MEMP PhD Thesis Defense: Morgan Talbot

Emulating and Enhancing Human Visual Perception and Learning with Image-Computable Models

The convergence of AI and neuroscience has produced powerful image-computable models that encode visual stimuli in ways that predict human brain activity and behavioral responses. Recent work has explored whether predictive AI can be inverted to modulate human neural activity and behavior: what is the ideal stimulus to provide to a model, and ultimately to a human, in order to elicit a desired outcome? For example, can we use models to generate images that induce more accurate human judgments, or organize stimuli into sequences that catalyze more effective learning? However, standard AI models fundamentally diverge from human perception and cognition, rendering stimuli produced by inverting these models unintelligible and therefore ineffective for humans. This thesis explores strategies for functionally aligning AI with human constraints and capabilities in order to augment human visual perception and learning. I show that models aligned with human perceptual robustness can accurately predict image difficulty and can be inverted to generate stimuli that accelerate visual learning. I validate these techniques in a real-world clinical training setting with first-year pathology residents. Finally, I close the loop between perception and learning by casting AI models as “surrogate learners” that simulate human learning dynamics. By inverting these simulations, I optimize instructional curricula that significantly enhance human performance in a controlled experiment. Collectively, these findings suggest that the fidelity of a model's alignment with human cognition is the key determinant of its utility as an educational tool, establishing a new paradigm of model-driven learning optimization.

Thesis Supervisor:
Gabriel Kreiman, PhD
Professor of Ophthalmology, Boston Children's Hospital and Harvard Medical School

Thesis Committee Chair:
James J. DiCarlo, MD, PhD 
Peter de Florez Professor of Neuroscience, Department of Brain and Cognitive Sciences, MIT
Director, MIT Quest for Intelligence

Thesis Reader:
Richard N. Mitchell, MD, PhD
Professor of Pathology, Brigham and Women's Hospital and Harvard Medical School
Associate Director, Harvard-MIT Health Sciences and Technology

________________________________________________________________________________________

Zoom Invitation
Morgan Talbot is inviting you to a scheduled Zoom meeting

Topic: Morgan Talbot MEMP PhD Thesis Defense
Time: Friday, February 13, 2026, 1:00 PM Eastern Time (US and Canada)

Your participation is important to us: please notify hst [at] mit.edu (hst[at]mit[dot]edu), at least 3 business days in advance, if you require accommodations in order to access this event.

Join Zoom Meeting
https://harvard.zoom.us/j/2726278568

Password: inverse

Join by telephone (use any number to dial in)
        +1 646 931 3860
        +1 929 436 2866
        +1 301 715 8592
        +1 305 224 1968
        +1 309 205 3325
        +1 312 626 6799
        +1 689 278 1000
        +1 719 359 4580
        +1 253 205 0468
        +1 253 215 8782
        +1 346 248 7799
        +1 360 209 5623
        +1 386 347 5053
        +1 507 473 4847
        +1 564 217 2000
        +1 669 444 9171
        +1 669 900 6833

International numbers available: https://harvard.zoom.us/u/afZQnLUjG

One tap mobile
+16469313860,,2726278568# US (Password: 5503527)

Meeting ID: 272 627 8568 

Join by SIP conference room system
2726278568.5503527 [at] zoomcrc.com (2726278568[dot]5503527[at]zoomcrc[dot]com)