Solving Partially Observable Environments with Universal Search Using Dataflow Graph-Based Programming Model

Abstract

Universal search allows one to create an asymptotically optimal intelligent agent, which can act in a wide range of computable environments. However combinatorial explosion is still lurking behind though its dependency shifted from problem size to solution size. To combat this scenario, we propose a dataflow graph-based functional programming model to be used in the universal search for solution program generation. We have justified the superiority of our proposed model compared to sequential token-based languages when used in universal search-based intelligent agents. We have shown how applying an equivalent program pruning strategy can handle the problem of semantically redundant program generation. An incremental learning strategy based on gradient ascent is also proposed for our designed agent. Experimental results positively reinforced the theoretical justifications. We used our agent to solve some partially observable environments and compared them with the current state of art methods and it reveals the exceptional performance of our agent.

KEYWORDS:

• Artificial intelligence
• Dataflow graph programming
• Functional programming
• POMDP
• Reinforcement learning
• Universal search

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Notes on contributors

Swarna Kamal Paul

Swarna Kamal Paul received his bachelor’s degree in electronics and communication engineering from West Bengal University of Technology, India in 2008. He joined Tata Consultancy Services in 2008. He received master’s degree in software engineering from Jadavpur University, India in 2014. He is currently working in Tata Consultancy Services and pursuing PhD from Department of Information Technology, Jadavpur University. His research interests include artificial general intelligence, machine learning, and deep learning.

Parama Bhaumik

Parama Bhaumik is currently associate professor in the Department of Information Technology at Jadavpur University. She received her BTech in computer science and engineering in addition to BSc in physics (Hons) from Calcutta University. She has done her MTech in computer science and engineering from Calcutta University. She pursued her PhD in the year 2009 from Jadavpur University. Her teaching and research encompass challenged networks, artificial general intelligent systems with special emphasis on efficient design and management of ICT-based systems. She has over 45 publications in journals, conferences and book chapters. Email: [email protected]