| Summary: | <p>To structure knowledge efficiently, we need to acquire and store it in a way that retains the relations between different pieces of content. A cognitive map is a way of organising knowledge in a relational manner that provides us with behavioural flexibility when goal demands change. If we needed a new cognitive map for every situation we encounter, however, this would be very computationally expensive. In Chapter 2, we investigate how humans share or reuse relational maps across different contexts when it is adaptive to do so and how this process is encoded neurally. Using a novel task in which the transition function is shared across one set of contexts but not the other, we find that the human medial temporal lobe encodes transition probabilities in an efficient way by representing them more similarly when they can be shared across contexts.</p>
<p>But what factors affect how well and robustly a novel cognitive map is learned in the first place? In Chapters 3 and 4 of this thesis we manipulate the training curriculum used during novel cognitive map learning and aim to elucidate the effect of training curriculum on map learning behaviourally (Chapter 3) and in terms of neural representations (Chapter 4). We investigate map learning using maps consisting of discrete images. We find that the spatial sampling of transitions during training affects participants’ retention as well as multi-step inference ability. Spatially disjoint sampling of transitions within a block of training (i.e. experiencing transitions from across the map) was beneficial compared to random walks along rows or columns of the map (trajectory training). We hypothesise that this is due to better spatial grounding in the disjoint condition as well as weaker encoding and compression of the map into separate rows and columns in the trajectory condition. In Chapter 4, we find evidence for the latter, with trajectory training resulting in neural encoding of the map that is more similar to separate transitive lines along rows and columns, the extent of which is related to behavioural accuracy. It therefore appears that the order in which we experience parts of a map during initial exposure is crucial for not only map retention but also the nature of the resulting map.</p>
|
|---|