• B.A Philosophy, honors
  Advisor: Peter Carruthers


• B.Sc Psychology, honors
  Advisors: Michael Dougherty and Donald Bolger

• Intramural Research Training Award
  Supervisor: Dr. Judith Rapoport

This work presents a novel computational framework and supporting human empirical evidence (behavioral and neural) that the brain discriminates memories not by expanding representational space, as long assumed, but by compressing it. The ability to tell whether an experience is new or familiar lies at the core of perception and memory. For decades, theories of memory discrimination have rested on the Marr-Albus principle of dimensionality expansion, instantiated by an expanded encoding ensemble (i.e. expanded dimensionality) and a sparse firing code. Our work provides behavioral, computational, and neural evidence that dimensionality reduction, not expansion, underlies the capacity to tell similar experiences apart. We show that neural activity in the hippocampus and neocortex reflects this compaction of representational space. Beyond resolving a long-standing paradox in memory theory, our results have broad implications. They link efficient coding in the brain to modern machine learning and computer vision architectures, connect compression to changes in memory systems with development and aging, and provide a computational link between perceptual categorization, novelty detection, and one-shot episodic encoding. The concept that selective information loss can improve memory discrimination reframes how we understand the adaptive nature of memory, and opens a path towards unified theories of biological and artificial intelligence.

How do we distill a "blooming, buzzing confusion" into what matters? Information compression is an optimization process that determines what is relevant by constructing compact representations which strategically discard redundant details while retaining essential properties for a given task. For example, image compression optimizes for efficient storage by removing visual details imperceptible to humans. Compression has been hypothesized across perception, memory, learning, attention, decision-making, social cognition, and emotion, but its neural bases are not understood. I developed a dynamical systems theory explaining that brain regions better compress inputs when embedded in highly connected, recurrent anatomical networks, letting signals circulate more efficiently so fewer signals are needed. Our framework adapts the mathematics of information theory to understand the limits of sending and receiving packets of information with individually varying speed and reliability across white matter pathways of differing integrity. Longer pathways distort information flow, so brain regions with higher transmission fidelity send and receive packets with greater rate and reliability as a function of network topology prioritizing shortest paths. Our model parsimoniously explains communication as a function of network complexity, how highly connected hub regions integrate information, and the speed and accuracy of behavior.

"Being stuck in a hole" is a clinical metaphor that is used to convey the pull of perseverative thought in depression and anxiety. The current paper makes this clinical metaphor for uncontrollable negative thinking more concrete through the mathematical frameworks of dynamical systems and control theory. The persistent, "sticky" quality of perseverative thought may emerge from brain activity that evolves in a low-energy attractor basin that is easy to enter and stay within, but difficult to leave. In perseverative thought, especially rumination, deepened attractors are associated with persistent neural activity and heightened symptom severity. The pull of these attractors could help explain why depressed and anxious individuals must exert so much effort to disengage from negative thoughts.

Practitioners of mindfulness effortfully de-automatize habitual reactions and center the present moment. We investigated how to understand these defining components of mindful experience using formal theory and associated computational metrics of brain network structure and function. In an experiment, we instructed college students to mindfully regulate their responses to alcohol in an fMRI scanner. Analyzing their neural activity, we identified changes in brain function during mindfulness. These changes in brain function characterize how individuals exert effort to de-automatize natural reactions to alcohol. We found associated brain metrics that predicted how likely college students were to drink alcohol in the future. Analyzing how their brain activity naturally changes over time, we found that brain regions that used more effort tended to renew activity more quickly. Since renewing activity is updating the present, the brain may center the present moment when it uses more effort. Our findings advance our understanding of brain network processes that characterize elusive subjective experiences during mindfulness. The general framework can be applied to study other forms of meditation and attention, and when processes of attention go awry in psychiatric disorders involving worry, anxiety, and rumination.

Curiosity is a fundamental aspect of human nature, yet most studies rely on small, Western samples. Analyzing 482,760 Wikipedia readers across 50 countries, this study replicates two known curiosity styles—"busybody" and "hunter"—and reveals a third, the "dancer," characterized by creative exploration. By linking these styles to global indicators like education and well-being, the findings shed light on how digital knowledge-seeking behaviors are shaped by cultural and geographical factors.

Coverage in Nature briefing, Science podcast, and Penn Today article.

Throughout life, we might seek a calling, companions, skills, entertainment, truth, self-knowledge, beauty, and edification. In this review, we describe how the practice of curiosity can be viewed as an extended and open-ended search for valuable information with hidden identity and location in a complex space of interconnected information. We propose how a computational model of efficient search can be used to bridge curiosity, cognitive maps, and model-based reinforcement learning.

Description from the gallery: '"Dyslexic Dictionary" is a provocation — an invitation to explore and redefine. Nine dyslexic artists have been commissioned to produce a creative response to words, books, poems and even the alphabet itself, highlighting how their minds experience language. Participating artists include Governor Gavin Newsom, Gil Gershoni, Adeniyi Akingbade, Christian Boer, Adam Eli Feibelman, Sally Gardner, Martin Grasser, Rebecca Kamen and Kelsey Ann Kasom.'

Sparking Curiosity depicts the network dynamics of Rebecca Kamen's art process. This network was created from transcribed interviews, where each circle represents a word describing an idea that shaped her artistic journey; and each line indicates how similar those words are to each other. The colors represent communities of ideas that are more alike. Every frame of the video summarizes the spark of hundreds of ideas and their immediate neighbors in the same order as spoken during the interview.