دانلود کتاب مقدمه‌ای بر یادگیری ماشین

These lecture notes accompany a junior-level machine learning course (COS 324) at Princeton University. This course provides a broad introduction to machine learning paradigms including supervised, unsupervised, deep learning, and reinforcement learning as a foundation for further study or independent work in ML, AI, and data science. Topics include linear models for classification and regression, clustering, low rank representations (PCA), n-gram language models, matrix factorization, feedforward neural nets and convolutional neural nets, Markov decision process, and reinforcement learning. Interesting applications are presented for all these models. The course design was shaped by some constraints that may not exist at other universities. Background assumed: The formal prerequisites are the following courses, which all our majors have taken through sophomore year: An introduction to computer science course (COS 126), Data Structures and Algorithms (COS 226), Single-variable Calculus (MAT 103, 104) and Linear Algebra (MAT 202/204/217). While many majors also take multi-variable calculus and a probability course, not all do. Hence we do not include them in the list of prerequisites, although we do assume some exposure to elementary probability at high-school level. All our majors take a course on proof-based reasoning (COS 240) but many don’t take it before junior year. Hence that course is not a prerequisite and our course doesn’t rely on formal proofs per se. A side benefit of assuming minimal math and programming prerequisites is that this also makes our course accessible to hundreds of non-majors, many of whom only take introductory CS courses. Should provide a broad introduction to today’s AI and machine learning: Since AI and Machine Learning has been transformed by deep learning and related methods in the past decade, we wanted to provide a reasonable competence in deep learning as well as reinforcement learning. Thus students who don’t take another AI/ML course still leave with a skill set appropriate for applying ML techniques in their careers. Of course, devoting the second half of our course to these “advanced” topics required leaving out several topics that are classically taught in introductory ML. Essentially, the classical topics are condensed into the first half of the term. Provide a taste of today’s programming environments: Students can start with zero Python background, and slowly graduate to comfort in Python as well as dipping their toes into deep learning on Google CoLab and RL on OpenAI gym. Taste of interesting applications + discussion of broader issues of societal interest: An introductory course needs to introduce students to these issues, given machine learning’s ubiquitous role in research across disciplines, as well as throughout our economy and our society. Sometimes we invite guest lecturers to provide such perspectives.