GRK 2340: Computational Cognition

计算认知研究训练组致力于认知科学与人工智能的再整合。该项目的学生将在这两个领域接受培训，并将结合这两个领域的见解来理解人类和机器的智能。我们已经确定了两个研究领域，这种重新整合将特别富有成果。首先，低级认知和高级认知之间存在着分裂。我们对基本感觉运动过程的神经信号有了很多了解，对推理、解决问题或语言的认知过程也有了一定的了解。然而，解释高级认知如何从低级机制产生是认知科学中一个长期存在的开放性问题。机器学习最近在深度学习方法和循环神经网络方面取得了很大进展。与此同时，认知科学家也探索了类似的想法，比如统一神经学习理论的预测编码。第一个集群中的博士项目将解决诸如语法学习、结构化表示或使用神经建模生成复杂行为等问题。因此，整合认知科学和人工智能的思想将使我们最终能够弥合低级和高级认知之间的差距。第二，人类智能处理的是高度结构化但不完整的知识。因此，潜在的表示和过程能够产生新的概念，并考虑到不确定性。沿着这些思路，类比推理、语言、语用推理和概念形成被认为是理解人类智能的关键。第二个集群的博士项目将解决这些领域的典型问题，这些问题对人类来说很容易，但对人工智能来说仍然很难。它们包括类比推理、概念发明和语用推理。对于每一个认知过程，都有来自认知科学的可靠数据和见解，使我们能够对这些过程进行建模。与人工智能早期相比，这是一个巨大的优势，当时人们对人类认知知之甚少。新的研究训练组将被整合到2002年成立的认知科学博士项目中。研究训练组的学生将受益于高度跨学科的环境，然而，这种环境侧重于一个共同的主题，并在项目之间提供许多方法上的协同作用。

The Research Training Group in Computational Cognition pursues the re-integration of cognitive science and artificial intelligence. Students in the program will be trained in both fields and will combine insights from both fields to understand intelligence in humans and machines. We have identified two areas of research where this re-integration will be particularly fruitful.First, there is a schism between low- and high-level cognition. We understand a lot about the neural signals underlying basic sensorimotor processes, and we know a fair bit about the cognitive processes involved in reasoning, problem solving, or language. However, explaining how high-level cognition can arise from lowlevel mechanisms is a long-standing open problem in cognitive science. Machine learning has recently made great progress on deep learning methods and recurrent neural networks. At the same time, cognitive scientists have explored similar ideas, such as predictive coding for unified neural theories of learning. PhD projects in the first cluster will tackle problems, such as grammar learning, structured representations, or the production of complex behaviors with neural modeling. Thus, integrating ideas from cognitive science and AI will allow us to finally bridge the gap between low- and high-level cognition.Second, human intelligence deals with highly structured, yet incomplete knowledge. Thus, the underlying representations and processes are able to generate new concepts and to take into account uncertainty. Along these lines, analogical reasoning, language, pragmatic inference and concept formation have been proposed as being the key to understand human intelligence. PhD projects in the second cluster will tackle exemplary problems of these domains that are easy for humans, but still hard for AI. They include analogical reasoning, concept invention, and pragmatic inferences. For each of these cognitive processes, there are reliable data and insights from cognitive science that will allow us to model these processes. This is a large advantage compared to the early days of AI where little was known about human cognition.The new Research Training Group will be integrated into the Cognitive Science PhD program that was established in 2002. Students in the Research Training Group will benefit from a highly interdisciplinary environment that is, nevertheless, focused on a common theme and that provides many methodological synergies between projects.