Abstract
How does the brain learn to recognize objects? Although researchers have debated the origins of object recognition for decades, it had not been possible to examine how high-level visual abilities emerge in the newborn brain with high precision. To overcome this barrier, my lab developed an automated controlled-rearing method with a newborn animal model—the domestic chick. This method can be used to measure the development of newborn object recognition for extended periods of time in strictly controlled virtual environments. First, I describe controlled-rearing experiments demonstrating that newborn chicks have advanced object recognition abilities, including view-invariant recognition and background-invariant recognition. These abilities develop rapidly (within the first few days of life) and can emerge from sparse input (e.g., view-invariant object recognition can develop from a single view of an object). These findings indicate that newborn visual systems can be highly generative at the onset of vision. Next, I describe controlled-rearing experiments characterizing the role of visual experience in the development of object recognition. We have discovered that the development of object recognition requires at least four types of visual experience: (1) experience with slowly moving objects, (2) experience with smoothly moving objects, (3) experience with objects containing surface features, and (4) experience with objects on natural backgrounds. When chicks are reared in environments that lack these experiences, the chicks fail to develop accurate object recognition abilities. Finally, I describe how these controlled-rearing data can be linked to models of visual cortex for characterizing the computations and learning rules underlying newborn object recognition. I conclude that controlled rearing can serve as a critical tool for testing between different theories and models in the vision science and computational neuroscience communities.
Meeting abstract presented at VSS 2017