Diverse perceptual biases emerge from Hebbian plasticity in a recurrent neural network model
Schönsberg, F.; Giana, D.; Chopra, Y.; Diamond, M. E.; Goldt, S.
Show abstract
Perceptual biases offer a glimpse into how the brain processes sensory stimuli. While psycho-physics has uncovered systematic biases such as contraction (stored information shifts towards a central tendency), and repulsion (the current percept shifts away from recent percepts), a unifying neural network model for how such seemingly distinct biases emerge from learning is lacking. Here, we show that both contractive and repulsive biases emerge from continuous Hebbian plasticity in a single recurrent neural network. We test the model in four different datasets, two sensory modalities and three experimental paradigms: two working memory tasks, a reference memory task, and a novel "one-back task" that we designed to test the robustness of the model. We find excellent agreement between model predictions and experimental data without fine-tuning the model to any particular paradigm. These results show that apparently contradictory perceptual biases can in fact emerge from a simple local learning rule in a single recurrent region of the brain.
Matching journals
The top 2 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Sampling-based Bayesian inference in recurrent circuits of stochastic spiking neurons 97%
- Spiking attractor model of motor cortex explains modulation of neural and behavioral variability by prior target information 96%
- Taming the chaos gently: a Predictive Alignment learning rule in recurrent neural networks 96%
Similar papers in this journal
- Dynamic control of sequential retrieval speed in networks with heterogeneous learning rules 97%
- Beta oscillations and waves in motor cortex can be accounted for by the interplay of spatially-structured connectivity and fluctuating inputs 97%
- Rate-distortion theory of neural coding and its implications for working memory 97%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.