainly) written by Andrew Ng so the concepts and notations are consistent with Coursera, making it a natural complement.
- As a textbook, Bishop’s Pattern Recognition and Machine Learning (get the 2nd edition) is a demanding, rewarding read and the best reference I could find. Machine learning algorithms are described and put in a perspective where the theoretical and mathematical assumptions and guarantees are systematically reviewed and demonstrated. I find it superior in rigor and clarity to the similar and maybe better known Elements of Statistical Learning by Hastie and al. (also an excellent textbook).
My recommendation is: learn machine learning algorithms with the general, mathematically correct material listed above or equivalent before reading about specialized applications in finance, or programming with Python or other languages.
Currently reading Sutton and Barto's book on reinforcement learning (RL):
This is a highly recommended reading, the best resource I could find on the subject and one of the best books I have seen on Machine Learning in general.
The complicated topic of RL is presented in an incremental, pedagogical and natural manner so that the concepts and equations flow naturally and stick in your mind. Somehow, the authors manage to teach RL in a captivating, even addictive manner without sacrificing rigor or completeness, leaving you with a fulfilling feeling of deep understanding of the subject, especially if you try to complete the numerous exercises and programming assignments. Unfortunately, solutions to exercises or assignments are not provided so readers cannot check the correctness of their understanding against the word of god (the authors), although a number of readers posted their own solutions online.