The use of computational models of memory has been effective in adaptive learning environments and in determining the memory capabilities of learners. However, these models have not been widely applied in clinical settings. Evaluation of memory loss still heavily relies on extensive neuropsychological testing performed by neurologists or psychiatrists, especially in the context of progressive neurodegenerative disorders. Current evaluation tools lack the necessary reliability, convenience, and repeatability to effectively capture key dynamics of memory decline, including the unique and changing nature of memory over time. The goal of this study was to predict and monitor memory decline in individuals diagnosed with Mild Cognitive Impairment (MCI) using a model-based adaptive fact learning system. Participants, aged 58 to 78 years, were divided into two groups based on their cognitive classification and completed weekly online learning assessments at home, tracking their individual speed of forgetting (SoF) across various study materials. The results showed that this method was effective in accurately diagnosing mild memory impairment, with a success rate of over 80% after a single 8-minute learning session. The study also demonstrated the model’s ability to distinguish MCI subtypes through computations of participants' SoF. These findings offer novel insights into the progressive nature of memory decline and could have implications for early detection and management of Alzheimer’s disease as well as other forms of dementia and cognitive impairment. Further development of this method could serve as an alternative or complement to established diagnostic procedures and be used in clinical settings.

Recent studies suggest that errors facilitate learning in certain conditions. Despite this, reinforcement paradigms dominate learning methods, subscribing to the narrative that errorless learning is the foundation of an ideal learning environment. If we continue to view learning from this restrictive perspective, we may fail to capture and apply the benefits of errors. Furthermore, although error learning is now a well-documented phenomenon, research on its underlying mechanisms is sparse. Two prominent theories have arisen out of this research; the elaborative hypothesis, in which meaningful connections are derived from errors, and the mediator hypothesis, in which errors act as secondary cues. To go beyond speculation, both must be examined empirically to successfully leverage error learning. Using a combined approach, data from computational models formulated to reflect different mechanisms of error learning were compared to behavioral data. In the behavioral task, participants (N = 61) learned word pairs in either a study or error trial before taking a final test. Supporting past error learning literature, errors before a study opportunity led to better performance on a final test. Differences in reaction times between conditions support the theory that errors increase learning through mediation by acting as a secondary cue rather than as a way to establish a deeper network between the cue and answer. Furthermore, comparing behavioral results to computational cognitive models provided insight into individual differences in mechanisms of error learning.

Memory loss is a debilitating symptom of neurodegenerative diseases. The exact process of memory decline or forgetting in the brain remains unclear. To address this issue, the development of technologies to preserve or improve memory is a continuous objective in translational medicine. This study aimed to uncover the brain networks involved in forgetting and investigate if reducing forgetting was possible through the use of transcranial alternating current stimulation (tACS; 60 Hz or sham) targeted to the dorsolateral prefrontal cortex (dlPFC). The dlPFC was considered a potential target for memory interventions as it had been linked to various aspects of memory function including working memory, executive control, encoding, and retrieval, as well as memory and attention functional connectivity networks. Participants took part in four visits, each consisting of three 8-minute memory tasks using an adaptive fact-learning software. The memory tasks assessed recognition memory (multiple choice), recall memory (fill-in-the-blank), and retrieval learning (response generation). The software used a neurocomputational model that adapted to each participant's performance. This cognitive model is based on established cognitive and biological principles and simulates memory encoding and passive forgetting. The model's α parameter, which represents the speed of forgetting and measures how quickly an individual's memories fade, was used as a dependent variable. Additionally, participants completed two resting state functional MRI (fMRI) scans to evaluate their brain's functional connectivity before and after stimulation. The study compared individual speeds of forgetting to individual patterns of functional connectivity to identify the neural networks most predictive of forgetting, and compared functional connectivity between participants who received tACS and those who received a sham stimulation. We hypothesize that tACS to the dlPFC will decrease forgetting rates and be associated with increases in functional connectivity. In conclusion, tACS has the potential to become a low-cost and non-invasive method to ameliorate memory impairments.