Humans have a remarkable ability to sense visual stimuli, adapt decisions and respond to rapid changes in the environment. This ability is thought to rely on learning-dependent plasticity in neural circuits. While some visual behaviors are learned in a supervised manner through explicit experience, some others are innate and develop in an unsupervised manner. In this thesis, we investigated neuronal representations of behavior formed through both supervised (visual categorization) and unsupervised learning (novel/familiar recognition). The central goal is to understand the neural activity and dynamics underlying visual categorization and learning in the dorsal and ventral visual streams of the cortical hierarchy. Our experimental approach involved electrophysiological recordings from populations of neurons in awake, behaving monkeys that were trained to perform complex, cognitive tasks. First, we investigated how flexible, categorical decisions are mediated by the dorsal stream (specifically, the lateral intraparietal cortex (LIP) in PPC). Prior studies have than shown that LIP encodes abstract learned visual categories of motion directions learned over months of supervised training. Visual motion categorization has only been studied in the context of the delayed match-to-category task with specific, fixed task-demands such as delayed decisions and match/non-match computations. So, we do not know whether LIP's role in categorization with stable, category rules generalizes to more flexible tasks that require animals to rapidly adapt their decisions to changes in context. To explore whether LIP plays a generalized role in category-based decision making, we developed a task-switching paradigm in which monkeys categorized the same random-dot motion stimuli in two different tasks. The tasks differed in the specific behavioral demands involved in making decisions, the effector used for reporting the decision (eye vs. hand movement) and in the sequence of task events upon which decisions were based. We found that neural populations showed remarkably similar categorical encoding in both tasks, with greater category selectivity in the task with memory and matching functions. These findings demonstrate that parietal neural circuits contribute to flexible task switching by dynamically changing information coding in a task-dependent manner. Second, we examined how unsupervised learning is mediated in the ventral visual pathway (area IT) as monkeys learned to recognize and become familiar with novel visual stimuli. Humans and other primates have an exceptional ability to learn to recognize familiar visual items, and to discriminate them from novel items. This ability is thought to rely on experience dependent changes in neuronal representations in the ventral visual pathway, particularly IT cortex. However, how neuronal familiarity encoding develops during learning and the link between neuronal familiarity effects and behavior is not well understood. To date, experimental studies have primarily sampled from the extremities of the familiarity spectrum, comparing highly familiar to highly novel images without capturing the dynamics in between. To probe this gap in knowledge, we familiarized rhesus macaques with multiple sets of novel images for up to 8 days each, while monitoring responses of IT neuronal ensembles during the familiarization process - both over long and short time scales. We found that familiarity encoding due to learning in IT cortex develops slowly over the course of several days, and familiarity-related changes in neuronal representations are minimal over short time scales of a single recording session. Our results give insight into how experience affects visual representations in IT across multiple timescales. Together, our results provide insights into how the primate brain learns to represent flexible behavior in the dorsal and ventral visual hierarchy. In the dorsal visual stream, our results demonstrate the task-dependence and flexibility of visual, category, and choice encoding in LIP, and reveal distinct roles for LIP in the two categorization tasks in the task switching paradigm. Through detailed comparisons between fundamentally similar categorization tasks with tractable differences, our contributions provide compelling evidence that dynamical formatting of relevant information provides a mechanistic basis for LIP to govern flexible, categorical decisions in the dorsal visual stream. In the ventral visual stream, our results demonstrate that familiarity-dependent plasticity in neural activity develops slowly over the course of several days, with a modest impact of familiarity on representations within a single recording session. Our results suggest that experience dependent plasticity occurs at multiple timescales in IT object representations.



Downloads Statistics