A Model of Semantic Completion in Generative Episodic Memory

A neural network model for online one-shot storage of pattern sequences

Based on the CRISP theory (Content Representation, Intrinsic Sequences, and Pattern completion), we present a computational model of the hippocampus that allows for online one-shot storage of pattern sequences without the need for a consolidation process. In our model, CA3 provides a pre-trained sequence that is hetero-associated with the input sequence, rather than storing a sequence in CA3. That is, plasticity on a short timescale only occurs in the incoming and outgoing connections of CA3, not in its recurrent connections. We use a single learning rule named Hebbian descent to train all plastic synapses in the network. A forgetting mechanism in the learning rule allows the network to continuously store new patterns while forgetting those stored earlier. We find that a single cue pattern can reliably trigger the retrieval of sequences, even when cues are noisy or missing information. Furthermore, pattern separation in subregion DG is necessary when sequences contain correlated patterns. Besides artificially generated input sequences, the model works with sequences of handwritten digits and natural images. Notably, our model is capable of improving itself without external input, in a process that can be referred to as 'replay' or 'offline-learning', which helps in improving the associations and consolidating the learned patterns.

A generative model of memory construction and consolidation

Episodic memories are (re)constructed, share neural substrates with imagination, combine unique features with schema-based predictions and show schema-based distortions that increase with consolidation. Here we present a computational model in which hippocampal replay (from an autoassociative network) trains generative models (variational autoencoders) to (re)create sensory experiences from latent variable representations in entorhinal, medial prefrontal and anterolateral temporal cortices via the hippocampal formation. Simulations show effects of memory age and hippocampal lesions in agreement with previous models, but also provide mechanisms for semantic memory, imagination, episodic future thinking, relational inference and schema-based distortions including boundary extension. The model explains how unique sensory and predictable conceptual elements of memories are stored and reconstructed by efficiently combining both hippocampal and neocortical systems, optimizing the use of limited hippocampal storage for new and unusual information. Overall, we believe hippocampal replay training generative models provides a comprehensive account of memory construction, imagination and consolidation.

Prediction error and memory across the lifespan

The influence of Prediction Errors (PEs) on episodic memory has generated growing empirical and theoretical interest. This review explores how the relationship between PE and memory may evolve throughout lifespan. Drawing upon the predictive processing framework and the Predictive, Interactive Multiple Memory System (PIMMS) model in particular, the paper highlights the hierarchical organization of memory systems and the interaction between top-down predictions and bottom-up sensory input, proposing that PEs promote synaptic change and improve encoding and consolidation processes. We discuss the neuroscientific mechanisms underlying PE-driven memory enhancement, focusing on the involvement of the hippocampus, the entorhinal cortex-hippocampus pathway, and the noradrenergic sympathetic system. Recognizing the divergent trajectories of episodic and semantic memory across the lifespan is crucial when examining the effects of PEs on memory. This review underscores the heterogeneity of memory processes and neurocognitive mechanisms underlying PE-driven memory enhancement across age. Future research is suggested to directly compare neural networks involved in learning from PEs across different age groups and to contribute to a deeper understanding of PE-driven learning across age.

EXPRESS: Where was the Toaster? A systematic investigation of semantic construction in a new virtual episodic memory paradigm

Retrieved memories of past events are often inaccurate. The scenario construction model (SCM) postulates that during encoding, only the gist of an episode is stored in the episodic memory trace and during retrieval, information missing from that trace is constructed from semantic information. The current study aimed to find behavioural evidence for semantic construction in a realistic, yet controlled setting by introducing a new paradigm and adjusted memory tests that measure semantic construction. Using a desktop virtual reality (VR) participants navigated through a flat in which some household objects appeared in unexpected rooms, creating conflicts between the experienced episode and semantic expectations. The manipulation of congruence enabled us to identify influences from semantic information in cases of episodic memory failure. Besides we controlled for objects to be task-relevant or -irrelevant to the sequence of action. In addition to an established old/new recognition task we introduced spatial and temporal recall measures as possible superior memory measures quantifying semantic construction. The recognition task and the spatial recall revealed, that both congruence and task-relevance predicted correct episodic memory retrieval. In cases of episodic memory failure semantic construction was more likely than guessing and occurred more frequently for task-irrelevant objects. In the temporal recall object-pairs belonging to the same semantic room-category were temporally clustered together compared to object-pairs from different semantic categories (at the second retrieval). Taken together, our findings support the predictions of the SCM. The new VR-paradigm, including the new memory measures appears to be a promising tool for investigating semantic construction.