Search Papers | Poster Sessions | All Posters
Poster C39 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Noise Correlations for Feature Learning
Xufeng Caesar Dai1 (), Joonhwa Kim1, Apoorva Bhandari1, Matt Nassar1; 1Brown University
In real-word learning, individuals continually encounter complex arrays of features, only some of which are crucial to the outcomes they experience. How do they manage to discern which combinations of features are relevant for learning? This study explores how dynamic noise correlations – contextually enhanced correlations in neuronal firing – can focus learning on the most relevant feature dimensions in the current context by leveraging prior experience with these features. Participants were tasked with discriminating multi-dimensional perceptual stimuli under various task conditions that specifically incentivized learning about distinct, combined feature dimensions. We found that people learned preferentially in relevant feature dimensions, but to a degree that differed across individuals. These results motivate ongoing work modeling human subject behavior with neural networks and probing noise correlations in feature representations with fMRI. Our approach provides a window into how adaptive neural mechanisms can enhance the efficiency of learning in complex environments.
Keywords: learning decision making noise correlations neural network