Neurocomputational mechanism of real-time distributed learning on social networks

Jiang, Yaomin; Mi, Qingtian; Zhu, Lusha

doi:10.1038/s41593-023-01258-y

Cited by 3 publications

(1 citation statement)

References 60 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Arguably, the most influential computational framework for understanding human behavioral learning is reinforcement learning (RL), which views intelligent behavior as subservient to the maximization of reward (Silver et al, 2021;Sutton & Barto, 2018). Grounded on the principle of reward maximization, the RL framework provides a normative understanding of a spectrum of human learning processes (Daw et al, 2011;Jiang et al, 2023;Niv & Langdon, 2016;Rescorla, 1972;Ribas-Fernandes et al, 2011;Tomov et al, 2021;van Opheusden et al, 2023;Xia & Collins, 2021) and offers theories on the underlying neural mechanisms (Barto, 1995;Montague et al, 1996;Niv, 2009;Schultz et al, 1997). However, the current RL framework provides very limited insights into human representation learning and generalization (Gershman & Daw, 2017;Ho et al, 2022;Mnih et al, 2015;Niv, 2019).…”

Section: Introductionmentioning

confidence: 99%

Learning generalizable representations through efficient coding

Fang¹,

Sims²

2023

Preprint

View full text Add to dashboard Cite

Generalization based on functional similarity is often summarized as acquired equivalence (AE). Acquired equivalence has been ubiquitously documented in humans and animals, but its nature remains not well understood. Here, we interpret the phenomenon by postulating that this generalization reflects the processes of representation compression. We formalize a representation compression framework based on the rate-distortion theory (RD), a branch of information theory that characterizes compression as a fundamental tradeoff between the complexity and accuracy of information processing. A reinforcement learning model derived from the RD theory was able to replicate human functional-similarity-based generalization as well as the behaviors in an extended acquired equivalence experiment paradigm where perceptual similarity was also incorporated. We identified from the model a set of low-level cognitive mechanisms proposed in the current AE theories to underlie the generalization in the acquired equivalence task. We conclude that the representation compression framework provides a unified explanation of human acquired equivalence.

show abstract