https://mingfei.io/Recent content on Mingfei's BlogHugoen-usSat, 15 Jun 2024 00:00:00 +0000https://mingfei.io/dqn/Sat, 15 Jun 2024 00:00:00 +0000https://mingfei.io/dqn/Deep Q-Network (DQN) is an Reinforcement Learning (RL) algorithm developed by Mnih et al. at DeepMind in 2013, which combines the classic RL algorithm Q-Learning with deep neural networks. It is considered as the first major success story of deep reinforcement learning due to its super-human level performance in Atari games – it takes only raw pixels as input (i.e., it “sees” exactly what a human sees), and is able to outperform humans in a wide range of Atari games.https://mingfei.io/leaving-fb/Sun, 24 Mar 2024 00:00:00 +0000https://mingfei.io/leaving-fb/I left Facebook/Meta on Friday. When I joined Facebook 11 years ago, I was just hoping to work with cool hackers for a few years, experience the Silicon Valley tech scene, make some money, and then go back to school to continue my theoretical computer science research. I soon became obsessed with the impact I got to make on billions of people, the coworkers I got to collaborate with and learn from, and the new skills and mindset shifts I got to pick up every day.https://mingfei.io/gcp-setup/Fri, 18 Aug 2023 00:00:00 +0000https://mingfei.io/gcp-setup/Note: Jun 16, 2024 Update Almost a year after writing this post, I realized that the best way to setup Google Cloud for deep learning (as an individual) is to NOT set it up at all! GPU rental platforms like vast.ai or runpod.io offer much cheaper and easier to use options for individual users, with a much wider selection of GPUs and shorter/no waiting times. I highly recommend that you check them out before you settle on a major cloud platform (unless you have credits from the cloud platform which changes the cost/benefit equation).