Humans have a remarkable ability to extract and learn abstract patterns from experience and organise them into mental models that guide behavior. This symposium brings together current research from studies using behavioural, fMRI and intracranial recordings alongside computational models to investigate the mechanisms driving the formation of abstract knowledge structures and asks how these structures might facilitate the learning of new episodes and action representations.

This symposium will start with a talk by Lukas Kunz, who will showcase how representations of objects and places interact during hippocampal ripples to support the formation of associative memory in humans. In a second talk, Irina Barnaveli will then demonstrate how the hippocampal cognitive maps can guide new motor interactions with the world by linking perception to action. The third talk by Andrea Greve then moves the focus to the question how knowledge structures or schema can guide the formation of new episodic memories. The talk will discuss how violations of expectations influence episodic memory, but also how expected events can improve episodic memory. Building on this, Salma Elnagar will address the question how spatial schemas support new episodic learning under varying degrees of uncertainty. This talk will discuss the conditions under which schema can either aid or hinder learning. Finally, Mirko Thalmann will conclude the symposium by presenting behavioural and computational evidence demonstrating how regularities shared between different chunks of information, i.e., motifs, can be used to retrieve sequences from memory.

The hippocampus and its surrounding medial temporal lobe regions combine the diverse features of our experiences into holistic memory representations. Objects and places are key features that the memory system needs to represent to facilitate associative memories for what happened where. In my talk, I will present evidence for single-neuron representations of objects and places in the human medial temporal lobe. I will describe how these representations interact with each other during hippocampal ripples to support associative memory in humans.

Efficiently interacting with the environment requires weighing and selecting among multiple alternative actions based on their associated outcomes. However, the neural mechanisms underlying these processes are still debated. In this talk, I will present evidence that forming relations between arbitrary action-outcome associations involve building a cognitive map. Using a novel immersive virtual reality paradigm, participants learned 2D abstract motor action-outcome associations and later compared action combinations while their brain activity was monitored with fMRI. We observed a hexadirectional modulation of the activity in entorhinal cortex while participants compared different action plans. Furthermore, hippocampal activity scaled with the 2D similarity between outcomes of these action plans. Conversely, the supplementary motor area represented individual actions, showing a stronger response to overlapping action plans. Crucially, the connectivity between hippocampus and supplementary motor area was modulated by the similarity between the action plans, suggesting their complementary roles in action evaluation. I will discuss the process of action evaluation and selection in the context of cognitive mapping and how our findings challenge classical models of memory taxonomy.

Episodic memories are not inscribed on a blank slate, rather encoding is shaped by what we already know. It is widely established that episodic memory is facilitated for events that are congruent with our prior knowledge, or schema. However, superior memory also occurs for novelty or events that conflict with a schema, i.e., cause a prediction error. In this talk I outline a theoretical framework, SLIMM, which predicts that episodic memory is a “U-shaped” function of expectancy, and that the two ends of this U-shape are functionally dissociable, since they depend on different neural systems. I will showcase empirical evidence which uses naturalistic stimuli and taps into well-established pre-experimental schemas. A series of pre-registered studies confirms that memory is better for both highly expected and highly unexpected events compared to neutral events. A second key prediction of SLIMM is that these two ends of the U-shape correspond to different memory processes: recollection of unexpected information and familiarity for expected information. I will explore new insights into the interplay between novelty, expectancy and episodic memory, highlighting theoretical approaches and future directions for understanding how prior expectations shape episodic learning.

Encoding new memories takes place against the backdrop of a rich library of schemas acquired through one’s life. Several studies show that schemas strengthen encoding and accelerate recall of new memories that are in agreement with it (congruent), while others show the opposite pattern where prediction violation facilitates learning. To reconcile the contradictory findings in the literature, the schema-linked interaction between the medial temporal and medial prefrontal regions (SLIMM) model postulates that both highly congruent and highly incongruent information with a schema benefit the process of consolidation due to different learning systems in the brain. However, the SLIMM model remains under scrutiny since empirical evidence is scarce to support its hypotheses. Furthermore, the neural underpinnings of such learning processes remain unknown. While some models suggest a trade-off between the medial prefrontal cortex (mPFC) and the medial temporal lobe (MTL) for congruent and incongruent effects respectively, other models predict an essential role of MTL structures in encoding information congruent to existing knowledge structures. We use behavioural methods and fMRI to understand whether and how the representation of prior knowledge enhance encoding and retrieval of new events. We developed a novel spatial schema paradigm, which compares three conditions of varying degrees of congruency to previous knowledge. Our results demonstrate a mnemonic advantage for congruent events, while incongruent events and those lacking a strong prior schema exhibit a disadvantage, suggesting that reaffirming expectations facilitates learning. In the concurrent fMRI study, we compare systems in the brain that support learning with a prior and without.

Previous work has investigated how transition probabilities and chunks help retrieve sequences from memory. Here, we ask whether people can make use of regularities shared between different chunks, which we call abstract motifs. In two experiments, we empirically tested whether people learn and use two different motifs in a sequence recall task: projectional motifs and variable motifs. Projectional motifs are defined as a common theme shared by sequences of different items. Variable motifs include portions with fixed items and a portion with items that are subject to change. The results show that people use both types of motifs in a transfer test in which sequences are constructed with items not encountered during training. The use of motifs leads to improved recall relative to a baseline, consistent with the predictions of our new computational model of motif learning.