Discrimination and generalisation of temporal information

心理学的一个核心问题是发展一套原则性的学习理论。在一个典型的学习任务中，一个事件之后会有一个结果，动物会知道这个事件预示着结果。联想理论解释说，这两个事件的心理表征是联系在一起的，所以第一个事件的呈现可以引出第二个事件的想法——这个理论可以解释动物是如何选择性地对因果相关的事件之间的预测关系进行编码的，而不是那些偶然发生的事件。但是，尽管联想理论具有解释力，但它不能同时解释动物如何学习结果何时发生（时机），而时机理论是独立发展的。显然，用一种理论来解释这两种学习形式会更简单，时间理论家们已经在尝试建立一个统一的模型。然而，由此产生的模型存在严重缺陷，因为它们无法解释动物如何追踪因果关系。出于这个原因，如果可以采用联想模型来解释时间，那就更好了。这就是当前提案的目的。联想理论无法解释时间的一个原因是，它们没有具体说明如何表示事件的时间属性，比如持续时间。在联想术语中，一个刺激被视为由不同的心理元素组成，对应于它的各种物理特征（如亮度、音调或音量），每个刺激元素都可以进入联想。这项提议的总体目标是测试一个新的假设，即时间也可以被概念化为刺激因素。它将通过提出三个更具体的问题来做到这一点。首先，这些假设的时间刺激元素是否能像更正统的刺激元素一样获得联想强度？第二，它们是否显示出联想性变化的证据——根据联想理论，联想性是刺激的另一个特性，它调节了经验对刺激可以调节的容易程度的影响。第三，我将研究一个被称为感知学习的过程，在这个过程中，暴露于两种刺激会增强区分它们的容易程度，这取决于两种刺激共有的元素的数量。如果我的假设是正确的，那么刺激的时间特征也可以产生这些感知学习效应。这些实验应该允许我们评估联想理论是否可以将时间概念化为另一个刺激因素，就像我所建议的那样。如果可以，那么这就可以形成一个修正的联想理论的基础，这个理论也可以解释时间效应。这项工作在许多层面上都有影响。结果将使我们能够评估时间效应的关联解释在多大程度上是可行的，因此与所有对学习感兴趣的人直接相关。它还将对依赖学习理论的心理学的许多方面以及心理学以外的领域产生影响——例如，研究人工智能体的学习使用的模型与心理学中使用的模型非常相似。因此，提出的实验产生的数据应该对那些发展机器学习理论的人有用。这项工作还具有重要的临床意义，因为时间异常与几种疾病密切相关，如注意力缺陷多动障碍（ADHD）、精神分裂症、抑郁症、帕金森病和阿尔茨海默病。由于本提案中描述的工作旨在更全面地了解时间的本质，这将有助于解释此类疾病的症状，从而增加用于模拟它们产生的认知缺陷的动物模型的复杂性。这将直接为确定其神经基础和开发治疗方法提供信息，并为早期诊断提供潜在的认知测试。

A core issue in psychology is developing a principled theory of learning. In a typical learning task an event is followed by an outcome, and the animal comes to learn that the event predicts the outcome. Associative theory explains this by saying that the mental representations of the two events become linked, so that presentation of the first can elicit the thought of the second - and this theory can explain how animals selectively encode the predictive relationships between events that are causally related, rather than those that co-occur by chance. But despite its explanatory power, associative theory cannot explain how, at the same time, the animal also learns when the outcome will occur (timing), and theories of timing have developed independently. It would clearly be simpler to have one theory to account for both forms of learning, and attempts to develop a unified model have been made by timing theorists. However, the resultant models are seriously flawed, because they are unable explain how animals can track causality. For this reason it would be better if associative models could be adapted to explain timing. This is the aim of the current proposal. One reason associative theories cannot explain timing is because they do not specify how to represent the temporal properties of an event, such as its duration. In associative terms, a stimulus is viewed as consisting of different mental elements, corresponding to its various physical characteristics (such as brightness, pitch or volume), and each stimulus element can enter into associations. The overall aim of this proposal is to test a new hypothesis, that time can also be conceptualised as a stimulus element. It will do so by asking three, more specific questions. First, can these postulated temporal stimulus elements gain associative strength, just like more orthodox stimulus elements? Second, do they show evidence of changes in associability - another property of a stimulus which, according to associative theory, mediates the effect of experience on the ease with which the stimulus can condition? Third, l will examine a process known as perceptual learning, in which exposure to two stimulus enhances the ease with which they can be told apart, and depends on the number of elements the two stimuli share. If my hypothesis is correct, then the temporal features of the stimuli can also produce these perceptual learning effects. These experiments should allow us to evaluate whether associative theory can conceptualise time as another stimulus element, in the way I have suggested. If it can, then this could form the basis of a modified associative theory that can also explain timing effects. This work has implications at a number of levels. The outcome will allow us to assess the extent to which an associative account of timing effects is feasible, and so be of direct relevance to all those interested in learning. It will also have implications for the many aspects of psychology that rely on learning theories, and as well as for areas outside psychology - for example, the study of learning in artificial agents uses models very similar to those used in psychology. Data generated by the experiments proposed should thus be of use to those developing theories of machine learning. The work also has significant clinical implications, as timing abnormalities are heavily implicated in several conditions, such as attention-deficit hyperactivity disorder (ADHD), schizophrenia, depression, Parkinson's and Alzheimer's disease. As the work described in this proposal aims at a more complete understanding of the nature of timing, this will aid interpretation of the symptoms of such diseases, and so increase the sophistication of the animal models used to model the cognitive deficits they produce. This will directly inform attempts to identify their neural basis, and to develop treatments, as well as suggesting potential cognitive tests for earlier diagnosis.