Neural-Symbolic Learning: How to Play Soccer

National Science Review, 2022

2001

This article contributes to the understanding of how symbols can profitably be integrated into a numeric reinforcement learning approach so as to realize a learning robot in a continuous problem setting. An initial experiment clearly revealed that the underlying task could not be learned merely on the basis of numeric reinforcement values. The unstructured utilization of sensory values of the reinforcement approach is dealt with by incorporating readily available domain knowledge. The domain knowledge pre-labels the task space into different regions and builds a connection between the direct sensory experience and the high level features. Teams of networks, each dedicated to a particular feature, exploit the differential characteristics of the regions to rapidly learn the sub-tasks. In addition, a pre-wired fuzzy controller whose function is to restrict the set of actions from which the learner composes a control policy is integrated into the learner. The task faced by the robot has...

A neuro-symbolic reasoning strategy for modelling a complex system is presented in which the aim is to forecast, in real time, the physical parameter values of a dynamic environment: the ocean. In situations in which the rules that determine a system are unknown the prediction of the parameter values that determine the characteristic behaviour of the system can be a problematic task. In such a situation it has been found that a case-based reasoning system, in combination with an artifical neural network, can provide a more effective means of performing such predictions than other connectionist or symbolic techniques. The case-based reasoning system incorporates a radial basis function artificial neural network for the case adaptation. The results obtained from experiments, in which the system operated in real time in the oceanographic environment, are presented.

Robot Soccer, 2010

Neural networks and symbolic systems are two different approaches to model cognitive functions. Both methods have certain strengths but also suffer from a number of weak points, which are however disparate. By combining characteristics of neural and symbolic processing, these weaknesses could be overcome. This paper introduces a new information processing principle based on so-called neuro-symbolic networks, which incorporates these two approaches. The utility of the suggested principle is outlined for applications in machine perception.

Proceedings of the 2012 Japan-Egypt Conference on Electronics, Communications and Computers, JEC-ECC 2012, 2012

During the last decade many research efforts have been directed towards studying the relative merits of the symbolic (rooted in logic, easily understandable) and nonsymbolic (numeric, difficult to understand) Artificial Intelligence (AI). Specifically, efforts have been directed towards discovering techniques to translate between knowledge available in one format to another; such as between Fuzzy Rule-based Systems (FRS) and Artificial Neural Networks (ANNs); combining both formats in a single hybrid system; such as Adaptive Neuro-Fuzzy Systems (ANFIS); or even equating both of them by introducing a new fuzzy logic operator [1]. The present paper proposes a new framework; based on a modification of the work given in [1]; that has several advantages over pure FRS, pure ANN systems and existing hybrid approaches. It is capable of producing meaningful plausible rules whether prior expert's knowledge is available or not. The theoretical foundation of this framework as well as its application to a path tracking case study are discussed. Its suitability for the solution of general optimization problems is highlighted and an example is given.

2015

This report documents the program and the outcomes of Dagstuhl Seminar 14381 “NeuralSymbolic Learning and Reasoning”, which was held from September 14th to 19th, 2014. This seminar brought together specialist in machine learning, knowledge representation and reasoning, computer vision and image understanding, natural language processing, and cognitive science. The aim of the seminar was to explore the interface among several fields that contribute to the effective integration of cognitive abilities such as learning, reasoning, vision and language understanding in intelligent and cognitive computational systems. The seminar consisted of contributed and invited talks, breakout and joint group discussion sessions. Seminar September 14–19, 2014 – http://www.dagstuhl.de/14381 1998 ACM Subject Classification I.2 Artificial Intelligence, I.2.4 Knowledge Representation Formalisms and Methods, I.2.6 Learning, I.2.10 Vision and Scene Understanding, I.2.11 Distributed Artificial Intelligence

Performance assessment in training simulators is a complex task. It requires monitoring and interpreting the student's behaviour in the simulator using knowledge of the training task, the environment and a lot of experience. Assessment in simulators is therefore generally done by human observers. To capture this process in an automated system is challenging and requires innovative solutions. This paper proposes a new module for automated assessment in simulators that is based on Neural-Symbolic Learning and Reasoning and the Recurrent Temporal Restricted Boltzmann Machine (RTRBM). The module is capable of using existing and learning new rules for performance assessment, by observing experts and students performing the training tasks. These rules are used to validate and support the assessment process and to automatically assess student performance in a training simulator. The module will be developed in a three year research project on assessment in driving simulators for testing and examination.

Recurrent Neural …

Smart Innovation, Systems and Technologies, 2011

Neurules are a kind of integrated rules integrating neurocomputing and production rules. Each neurule is represented as an adaline unit. Thus, the corresponding neurule base consists of a number of autonomous adaline units (neurules). Due to this fact, a modular and natural knowledge base is constructed, in contrast to existing connectionist knowledge bases. In this paper, we present an overview of our main work involving neurules. We focus on aspects concerning construction of neurules, efficient updates of neurule bases, neurule-based inference and combination of neurules with case-based reasoning. Neurules may be constructed from either symbolic rule bases or empirical data in the form of training examples. Due to the fact that the source knowledge of neurules may change with time, efficient updates of corresponding neurule bases to reflect such changes are performed. Furthermore, the neurule-based inference mechanism is interactive and more efficient than the inference mechanism used in connectionist expert systems. Finally, neurules can be naturally combined with case-based reasoning to provide a more effective representation scheme that exploits multiple knowledge sources and provides enhanced reasoning capabilities.