A new advanced nanodegree on Deep Reinforcement Learning from Udacity's School of AI has just started. So has another presentation of Coursera's Deep Learning Specialization taught by Andrew Ng. 

Disclosure: When you make a purchase having followed a link from this article, we may earn an affiliate commission.

There's a lot of interest and excitement about "Deep Learning" which is currently regarded as the hottest research field in artificial intelligence. The approach of combining reinforcement learning with neural networks first hit the headlines when DeepMind’s AlphaGo defeated the world champion Go player Lee Sedol. The most recent story we have followed is about OpenAI Five, a team of AI agents that have learn to play at Dota 2, one of the most complex esports games in the world. As Udacity claims, Deep Reinforcement Learning is:

the closest we’ve yet come to developing AI that can learn and develop like a human does!

We first heard of the Deep Reinforcement Learning Nanodegree when Udacity launched its School of AI back in March. It is at advanced level and requires experience with Python, probability and machine learning, all of which can be acquired through existing Udacity courses, in particular its Machine Learning Engineer Nanodegree, see More Machine Learning From Udacity. It also needs you to already have knowledge of deep learning, which can be gained by taking Udacity's intermediate-level Deep Learning Nanodegree which covers cutting-edge topics such as neural, convolutional, recurrent neural, and generative adversarial networks and has projects in Keras and NumPy, in addition to TensorFlow. An updated version which has its main examples in PyTorch starts on September 11 and this trailer tells you more:

Enrollment for the first presentation of Deep Reinforcement Learning closes on August 28. It can be completed in 4 months by studying 10-15 hours per week and costs $999. As in other nanodegrees, students work on projects and view video lectures. In this case there are four course and three projects:.

• Course 1: Foundations of Reinforcement Learning
  Master the fundamentals of reinforcement learning by writing your own implementations of many classical solution methods.
  
  
• Course 2: Value-Based Methods
  Apply deep learning architectures to reinforcement learning tasks. Train your own agent that navigates a virtual world from sensory data.
  
  
• Project 1: Navigation
  Use neural networks to train an agent that learns intelligent behaviors from sensory data
  
  
• Course 3: Policy-Based Methods
  Learn the theory behind evolutionary algorithms and policy-gradient methods. Design your own algorithm to train a simulated robotic arm to reach target locations.
  
  
• Project 2: Continuous Control
  Train a robotic arm to reach target locations, or train a four-legged virtual creature to walk.
  
  
• Course 4: Multi-Agent Reinforcement Learning
  Learn how to apply reinforcement learning methods to applications that involve multiple, interacting agents. These techniques are used in a variety of applications, such as the coordination of autonomous vehicles.
  
  
• Project 3: Collaboration and Competition
  Train a system of agents to demonstrate collaboration or cooperation on a complex task.

In creating this nanodegree Udacity collaborated with Unity and the NVIDIA Deep Learning Institute to build a program with a balance of theory and practical application, and which enables participants to explore compelling challenges in fields ranging from gaming to finance to robotics. Over the course of the program, students will implement several deep reinforcement learning algorithms using a combination of Python and PyTorch, to build projects that will serve as GitHub portfolio pieces to showcase your mastery of this advanced field. The GitHub repo for the course contains tutorials that make up the course. 

We looked at Coursera's Deep Learning Specialization taught by Andrew Ng and others from his Deeplearning.ai company when it first started a year ago and before there were details of all its modules. Now it is well established and each of the courses is rated 4.8 out of 5 or above.

At intermediate level and using Python and Tensorflow, this specialization consists of 5 short courses, each of which can also be taken standalone. Much of the material can be audited for free but to get the full benefit, and official certificates on successful completion of courses, the monthly fee of $49 gives unlimited access to the entire Coursera catalog and there is a 7-day free trial. 

Neural Networks and Deep Learning 
4 weeks, 3-6 hours per week

Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization 
3 weeks, 3-6 hours per week

Structuring Machine Learning
2 weeks, 3-4 hours per week

Convolutional Neural Networks 
4 weeks, 4-5 hours per week

Sequence Models
3 weeks, 4-6 hours per week

Over these five courses, you will learn the foundations of Deep Learning, understand how to build neural networks, and learn how to lead successful machine learning projects. You will learn about Convolutional networks, RNNs, LSTM, Adam, Dropout, BatchNorm, Xavier/He initialization, and more. You will work on case studies from healthcare, autonomous driving, sign language reading, music generation, and natural language processing. You will master not only the theory, but also see how it is applied in industry. 

UPDATE June 2020:

They also provide to moving on to the second specialization from deeplearning.ai, Natural Language Specialization, which was launched on June 19, 2020 - see New Natural Language Processing Specialization on Coursera.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info