Contact
How to shop
Delivery
Shopping guide
English
106 b
Delivery to Austria
Up to 30 days for returns
Customers also purchased
/
Paperback
8.49
€
Reinforcement learning is the problem faced by an§agent that must learn behavior through§trial-and-error interactions with a dynamic§environment. Usually, the problem to be solved§contains subtasks that repeat at different regions of§the state space. Without any guidance§an agent has to learn the solutions of all subtask§instances independently, which in turn degrades the§performance of the learning process. In this work, we§propose two novel approaches for building the§connections between different regions of the search§space. The first approach efficiently discovers§abstractions in the form of conditionally terminating§sequences and represents these abstractions compactly§as a single tree structure; this structure is then§used to determine the actions to be executed by the§agent. In the second approach, a similarity function§between states is defined based on the number of§common action sequences; by using this similarity§function, updates on the action-value function of a§state are re ected to all similar states that allows§experience acquired during learning be applied to a§broader context. The effectiveness of both approaches§is demonstrated empirically over various domains.
About the book
English
Give this book today
It's easy
1 Add to cart and choose Deliver as present at the checkout 2 We'll send you a voucher 3 The book will arrive at the recipient's addressYou might also be interested in
/
Paperback
12.09
€
/
Paperback
20.79
€
/
Paperback
10.89
€
/
Hardback
108.89
€
/
Paperback
17.09
€
