Search

This report documents the program and the outcomes of Dagstuhl Seminar 17192 "Human-Like Neural-Symbolic Computing", held from May 7th to 12th, 2017. The underlying idea of Human-Like Computing is to incorporate into Computer Science aspects of how humans learn, reason and compute. Whilst recognising the relevant scientific trends in big data and deep learning, capable of achieving state-of-the-art performance in speech recognition and computer vision tasks, limited progress has been made towards understanding the principles underlying language and vision understanding. Under the assumption that neural-symbolic computing - the study of logic and connectionism as well statistical approaches - can offer new insight into this problem, the seminar brought together computer scientists, but also specialists on artificial intelligence, cognitive science, machine learning, knowledge representation and reasoning, computer vision, neural computation, and natural language processing. The seminar consisted of contributed and invited talks, breakout and joint group discussion sessions, and a hackathon. It was built upon previous seminars and workshops on the integration of computational learning and symbolic reasoning, such as the Neural-Symbolic Learning and Reasoning (NeSy) workshop series, and the previous Dagstuhl Seminar 14381: Neural-Symbolic Learning and Reasoning.

We discuss and illustrate Neural Blackboard Architectures (NBAs) as the basis for variable binding and combinatorial processing the brain. We focus on the NBA for sentence structure. NBAs are based on the notion that conceptual representations are in situ, hence cannot be copied or transported. Novel combinatorial structures can be formed with these representations by embedding them in NBAs. We discuss and illustrate the main characteristics of this form of combinatorial processing. We also illustrate the NBA for sentence structures by simulating neural activity as found in recently reported intracranial brain observations. Furthermore, we will show how the NBA can account for ambiguity resolution and garden path effects in sentence processing.

This report documents the program and the outcomes of Dagstuhl Seminar 14381 "Neural-Symbolic 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.

Providing personalized feedback in Intelligent Transport Systems is a powerful tool for instigating a change in driving behaviour and the reduction of CO2 emissions. This requires a system that is capable of detecting driver characteristics from real-time vehicle data. In this paper, we apply the architecture and theory of a Neural-Symbolic Cognitive Agent (NSCA) to effectively learn and reason about observed driving behaviour and related driver characteristics. The NSCA architecture combines neural learning and reasoning with symbolic temporal knowledge representation and is capable of encoding background knowledge, learning new hypotheses from observed data, and inferring new beliefs based on these hypotheses. Furthermore, it deals with uncertainty and errors in the data using a Bayesian inference model, and it scales well to hundreds of thousands of data samples as in the application reported in this paper. We have applied the NSCA in an Intelligent Transport System to reduce CO2 emissions as part of an European Union project, called EcoDriver. Results reported in this paper show that the NSCA outperforms the state-of-the-art in this application area, and is applicable to very large data.

Deep Boltzmann Machines (DBM) have been used as a computational cognitive model in various AI-related research and applications, notably in computational vision and multimodal fusion. Being regarded as a biological plausible model of the human brain, the DBM is also becoming a popular instrument to investigate various cortical processes in neuroscience. In this paper, we describe how a multimodal DBM is implemented as part of a Neural-Symbolic Cognitive Agent (NSCA) for real-time multimodal fusion and inference of streaming audio and video data. We describe how this agent can be used to simulate certain neurological mechanisms related to hallucinations and dreaming and how these mechanisms are beneficial to the integrity of the DBM. Finally, we will explain how the NSCA is used to extract multimodal information from the DBM and provide a compact and practical iconographic temporal logic formula for complex relations between visual and auditory patterns.

We discuss the purpose of neural-symbolic integration including its principles, mechanisms and applications. We outline a cognitive computational model for neural-symbolic integration, position the model in the broader context of multi-agent systems, machine learning and automated reasoning, and list some of the challenges for the area of neural-symbolic computation to achieve the promise of effective integration of robust learning and expressive reasoning under uncertainty.

International audience