At the beginning of September 2018, I made a plan to spend the next 25 weeks studying deep learning. This page gives updates on my progress.
I have finished my first month more-or-less according to schedule. You can read my one month review here
Below I keep track of what I worked on, week-by-week.
I spent this week reviewing prerequisites. To brush up on machine learning, I completed Andrew Ng’s course (I only had 4 weeks of material left). I skimmed part 1 of Ian Goodfellows book for a more technical review. Up until this point I find his book too theoretical, parts like the “no free lunch theorem” could have easily been skipped. For python syntax, I watched some lectures on Udemy, and completed the associated exercises.
I finished with my plan ahead of schedule, watching the fast.ai videos on single label image classification. I coded up some of my own experiments, such as a Jupyter notebook that downloads pictures of Obama and Trump and builds an image classifier based on them.
Completed much of the fast.ai course. Created a Harry Potter image classifier. Coded up toy drafts on structured learning, recommender systems, and nlp using different datasets than in the course (didn’t optimize for good results though). Through them I familiarized myself with fast.ai and pytorch code.
I watched the first two parts of the Coursera deep learning lecture by Andrew Ng. Completed the first two weeks of material for Siraj Raval’s School of AI course. Slightly behind completing part 1 of fast.ai.
I finished part 1 of the fast.ai course, and watched the Coursera DL lectures up to part 5. Completed week 3 of the Move37 RL course.
I also started working on the Inclusive Images competition on Kaggle. Thanks to google, all participants received $1000 worth of computing credits on google cloud. I set up my google cloud machine and downloaded the images (which, taking more than 500 GB, needed more than a day to download).
This week I started blitzing through Coursera’s deep learning specialization. I finished the first four out of five parts/modules. Each part is supposed to take a few weeks; however, I have already watched most of the lectures, and the programming assignments are extremely easy. I will finish early next week.
I finished the Coursera DL specialization early this week. I found the course great, I’ll write up a detailed review soon, along with some highlighted tips and tricks. I also watched lectures 9-10 of fast.ai, but didn’t give it any serious effort. Other than that, I started programming a language detection task, and gave my website a new design.
I didn’t achieve much on the learning front this week. I watched the fastai part 2 lectures (11 to 14), and completed the midterm for the RL course. The rest of the time I spent doing administrative tasks, such as updating and formatting my resume. I also took some time off for personal reasons (relatives visiting).
These two weeks I was taking time off from my projects (due to family visit and a trip to China). Plenty to catch up on.
Created a language detection model that can classify the 21 languages spoken in the EU.
Another week off from my projects. Plenty more to catch up on.