Dartmouth Events

Human cognition is characterized by a remarkable ability to transcend the specifics of limited experience to entertain highly general, abstract ideas. Efforts to explain this capacity have long fueled debates between proponents of symbol systems and statistical approaches, most recently embodied by the debate concerning the emergent cognitive capacities of large-scale AI systems. In this talk, I will present the results of several interrelated projects that aim to unify these two traditions, illustrating how many of the key properties of traditional symbol systems can be implemented in deep neural networks, and pointing to a common set of principles that govern abstract reasoning in both human cognition and artificial systems.

I am broadly interested in the question of how the brain extracts structured, abstract representations from noisy, high-dimensional perceptual inputs, and uses these representations to achieve intelligent behavior. To better understand these processes, my work exploits a bidirectional interaction between cognitive science and artificial intelligence, with an emphasis on the visual domain. This involves two major components. First, I use recently developed neural network modeling techniques to build models of higher-order cognitive processes (e.g., metacognition, analogical reasoning) that are grounded in realistic perceptual inputs (images). Second, I take inspiration from cognitive science to design novel inductive biases aimed at imbuing deep learning algorithms with a more human-like capacity for reasoning and abstraction.

Taylor Webb, UCLA Reasoning Lab

Free and open to the public

Sponsored by the Cognitive Science Program