Scott Steinmetz

This study compares supervised and unsupervised neural network models to replicate MSTd's optic flow tuning, finding that autoencoders with MT-like inputs achieve better alignment than accuracy-optimized models.

This paper compares CNNs and NNMF models in modeling optic flow tuning in MSTd neurons, finding that while CNNs are more accurate, NNMF better matches observed tuning properties.

The paper uses a computational model of MSTd to decode self-motion parameters from neural signals, showing that spiral and radial cells can estimate curvilinear path direction, gaze direction, and curvature.

This paper introduces a dynamic efficient sensory encoding mechanism in a neural model of visual motion processing, showing improved accuracy and latency in heading estimation compared to static tuning.