## Relational Reinforcement Learning (2001)

Citations: | 103 - 6 self |

### BibTeX

@MISC{Dzeroski01relationalreinforcement,

author = {Saso Dzeroski and Luc De Raedt and Kurt Driessens},

title = { Relational Reinforcement Learning},

year = {2001}

}

### Years of Citing Articles

### OpenURL

### Abstract

Relational reinforcement learning is presented, a learning technique that combines reinforcement learning with relational learning or inductive logic programming. Due to the use of a more expressive representation language to represent states, actions and Q-functions, relational reinforcement learning can be potentially applied to a new range of learning tasks. One such task that we investigate is planning in the blocks world, where it is assumed that the effects of the actions are unknown to the agent and the agent has to learn a policy. Within this simple domain we show that relational reinforcement learning solves some existing problems with reinforcement from specific goals pursued and to exploit the results of previous learning phases when addressing new (more complex) situations.

### Citations

4931 | C4.5: Programs for Machine Learning - Quinlan - 1993 |

3905 |
Classification and Regression Trees
- Breiman, Friedman, et al.
- 1984
(Show Context)
Citation Context ...on(a,floor). on(a,floor). on(a,floor). on(c,floor). on(c,floor). on(c,floor). 4.3 TOP-DOWN INDUCTION OF LOGICAL REGRESSION TREES Logical regression trees are similar to propositional regression trees =-=[3]-=-: leaves predict a value for a continuous class, while internal nodes contain conditions that partition the example space. The difference is that examples here are not feature or attribute-value vecto... |

3351 | Induction of decision trees - Quinlan - 1986 |

1740 |
Strips: A new approach to the application of theorem proving to problem solving
- Fikes, Nilsson
- 1972
(Show Context)
Citation Context ...i(s; a 1 )):::); :::a i )) = t. The agent can be in one of the states of S. It can execute action a 2 A in a given state s if the preconditions for a are true in s (pre(s; a) = t), e.g., as in STRIPS =-=[7]-=-. Executing an action a in a state s will put the agent in a new state ffi(s; a). When placed in a state s the task of the agent is to find a (shortest) sequence of actions a 1 ; :::; an that will lea... |

1303 | Reinforcement Learning: A Survey
- Kaelbling, Littman, et al.
- 1996
(Show Context)
Citation Context ...cific goals pursued and exploit the results of previous learning phases when addressing new (more complex) situations. 1 INTRODUCTION Within the field of machine learning, both reinforcement learning =-=[8]-=- and inductive logic programming (or relational learning) [12, 10] have received a lot of attention since the early nineties. It is therefore no surprise that both Leslie Pack Kaelbling and Richard Su... |

853 | Leaming Logical Definitions from Relations - Quinlan - 1990 |

455 | Inductive logic programming: Theory and methods - Muggleton, Raedt - 1994 |

444 | The discipline of machine learning - Mitchell - 2006 |

426 | Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning - Sutton, Precup, et al. - 1999 |

373 |
Temporal difference learning and TD-gammon
- Tesauro
- 1995
(Show Context)
Citation Context ...considering problems outside the blocks world. This work is related to work on generalization in reinforcement learning, which has however mainly addressed the use of neural networks for this purpose =-=[13]-=-. The closest related work is probably Chapman's and Kaelbling's decision tree algorithm that was specifically designed for reinforcement learning [5]. Note however that our approach is distinguished ... |

276 | Self-Improving Reactive Agents Based On Reinforcement Learning, Planning and Teaching - Lin - 1992 |

134 | Input generalization in delayed reinforcement learning: an algorithm and performance comparisons
- Chapman, Kaelbling
- 1991
(Show Context)
Citation Context ...ssed the use of neural networks for this purpose [13]. The closest related work is probably Chapman's and Kaelbling's decision tree algorithm that was specifically designed for reinforcement learning =-=[5]-=-. Note however that our approach is distinguished from the mainstream work in reinforcement learning by the use of a relational representation. Relational representations are commonly used in planning... |

126 | Top-down Induction of First Order Logical Decision Trees - Blockeel - 1998 |

122 | S.: Inductive Logic Programming: Techniques and Applications - Lavrac, Dzeroski - 1994 |

122 | Learning by experimentation: Acquiring and refining problem solving heuristics - Mitchell, Utgoff, et al. - 1983 |

120 | Decision tree induction based on efficient tree restructuring - Utgoff, Berkman, et al. - 1997 |

119 | Approaches to Machine Learning - Langley, Carbonell - 1984 |

101 | Top-down induction of clustering trees - Blockeel, Raedt, et al. - 1998 |

69 | Lazy incremental learning of control knowledge for efficiently obtaining quality plans
- Borrajo, Veloso
- 1996
(Show Context)
Citation Context ...al representations are commonly used in planning approaches. There have also been some attempts to combine planning with relational learning within those approaches, e.g., within the PRODIGY approach =-=[2]-=-. Our approach is related to them through the use of a relational representation. However, it seems that the combination of planning, reinforcement learning and relational learning has not been addres... |

69 | Induction of first-order decision lists: Results on learning the past tense of English verbs - Mooney - 1995 |

64 | Structural regression trees
- Kramer
- 1996
(Show Context)
Citation Context ...does not know the effects of its actions. Relational reinforcement learning employs the Q-learning method [14, 8, 11] where the Q-function is learned using a relational regression tree algorithm (see =-=[6, 9]-=-). A state is represented relationally as a set of ground facts. A relational regression tree in this context takes as input a relational description of a state, a goal and an action, and produces the... |

49 | R.G.: The effect of representation and knowledge on goaldirected exploration with reinforcement-learning algorithms - Koenig, Simmons - 1996 |

42 | Scaling up inductive logic programming by learning from interpretations - Blockeel, Raedt, et al. - 1999 |

37 | First order regression - Karalic, Bratko - 1997 |

29 | Toward a model of mind as a laissez-faire economy of idiots - Baum - 1996 |

27 | Raedt. Lookahead and discretization in ilp - Blockeel, De - 1997 |

27 | Declarative bias in ILP - Nédellec, Rouveirol, et al. - 1996 |

26 | Explanation based generalization: A unifying view - Mitchell, Keller, et al. - 1986 |

22 | Using logical decision trees for clustering
- Raedt, Blockeel
- 1997
(Show Context)
Citation Context ...does not know the effects of its actions. Relational reinforcement learning employs the Q-learning method [14, 8, 11] where the Q-function is learned using a relational regression tree algorithm (see =-=[6, 9]-=-). A state is represented relationally as a set of ground facts. A relational regression tree in this context takes as input a relational description of a state, a goal and an action, and produces the... |

9 | Team-Partitioned, Opaque Transition Reinforcement Learning - Stone, Veloso - 1999 |

7 | Strategy acquisition governed by experimentation - Langley - 1982 |

6 | M.: Special issue on context sensitivity and concept drift. Machine Learning 32 - Widmer, Kubat - 1998 |

2 |
Experiments with Top-down Induction of Logical Decision Trees
- Blockeel, Raedt
- 1997
(Show Context)
Citation Context ...ppear in the root of the tree (goal(on(C,D))). The Prolog program corresponding to the tree is shown in the lower part of Figure 2. The semantics of logical decision trees is extensively discussed in =-=[1]-=-, as well as the correspondence between a tree and a Prolog program. The method to induce the trees is described in [6] and is - for the case of regression trees - very similar to Kramer's SRT system ... |