دانلود کتاب هندسه یادگیری عمیق – دیدگاه پردازش سیگنال

It was a very different, unprecedented, and weird start of the semester, and I did not know what to do. This semester, I was supposed to offer a new senior-level undergraduate class on Advanced Intelligence to jointly teach students at the Department of Bio/Brain Engineering and the Department of Mathematical Sciences. I had initially planned a standard method for teaching machine learning, the contents of which are practical, experience-based lectures with a lot of interaction with the students through many mini-projects and term projects. Unfortunately, the global pandemic of COVID-19 has completely changed the world and such interactive classes are no longer an option most of the time. So, I thought about the best way to give online lectures to my students. I wanted my class to be different from other popular online machine learning courses but still provide up-to-date information about modern deep learning. However, not many options were available. Most existing textbooks are already outdated or very implementation oriented without touching the basics. One option would be to prepare presentation slides by adding all the up-to-date knowledge that I wanted to teach. However, for undergraduate-level courses, the presentation files are usually not enough for students to follow the class, and we need a textbook that students can read independently to understand the class. For this reason, I decided to write a reading material first and then create presentation files based on it, so that the students can learn independently before and after the online lectures. This was the start of my semester-long book project on Geometry of Deep Learning. In fact, it has been my firm belief that a deep neural network is not a magic black box, but rather a source of endless inspiration for new mathematical discoveries. Also, I believed in the famous quote by Isaac Newton, “Standing on the shoulders of giants,” and looking for a mathematical interpretation of deep learning. For me as a medical imaging researcher, this topic was critical not only from a theoretical point of view but also for clinical decision-making, because we do not want to create false features that can be recognized as diseases.