Loading...
Please wait, while we are loading the content...
Similar Documents
Based Reinforcement Learning in DEC-POMDPs with Bayesian Nonparametrics
| Content Provider | Semantic Scholar |
|---|---|
| Author | Liu, Miao |
| Copyright Year | 2015 |
| Abstract | Decentralized POMDPs (Dec-POMDPs) are rich models for multiagent decision making under uncertainty, but generating high-quality solutions is difficult. Recent work has shown that Expectation maximization (EM) is an efficient algorithm for learning finite-state controllers (FSCs) in large decentralized POMDPs (Dec-POMDPs). However, current methods use fixed-size FSCs and often converge to maxima that are far from the optimal value. A recent work decentralized stick-breaking policy representation (Dec-SBPR) [16] addresses this problem by representing the local policy of each agent using variable-sized FSCs that are constructed using a nonparametric prior, stick-breaking processes. This approach infers controllers with a variational Bayesian algorithm without having to assume that the Dec-POMDP model is available. In this paper, we provide additional theoretical analysis to establish the relation between exploration rates and policy value, and provide additional experimental results demonstrating the scalability of the proposed method. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://web.mit.edu/miaoliu/www/publications/nips_malic15.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
