Machine learning aims at building intelligence by modeling patterns in observed training data. It is a powerful tool of artificial intelligence that is used in many other research areas, such computer vision, robotics, audio and speech recognition, biometrics, and mutimedia analytics, to name a few.
Deep neural networks have been trailblazing in multiple artificial intelligent tasks. This development has been enabled by advances in many aspects related to these network, such as building blocks, activation functions, overall architectures, optimization methods, loss functions, and training paradigms. Despite the significant advances, research in all these aspects is not winding down any time soon. They constitute core areas of VIMAL’s research.
Whether it is image, audio, text, or video, the input to a machine learning model has to be put in a numeric representation that is suitable for the task at hand. In the deep learning era, this representation is no longer manually crafted; instead, it is automatically learned by the machine learning model. Influencing this learned representation to encourage desirable properties, such as class-compactness, and discourage unwanted ones, such as confounding factors, constitutes another core area of VIMAL’s research.