Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Estimating the Synaptic Efficacies of the Drosophila Optical Lobe Full Connectome with Predictive Coding

Rintaro Kai¹, Keisuke Toyoda¹, Naoya Nishiura², Masataka Watanabe; ¹The University of Tokyo, Tokyo Institute of Technology, ²The University of Tokyo

Presenter: Rintaro Kai

The neural circuitry of the drosophila brain is moderately complex and has long served as a model in neuroscience research; however, many existing models of the Drosophila brain rely on extreme simplifications or biologically implausible assumptions. Recently, Drosophila has gained further attention because the drosophila full connectome has been revealed. In this study, we constructed a biologically plausible autoencoder for visual processing using the complete connectome of the drosophila brain. By computing prediction errors between two anatomically closely related visual neurons, we implemented predictive coding. We also built an autoencoder with a randomly initialized connectivity matrix and found that it is harder to train than the model initialized with the real connectome. These findings suggest that the original connectome already has mechanisms akin to predictive coding. We hope that, in the future, initializing models with real connectome data will show biological characteristics that would not be shown in other initializations.

Topic Area: Predictive Processing & Cognitive Control

Extended Abstract: Full Text PDF