Deep Learning Travels
Don't panic

Design of Digital Circuits
Summary of Design of Digital Circuits course by Onur Mutlu in ETH Zurich. Thank you very much for opening up the great lectures and materials for selflearners like me. This course provided me invaluable insight to understand computers. Introduction and Basics Mysteries in Comp Arch Meltdown and Spectre RowHammer Introduction...

Distributed Prioritized Experience Replay
WHY? Gorila framework separated several actors and learners with a centralized parameter server to parrallelize the learning process. This framework required one GPU per learner. WHAT? ApeX architecture only consists of two parts: many actors and one learner. Given a model from model, many actors generate experience simultaneously. The learner...

A Hierarchical Latent Variable EncoderDecoder model for Generating Dialogues
WHY? Hierarchical recurrent encoderdecoder model(HRED) that aims to capture hierarchical structure of sequential data tends to fail because model is encouraged to capture only local structure and LSTM often has vanishing gradient effect. WHAT? Latent Variable Hierarchical Recurrent EncoderDecoder(VHRED) tried to improve HRED by forcing to learn z with variational...

ForwardBackward Reinforcement Learning
WHY? Reinforcement learning with sparse reward often suffer from finding rewards. WHAT? ForwardBackward Reinforcement Learning(FBRL) consists of forward and backward process. Forward process is like normal rl using memory to update Q function. In backward process, new model is introduced called backward model b. b is a neural network that...

A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learning
WHY? This paper wanted to catch nonlinear dynamics of the object in video. WHAT? KVAE(Kalman Variational Autoencoder) combined Kalman filter with VAE to model dynamic latent variables. Linear Gaussian state space models are used to model Kalman filter and stable latent variables of the variational autoencoder. Matrices are the state...