Emergent perceptual biases from state-space geometry in trained spiking recurrent neural networks

• Published in: Cell reports (Cambridge) Vol. 43; no. 7; p. 114412
• Main Authors: Serrano-Fernández, Luis, Beirán, Manuel, Parga, Néstor
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 23.07.2024
• Subjects: Action Potentials - physiology; Animals; Bayes Theorem; Bayesian computations; decision-making; dynamical systems; Memory, Short-Term - physiology; Models, Neurological; Nerve Net - physiology; Neural Networks, Computer; Neurons - physiology; perceptual biases; recurrent neural networks; state-space geometry; dynamical systems; state-space geometry; perceptual biases; recurrent neural networks; decision-making; CP: Neuroscience; Bayesian computations
• DOI: 10.1016/j.celrep.2024.114412
• ISSN: 2211-1247

A stimulus held in working memory is perceived as contracted toward the average stimulus. This contraction bias has been extensively studied in psychophysics, but little is known about its origin from neural activity. By training recurrent networks of spiking neurons to discriminate temporal intervals, we explored the causes of this bias and how behavior relates to population firing activity. We found that the trained networks exhibited animal-like behavior. Various geometric features of neural trajectories in state space encoded warped representations of the durations of the first interval modulated by sensory history. Formulating a normative model, we showed that these representations conveyed a Bayesian estimate of the interval durations, thus relating activity and behavior. Importantly, our findings demonstrate that Bayesian computations already occur during the sensory phase of the first stimulus and persist throughout its maintenance in working memory, until the time of stimulus comparison. [Display omitted] •Perception is influenced by sensory history, short-term memory, and long-term plasticity•Recurrent neural networks of spiking neurons trained to discern temporal intervals•Geometric properties of neural activity encoded distorted representations of first interval•These representations conveyed a Bayesian estimation of first stimulus In delayed comparison tasks, contraction bias compresses stimuli in working memory toward their mean. Trained recurrent networks of spiking neurons exhibit this effect, encoding distorted representations of first intervals, influenced by sensory history. Serrano-Fernández et al. link these representations to Bayesian estimation, connecting neural activity with behavior.