UC San Diego

Human perception and behavior are shaped by past experiences. Neural representations are constrained to utilize statistical regularities to encode the world efficiently while decision making should utilize heuristics to optimize the use of uncertain information. Recently, there has been heightened interest in serial dependence – a feature-specific attraction towards previously seen stimuli – to better understand these distinct objectives. Serial dependence differs from the more familiar perceptual adaptation effects as it is attractive, can be induced by weak stimuli, and can persist for 10s of seconds. Accounts explaining serial dependence have varied greatly, with some positing a low-level perceptual phenomenon and others positing a post-perceptual origin operating during decision-making. That said, most existing studies did not separate the influence of previous stimuli, decisions, and motor actions so directly comparing possible mechanisms is challenging. We first examined the neural underpinning of serial dependence by having participants complete a delayed orientation discrimination task while measuring brain activity with functional magnetic resonance imaging (fMRI, Chapter 1). While behavioral responses indicated an attraction towards the previous stimulus, orientation-specific activation patterns in visual cortex exhibited a repulsive bias. We reconciled these apparently divergent findings with an ideal-observer model in which readout from perceptual areas during decision-making accounts for the attractive biases. We next developed a technique to distinguish stimulus from response generated biases using a simulated observer (Chapter 2). Utilizing this approach, we consistently found that reports were attracted towards previously reported – as opposed to previously presented – stimuli in an orientation report task. Finally, we sought to experimentally disentangle the role of sensory, decisional, and motor contributions to serial dependence (Chapter 3). Through a series of experiments, we found that attraction operates on a perceptual level, unrelated to attention or decisions, as well as on a decisional level, unrelated to motor or sensory experiences. We develop a model in which serial dependence is not the result of processing at a single stage. Instead, all levels of processing are influenced by a canonical prior for stability to optimize the efficiency of neural circuits that contribute to different cognitive operations.