Representation Learning for Vision & Language

The performance of deep learning models depends upon a strong neural architecture design that learns latent representations from data by using different transformation functions, such as convolutions and fully connected layers. In our lab, we develop new transformations and neural architectures that allow models to learn richer representations effectively across different domains. A special focus of our group is on developing light-weight and power-efficient architectures for edge devices, including mobile phones, with good generalization abilities.

Related Publications