JNP Micro Podcasts: Bidirectional Generative Adversarial Representation Learning for Natural Stimulus Synthesis

In this episode, coauthors Andriy S. Kozlov, Johnny Reilly, and John D. Goodwin discuss their recently published research titled "Bidirectional Generative Adversarial Representation Learning for Natural Stimulus Synthesis." The authors introduce a groundbreaking artificial neural network that generates animal vocalization waveforms and interpolates between them to create new, realistic vocalizations. The team shares how their synthetic stimuli drive auditory cortical neurons in mice just as effectively as natural vocalizations, producing receptive field features with equal predictive power. Tune in as Andriy, Johnny, and John explain the significance of their findings and the potential implications of this innovative technology for sensory neuroscience.

 

Bidirectional generative adversarial representation learning for natural stimulus synthesis

Johnny Reilly, John D. Goodwin, Sihao Lu, and Andriy S. Kozlov

Journal of Neurophysiology 2024 132:4, 1156-1169