BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Humans learn generalizable representations through efficient codin
 g - Daksh Mehta
DTSTART:20251128T140000Z
DTEND:20251128T150000Z
UID:TALK241450@talks.cam.ac.uk
CONTACT:Adam Triabhall
DESCRIPTION:This week we will discuss and debate a very recent preprint by
  Fang and Sims (2025).\n\nAbstract: “Reinforcement learning theory expla
 ins human behavior as driven by the goal of maximizing reward. Conventiona
 l approaches\, however\, offer limited insights into how people generalize
  from past experiences to new situations. Here\, we propose refining the c
 lassical reinforcement learning framework by incorporating an efficient co
 ding principle\, which emphasizes maximizing reward using the simplest nec
 essary representations. This refined framework predicts that intelligent a
 gents\, constrained by simpler representations\, will inevitably: 1) disti
 ll environmental stimuli into fewer\, abstract internal states\, and 2) de
 tect and utilize rewarding environmental features. Consequently\, complex 
 stimuli are mapped to compact representations\, forming the foundation for
  generalization. We tested this idea in two experiments that examined huma
 n generalization. Our findings reveal that while conventional models fall 
 short in generalization\, models incorporating efficient coding achieve hu
 man-level performance. We argue that the classical RL objective\, augmente
 d with efficient coding\, represents a more comprehensive computational fr
 amework for understanding human behavior in both learning and generalizati
 on” (Fang & Sims\, 2025).\n\nReference: Fang\, Z.\, & Sims\, C. R. (2025
 ). Humans learn generalizable representations through efficient coding. Na
 ture Communications\, 16(1)\, Article 3989. https://doi.org/10.1038/s41467
 -025-58848-6
LOCATION:https://cam-ac-uk.zoom.us/j/92612577704?pwd=MUtqMjVQdXNmUTVIYjRkM
 G1NUW9GZz09
END:VEVENT
END:VCALENDAR