Integrating reinforcement learning into a programming language
نویسنده
چکیده
My thesis work combines AI, programming language design, and software engineering. I am integrating reinforcement learning (RL) into a programming language so that the language achieves three primary goals: accessibility, adaptivity, and modularity. My language, AFABL (A Friendly Adaptive Behavior Language), will be an agent programming language designed to be accessible to nonprogramming experts like behavioral scientists, game designers, and intelligence analysts. If I am successful, my work will enable a discipline of modular large-scale agent software engineering while making advanced agent modeling accessible to authors of agent-based systems who are not programming experts. There is currently a spectrum of agent-based simulation and programming tools available to social scientists and game designers that runs from the very simple, like NetLogo (Gilbert and Troitsch 2005), to the very powerful, like ABL (A Behavior Language), a state of the art language developed specifically for believable agents (Mateas and Stern 2004). However, the current landscape of agent programming systems suffers from two fundamental and opposing weaknesses. On one hand, simple tools like NetLogo that are accessible to non-programming experts are poorly suited to large-scale agent systems in which the agents are very complex. On the other hand, powerful agent programming languages like ABL are too difficult for non-programming experts to understand and use effectively. With AFABL I intend to close this gap by creating an agent programming language that is accessible to non-programming experts while sacrificing none of the power of advanced languages like ABL. AFABL will be designed for writing adaptive software agents, that is, agents that learn to adapt to their environment during run-time, not software that is written to be easily changed by modifying the source code and recompiling. I am particularly interested in programming intelligent agents that operate in real environments, and in virtual environments that are designed to simulate real environments. Examples of these kinds of agents include robots, and nonplayer characters in interactive games and narratives. Unlike
منابع مشابه
Reinforcement Learning in Partially Observable Markov Decision Processes using Hybrid Probabilistic Logic Programs
We present a probabilistic logic programming framework to reinforcement learning, by integrating reinforcement learning, in POMDP environments, with normal hybrid probabilistic logic programs with probabilistic answer set semantics, that is capable of representing domain-specific knowledge. We formally prove the correctness of our approach. We show that the complexity of finding a policy for a ...
متن کاملIntegrating Relational Reinforcement Learning with Reasoning about Actions and Change
This paper presents an approach to the integration of Relational Reinforcement Learning with Answer Set Programming and the Event Calculus. Our framework allows for background and prior knowledge formulated in a semantically expressive formal language and facilitates the computationally efficient constraining of the learning process by means of soft as well as compulsive (sub-)policies and (sub...
متن کاملTD Models: Modeling the World at a Mixture of Time Scales
Temporal-diierence (TD) learning can be used not just to predict rewards, as is commonly done in reinforcement learning, but also to predict states, i.e., to learn a model of the world's dynamics. We present theory and algorithms for intermixing TD models of the world at diierent levels of temporal abstraction within a single structure. Such multi-scale TD models can be used in model-based rein...
متن کاملIntegrating Reinforcement Learning and Declarative Programming to Learn Causal Laws in Dynamic Domains
Robots deployed to assist and collaborate with humans in complex domains need the ability to represent and reason with incomplete domain knowledge, and to learn from minimal feedback obtained from non-expert human participants. This paper presents an architecture that combines the complementary strengths of Reinforcement Learning (RL) and declarative programming to support such commonsense reas...
متن کاملThe Effect of Computer Assisted Cooperative Language Learning on Iranian High School Students' Language Anxiety and Reading Comprehension
This study explored the effectiveness of the two computer-assisted modes: cooperative and individual on improving Iranian high school students’ reading comprehension. It was also concerned with investigating the effectiveness of the two computer-assisted modes on the participants’ foreign language learning anxiety (FLLA). The sample of the study consisted of two intact groups, each containing 2...
متن کامل