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

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Unsupervised Future-Predictive Learning in the Connectome-Constrained Drosophila Optical Lobe

Naoya Nishiura¹, Keisuke Toyoda², Masataka Watanabe; ¹The University of Tokyo, ²The University of Tokyo, Tokyo Institute of Technology

Presenter: Naoya Nishiura

Recent advances in connectome-based modeling of the fruit fly brain have enabled neural network architectures that mirror the anatomical wiring of the optic lobe. Despite the success of supervised approaches in predicting neural activity and performing optic flow estimation, many natural settings lack explicit labels such as motion vectors. In this work, we propose an unsupervised learning strategy in which the input layer (R1--8) receives feedback signals to predict future visual inputs $I_{t + \Delta t}$ without any external label. We show that this approach partially reproduces ON/OFF direction selectivity in T4/T5 neurons, a hallmark of the fly visual system.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF