Abstract
Recent human behavioral and neuroimaging results suggest that people are selective in when they encode and retrieve episodic memories. To explain these findings, we trained a memory-augmented neural network to use its episodic memory to support prediction of upcoming states in an environment where past situations sometimes reoccur. We found that the network learned to retrieve selectively as a function of several factors, including its uncertainty about the upcoming state. Additionally, we found that selectively encoding episodic memories at the end of an event (but not mid-event) led to better subsequent prediction performance. In all of these cases, the benefits of selective retrieval and encoding can be explained in terms of reducing the risk of retrieving irrelevant memories. Overall, these modeling results provide a resource-rational account of why episodic retrieval and encoding should be selective and lead to several testable predictions.
Highlights
In a natural setting, when should an intelligent agent encode and retrieve episodic memories? For example, suppose I am viewing the BBC television series Sherlock
Most of what we know about episodic memory has, by design, come from experiments where performance depends primarily on episodic memory, and participants are given clear instructions about when episodic memories should be stored and retrieved; likewise, most computational models of human memory have focused on explaining findings from these kinds of experiments
Our approach is built on the principle of resource rationality, whereby human cognition is viewed as an approximately-optimal solution to the learning challenges posed by the environment, subject to constraints imposed by our cognitive architecture (Griffiths et al, 2015; Lieder and Griffiths, 2019); according to this principle, the approximately-optimal solutions obtained by our model can be viewed as hypotheses about how humans use episodic memory in complex, real-world tasks
Summary
In a natural setting, when should an intelligent agent encode and retrieve episodic memories? For example, suppose I am viewing the BBC television series Sherlock. A typical episodic memory experiment could ask participants to remember a set of random wordpairs; later on, given a word-cue, the participants need to report the associated word (Kahana, 2012) In this kind of word-pair experiment, the optimal timing for encoding and retrieval is clear: The participant should encode an episodic memory when they study a word-pair and retrieve the associate when they are prompted by a cue. There has been increasing interest in using naturalistic stimuli such as movies or audio narratives in psychological experiments, to complement results from traditional experiments using simple and well-controlled stimuli (Sonkusare et al, 2019; Nastase et al, 2020) These experiments have the potential to shed light on when encoding and retrieval take place during event perception in a naturalistic context, where no one is explicitly instructing participants about how to use episodic memory. Selectivity effects can be observed in the realm of more traditional list-learning studies – for example, there is extensive behavioral and neuroscientific evidence that stimuli that trigger strong prediction errors are preferentially encoded into episodic memory (for reviews, see Frank and Kafkas, 2021; Quent et al., 2021b)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
