斯坦福大学计算机科研项目

Stanford University CS Research

科研主题

Game theory  [博弈论]

Artificial intelligence  [人工智能]

Machine learning  [机器学习]

Software robots  [软体机器人]

Computer  [计算机]

Big data and its application  [大数据及应用]

参考课题

Title: Playing Games and Controlling Robots with Deep Reinforcement Learning

Introduction

Reinforcement Learning (RL) is a class of methods that train an agent to maximize reward by interacting with the environment. By learning from experience and exploring the environment, the agent can learn about the dynamics of the environment and figure out best ways to accomplish tasks.

Deep Learning (DL) is a class of methods that take inspiration from signal processing in the brain.

Through massively parallel computation with millions of neurons, the system can learn to accomplish complex tasks such as visual perception, audio understanding, natural language translation, and even reasoning.

In this project, we combine reinforcement learning and deep learning to train an agent that can interact with a complex environment. With deep learning, the agent can process complex visual input typically associated with a robotics or game environment, and with reinforcement learning, the agent can learn to accomplish goals based on its processing of the environment.

Example applications of combination of RL and DL include AlphaGo from Google Deepmind, and

advanced robotics control from OpenAI and UC Berkeley.

Requirement

To accomplish the project, participants are expected to posess the following skills

※Required:

Strong programming skills, familiar with at least one programming language.

Good math skills, familiar with algebra and probability.

※Recommended:

Experience with using linux-based shell environments.

Basic knowledge of linear algebra.

Familiarity with python.

Basic knowledge of machine learning.

This project is advanced and challendging. If the student do not have enough prior experience to

finish the project, he or she may participate in a simplied version of the project. i.e. reinforcement

learning only or deep learning only, depending on the specific situation of the student.

Schedule

Week 1: Basic Knowledge, Python, Introduction to Machine Learning

Week 2: Reinforcement Learning

Week 3: Deep Learning

Week 4: Deep Reinforcement Learning