Test of an integrated associative theory of timing

理解学习是心理学的一个中心问题，已经被广泛研究，特别是在动物学科中。联想学习理论认为，共同发生的事件的心理表征是相互联系的，因此一个事件的呈现可以引发另一个事件的思考。这可能是该领域中最详尽的学习理论，具有相当强的解释力。例如，控制关联形成条件的规则解释了动物对环境事件之间的相关性的敏感性，因此它们有能力选择性地学习因果相关的事件，而不是那些偶然共同发生的事件。然而，目前有一些现象超出了联想学习模型的范围。例如，在一个标准的联想学习任务中，一个事件，通常是固定的持续时间，后面是一个结果，动物不仅会学习到这个事件后面会有结果（条件反射），而且还会学习到什么时候会有结果--一种被称为时机的效应。然而，尽管这两种类型的学习之间有着密切的关系，条件反射理论和时机理论却各自独立地发展起来。考虑到在相同的学习任务中，时间和条件作用是同时出现的，人们可能会认为这种理论框架从根本上是错误的，近年来，关于时间的信息处理理论已经发展起来，试图解释这两种现象。这种方法在理论上是吝啬的--但在实践中，这些理论解释条件作用的能力是有限的，联想理论的大部分解释力也丧失了。在条件反射理论框架内解释时间的另一种方法相对被忽视了。这个建议的目的是测试一个版本的联想理论，已专门开发来解释时间效应，与发展一个统一的条件反射和时间的联想帐户的最终目标。修正后的联想理论的预测将与那些已被改编来解释条件反射的信息处理帐户进行比较。这将部分通过研究一系列联想学习任务的时间操纵的影响来实现。我们还将使用病变技术。我们用来解释时间的联想理论，也解释了海马损伤所导致的认知障碍，应用它可以对海马动物出现时间和相关缺陷的条件做出一些具体的预测。另一组实验将测试这些预测，作为评估我们的联想理论版本的一种手段。这项工作在若干层面上具有影响。理论上，通过评估修改后的联想理论，我们将评估在何种程度上的联想帐户的时间效应是可行的。实际上，关于时间的研究具有重要的临床意义，因为时间异常与注意力缺陷多动障碍（ADHD）以及精神分裂症和抑郁症密切相关。由于本提案中所描述的工作旨在全面理解时间的性质，这将有助于解释此类异常，并直接为识别其神经基础的尝试提供信息。此外，海马功能的部分将增加我们对这种结构的心理作用的理解，这种结构与阿尔茨海默病等疾病的介导密切相关。因此，我们的工作将在许多领域具有治疗意义。最后，人工智能体学习的研究使用的模型与心理学中使用的模型非常相似。因此，我们提出的实验产生的数据应该有助于发展机器学习理论。

Understanding learning is a central issue in psychology, and has been extensively studied, especially in animal subjects. The theory of associative learning asserts that the mental representations of co-occurring events become linked, such that presentation of one can elicit the thought of the other. This is probably the most elaborated theory of learning in the field, and has considerable explanatory power. For example, the rules governing the conditions under which associations may be formed explains the sensitivity of animals to correlations between environmental events, and hence their ability to learn selectively about events that are causally related, rather than those that co-occur by chance. However, there are some phenomena that are at present outside the scope of associative learning models. For example, in a standard associative learning task an event, which is usually of fixed duration, is followed by an outcome, and the animal will learn not only that the event will be followed by the outcome (conditioning) but when - an effect called timing. Yet despite the intimate relation between these two types of learning, theories of conditioning and of timing have developed independently. Given the fact that timing and conditioning arise alongside each other in the same learning tasks, one might argue that this theoretical framework is fundamentally misconceived, and in more recent years information processing theories of timing have been developed that seek to account for both phenomena. This approach is parsimonious in theory - but in practice, the ability of these such theories to explain conditioning is limited, and much of the explanatory power of associative theory is lost. The alternative approach, to explain timing within the framework of conditioning theory, has been relatively neglected. This proposal aims to test a version of associative theory that has been specifically developed to explain timing effects, with the ultimate objective of developing a unified associative account of conditioning and timing. The predictions of the modified associative theory will be compared with those of an information processing account that has been adapted to explain conditioning. This will be achieved in part by examining the effects of temporal manipulations on a series of associative learning tasks. We will also use lesion techniques. The associative theory we have adapted to explain timing also provides an explanation of some of the impairments in cognition produced by damage to the hippocampus, and applying it results in some specific predictions about the conditions under which timing and related deficits will be found in hippocampal animals. Another set of experiments will test these predictions as a means of evaluating our version of this associative theory. This work has implications at a number of levels. Theoretically, by evaluating the modified associative theory, we will assess the extent to which an associative account of timing effects is feasible. Practically, the work on timing has significant clinical implications, insofar as timing abnormalities are heavily implicated in attention-deficit hyperactivity disorder (ADHD), as well as in schizophrenia and depression. As the work described in this proposal aims at a complete understanding of the nature of timing, this will aid interpretation of such abnormalities, and so directly inform attempts to identify their neural basis. In addition the section looking at hippocampal function will increase our understanding of the psychological role of this structure, which is heavily implicated in mediating conditions such as Alzheimer's disease. Thus our work will have therapeutic implications in a number of areas. Finally, the study of learning in artificial agents uses models very similar to those used in psychology. The data generated by the experiments we propose should thus be of use in helping to develop theories of machine learning.