From Short-Term Tolerance to Long-Term Recognition in Human Visual Memory

Humans have a remarkable ability to recognize visual objects following limited exposure and despite changes at the image-level. How humans acquire this ability remains a mystery, and it remains one area in which artificial intelligence has yet to match human performance. I sought to understand this fundamental cognitive ability by leveraging theories and methods from multiple fields. In particular, I examined how rules guiding the perception of objects in visual working memory assist in the construction of visual long-term memories. In four experiments, I reveal that our expectations for how objects move in the world are used to learn and integrate object information into visual long-term memory. Next, I further examined how aspects of memory over the short-term may actually be features used to construct appropriately constrained representations in the long-term. I demonstrate in three experiments that visual working memory is highly tolerant to variability at test, in order to act as a venue to integrate information into long-term memory. Finally, I moved past investigating memory following singular experiences to understand how our memories change over repeated exposures. I discovered across three experiments that the initial quality of an experience, as well the amount of time between repeated encounters, affects our ability to integrate and remember objects we encounter multiple times. This work contributes to our understanding of the growth process of visual memory, and attempts to form bridges between traditionally disparate fields of vision scientist studying object perception, neuroscientists studying long-term memory, and engineers designing artificial intelligence recognition systems