about YSDA

In this year I ended 2-year data science program in Yandex School of Data Analysis (ШАД) in Moscow. Here I want to make a short overview of all courses which I've studied.

At first some words about YSDA:

You can study here at any age. You will always be able to find things which are challenging and fun here  regardless of your experience. A lot of lecture/homework material is recent state of the art in machine learning. Reserve a lot of study time. Take minimal amount of courses but study them very deep with additional literature. Taking a lot of courses is not efficient at all. 

Below I will list individual courses:

• algorithms and data structures part1 (babenko) lectures are exceptionally good, homeworks are hard/olympiad level tasks.
• computer vision (konushin) - review lectures are very good (computer vision is broad field, so this course is compressed enough without losing much detail). homeworks are fundamental (including classic and deep learning methods).
• math for data science (trushin) - very good review lectures (linear algebra/math analysis/probability/statistics), homeworks are good to recall forgotten math fields.
• deep learning (lempitsky) - very good review lectures (dl - is quite huge field, so the lectures are very compressed without losing important ideas), homeworks are good also for practice.
• statistics in machine learning (burnaev) - good lectures based on several books (one of them - All of statistics by Wasserman), homeworks are heavy on theory and math (so they are hard).
• natural language processing (different lecturers) - good review lectures on nlp with a lot of advanced topics such as domain adaptation/summarization/translation/relation extraction and so on, homeworks are tough too.
• convex optimization (dorn,katrutsa) - dense compressed good lectures on convex optimization (covering classic book by Boyd and  then deepening into modern convex optimization fields - proximal gradient descent, robust optimization, constraint optimization, etc). the material is hard to comprehend and need a lot of time to study, one of the toughest courses in YSDA.
• self driving cars (different lecturers) - good review lectures on autonomous driving, hard homeworks (which require a lot of skills - from deep learning and optimization methods to control theory and mechanics). Because of broad topics - one of the toughest courses in YSDA.
• reinforcement learning (different lecturers) - good lectures by which you can understand fundamentals of rl. homeworks are very good to practice by implementing basic rl algorithms like dqn, policy gradients, bandits, actor critic, ppo, etc. A lot of bonus tasks in  homeworks.
• machine learning part1 (different lecturers) - good lectures for learning fundamentals of machine learning - classic methods/metrics/clustering/ensembles/bayes methods/svm. Some homeworks are actually competitions on kaggle - and they are very fun to compete.
• machine learning part2 (different lecturers) - good lectures for more modern methods like metric learning/deep learning/self-supervised approaches/ranking methods/recommendation systems/time series analysis/graph deep learning. Additional plus is homeworks for paper review.
• ml engineering practice (yozh) - for me this was a challenging task to reduce gpu memory in self-attention for long sequences which I've completed successfully.