Interactions Between Working Memory and Long-Term Memory Within and Across Individuals

In computer systems, random access memory (RAM) temporarily stores a limited amount of information for immediate use, whereas the hard disk provides larger, more durable storage. Analogously, human cognition relies on multiple memory systems, including a capacity-limited visual working memory (VWM) that maintains recently encountered information and a visual long-term memory (VLTM) system that supports the storage and retrieval of accumulated knowledge. A central question in cognitive neuroscience concerns how these systems interact within and across individuals. In this dissertation, I integrate behavioral, electroencephalography (EEG), and computational approaches to investigate the relationship between VWM and VLTM. In Chapter 1, I demonstrate that individual differences in working memory predict long-term memory performance when encoding time is sufficient, but this relationship diminishes under rapid presentation conditions (250 ms per item). This finding suggests that the correlation between VWM and VLTM is driven by a shared encoding bottleneck that constrains both systems. In Chapter 2, I characterize the neural mechanisms underlying memory retrieval, showing that both univariate activity and multivariate decoding of EEG signals reveal shared access to representations in VWM and VLTM during the test phase. In Chapter 3, I examine repeated learning and show, across a large sample of participants, that working memory capacity consistently predicts long-term memory performance across repetitions, indicating a stable role of VWM in VLTM acquisition. Finally, in Chapter 4, I investigate how neural representations differ as a function of repetition strength, comparing well-practiced and less-practiced long-term memories, and contrasting both with working memory representations at retrieval. Together, these findings provide converging evidence that interactions between working memory and long-term memory are shaped by a shared encoding bottleneck and overlapping retrieval processes, and that these interactions remain stable across repeated learning. This work advances our understanding of how multiple memory systems coordinate to support human learning and memory.