Li Feifei like Andrej karpathy for everyone answering questions – Science and technology Sohu x3210

Li Feifei like Andrej Karpathy answering questions for everyone – the Sohu technology editor’s note: Li Feifei like Andrej KarPathy2015 in Stanford University received a Ph.D. in computer science, 2016 in OpenAI, the main research interest: deep learning, generative model and reinforcement learning. From 2011 to 2015 in Google Brain, Deepmind and the major DL laboratory practice, in learning and working experience. This is his study field for the majority of students answering machine in Quora, we hope to inspire. How do you learn to study ML DL? What are your favorite books in the process of learning DL? I’ve been talking about this topic in an interview with the data science journal. We want to do is to make a long story short, I used quantum computation in this together, after that a little bit and realized that AI is get disheartened, I want to study the most important "Yuan" problem. During my PhD study, there was not much reference to DL. There are now books and other sources of information such as Ian Goodfellow depth learning (e.g., many lectures, CS231n, etc.). I personally don’t rely too much on reference books. I love Bishop’s book, in read during the period from A to Z had read many times, and Sutton reinforcement learning (Reinforcement Learning) a book, the book I read in a few weeks fast again, repeated practice on ReinforceJS chapter of knowledge. But, unfortunately, this book in the policy gradient (Policy Gradients) this together to speak less, and we used this knowledge is still pretty much in the study, however, this book lays a good foundation for my DL learning. Up to now, I have found that the trick to learn a lot of knowledge is to practice the theoretical knowledge that has been learned. No matter when, when I read some knowledge, I would imagine they have a thorough understanding of the practice, and then forced himself repeatedly, this method always brings new and interesting insights to me. This is my favorite way to learn. What is your personal preference for a deep learning framework? During the study, I have experienced several transition periods in the process of learning the deep learning framework. At first it was Matlab, when everyone used the software. However, unfortunately, Matlab is not a real language, when all people laugh at me, so I turned to learn Python numpy, handwritten out all my own back propagation algorithm code. Unfortunately, however, numpy does not apply to GPU, so learning Python also does not work. After that, I began to learn Torch, I was very fond of this calculation framework, to now still like. Torch is simple: you!相关的主题文章: