I got fed up with listening to charlatans, so now I'm deep diving into AI, particularly deep learning. I'm currently dissecting the fastai videos, the book, and the code, and experimenting to discover the truth about deep learning, learn how to build practical solutions with it, and hopefully monetize a solution.
Having watched the first videos of fast.ai and having verified basic but powerful models on my own computer, I have that matrix-waking-up-blowing-mind feeling, similar to when I discovered Bitcoin: a rare gem hidden in plain sight that defies mainstream ideas. It might be another great opportunity for those who take action instead of just limiting themselves to repeating theory or blindly believing authorities. I'm excited to explore this material that the fast.ai team has released for free.
fast.ai is an organization that aims to make deep learning easier to use through free online courses and software. Their main software is the fastai library, which is built on top of PyTorch and allows programmers to train models quickly and easily, removing the need for PyTorch's more "hairy code". In the videos, co-founder Jeremy Howard also teaches how to apply it to practical problems and, more importantly, the fundamentals behind deep learning, in a very practical way, providing theory only as required. The book is based on the course.
Jeremy Howard affirms that there are myths circulating around deep learning, such as the idea that a lot of math is required to create with it. He also suggests that disinformation is spread by big tech companies, including the notion that one needs vast amounts of data and computing power. The only requirement to use fastai is having some coding experience, as Python is heavily used throughout the course. He encourages being tenacious in order to finish the course, as many students abandon it, and highligths the importance of being an active learner, constantly experimenting rather than just reading and watching videos.
- Organization Website: https://www.fast.ai
- Book: "Deep Learning for Coders with fastai and PyTorch: AI Applications Without a PhD", by Jeremy Howard and Sylvain Gugger. They released both paid and free options.
- Course: Practical Deep Learning for Coders
- Forums: https://forums.fast.ai/
- Docs: https://docs.fast.ai/
Each of my experiments is in a separate folder, starting with 3 digits for quick identification; some of them include various alternatives that I explored. The apps are deployed on Hugging Face Spaces, and you can find the model files (.pkl) either through the corresponding HF Space link or on Kaggle.
001-is-it-a-tree: tree detector. Deployed.002-is-it-a-russian-text: Russian text detector. Deployed.003-light-bulb-type: light bulb type detector. Kaggle.004-bsas-city-landmarks: Buenos Aires City landmarks detector. Kaggle.
I set my previous version of this repository, which contained my own chapter summaries and raw notes from the book and video, to private as a precaution after carefully reviewing the fastbook terms. I respect the work that the fastai team is publishing for free, and I also want to ensure that I do not compromise my entire portfolio repository. This new public repository will only contain my own experiments.
Code licensed under the Apache License 2.0. All third-party images have their own licenses.