Integrating Guidance into Relational Reinforcement Learning

By Kurt Driessens, Saso Dzeroski. 2004

106 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Fitted Q-Learning for Relational Domains (2020, methodology): “This work was later extended by others (Driessens and Ramon 2003; Driessens and Dˇzeroski 2004; Driessens, Ramon, and Blockeel 2001; G¨artner, Driessens, and Ramon 2003) w.r.t building incremental RRTs, providing guidance to the learner and regression algorithms using kernels and gaus-sian…”
Relational Deep Reinforcement Learning (2018, background): “The core idea behind RRL is to combine reinforcement learning with relational learning or Inductive Logic Programming [16] by representing states, actions and policies using a first order (or relational) language [8, 9, 17, 18].”
Deep reinforcement learning with relational inductive biases (2018, background): “Various structured approaches to RL (e.g. Dzeroski et al. (2001); Driessens & Dzeroski (2004); Diuk et al. (2008); Garnelo et al. (2016)) have attempted to overcome these limitations by explicitly incorporating entity-based and symbolic representations, and specialized building blocks for solving…”
Agent-Agnostic Human-in-the-Loop Reinforcement Learning (2017, background): “Prior literature has investigated how people can help RL agents learn more efﬁciently through different methods of interaction [32, 24, 28, 36, 42, 48, 21, 17, 25, 43, 23, 49, 46, 44, 30, 10].”
Non-parametric policy gradients: a unified treatment of propositional and relational domains (2008, methodology): “To counter this di.culty of reaching any reward, we adopted the active guidance approach as proposed by Driessens and Dzeroski (2004), presenting the RL agent with 10% of expert traces during exploration in all experiments.”
Neural Logic Reinforcement Learning (2019, background): “There are variants of this work (Driessens & Ramon, 2003; Driessens & Dˇzeroski, 2004) that extend the work, however, all these algorithms employ non-differential operations which makes it hard to apply new breakthroughs happened in DRL community.”
A comprehensive survey on safe reinforcement learning (2015, methodology): “This prior knowledge can be used to provide initial information to the RL algorithm biasing the subsequent exploratory process (Driessens and Džeroski, 2004; Martín H. and Lope, 2009; Koppejan and Whiteson, 2011), to provide a finite set of demonstrations on the task (Abbeel and Ng, 2005; Abbeel et…”
Graph kernels and Gaussian processes for relational reinforcement learning (2006, unknown): “Whether, for a given function k : X × X → R, a feature transformation φ : X → H into some Hilbert space H exists, such that k(x, x′) = 〈φ(x), φ(x′)〉 for all x, x′ ∈ X can be checked by verifying that the function is positive definite [1].”
Safe Exploration Algorithms for Reinforcement Learning Controllers (2018, background): “…approaches: “providing initial knowledge,” directing the learning in its initial stage toward more proﬁtable and safer regions of the state space [26]; “deriving a policy from demonstrations” by means of learning from demonstration [27]; “providing teacher advice” by including an external…”
Relational reinforcement learning (2001, unknown): “The ﬁrst is a methodology to supply guidance to the RRL system through the use of ex-ternal, reasonable policies that enables the RRL sys-tem to perform better on tasks with sparse and hard to reach rewards [2,3].”
An Overview of Cooperative and Competitive Multiagent Learning (2005, methodology): “One such succesfull method in singleagent learning has been the cross fertilization between reinforcement learning and inductive logic programming [41, 39, 168].”