Abstract: Due to the availability of data in multiple domains, matching across domains has become an important area of research, with several applications in surveillance, e-commerce, etc. Also, since new categories of data are constantly being discovered, it is important to account for the previously unseen data. First, we will discuss how we can develop generalized models which can handle both unseen domains and classes during testing. The second part focusses on how the models can be incrementally updated to incorporate the information of the new classes. For the third part, we will look at the challenging task when the model encounters changing domains during testing. Finally, we will discuss the future research directions in this exciting area.

Abstract: Artificial Intelligence (AI) intricately weaves throughout the fabric of modern society. From judiciaries to policing, from commerce to education, from airports to our private study rooms-- our society is becoming more and more dependent on AI and its seamless functioning. However, with such a far reaching effect it also can potentially bring in some inadvertent consequences in the form of unfairness and biases toward different sectors of the population. Hence a narrow outlook of AI, through the lenses of technicality, is no longer enough. To better understand the root of these inadvertent consequences and mitigate them, a new research paradigm which has gained momentum recently is AI and Ethics. In this tutorial, we attempt to provide an in-depth overview, through case studies and practical examples, of the many ways to operationalize Ethics in Artificial Intelligence. We provide both theoretical definitions as well as practical case studies through peer-reviewed publications that study bias and discrimination in various domains ranging from vision to speech and information retrieval. The goal of the tutorial is to prepare the audience to become better at auditing platforms for identifying biases and designing mitigation measures to rid the systems of these biases.

Abstract: In this tutorial we will focus on the biological complexity of neural hardware (neurons and synapses) and brain activity which have important computational consequences and form the neural basis of brain dysfunction. We will start the tutorial with mathematical models of biological neurons, connections (synapses) and networks. Next, we will focus on how we can use the spiking neurons to construct network that can do computing for example track an object. At this point we will briefly delve into the synaptic plasticity rules. Finally, we will discuss how changes in the neural hardware give rise to brain activity which then manifests as a certain brain disorder.

Abstract: Whether it be recommending songs to fit our current mood or teaching our personal robots to find the bunch of keys that we keep misplacing in our homes, Reinforcement Learning (RL) is being used to develop new applications that can make our lives easier and more enjoyable. RL agents explore large state-action spaces to discover the most useful sequence of actions to accomplish a given task. However, there are a number of challenges to applying RL in real-world settings. These include sample efficiency, safety, and generalization. This tutorial will discuss these challenges and introduce approaches to address them. We will introduce the traditional approach to training RL agents in simulation followed by sim-to-real transfer, and discuss the limitations of this approach. We will also explore how the challenges of sim-to-real transfer can be circumvented by safe and sample-efficient exploration directly in the real world. The tutorial will cover approaches to make RL in the real world feasible, such as modular design, exploration in a high-level action space, object-oriented representation of the world, model-based planners that leverage strong inductive biases, and utilization of prior knowledge and common sense from large language models. We will also study the important problem of domain generalization and how it can be addressed in a sample-efficient manner. The topics discussed will be motivated by applications in recommendation systems, advertising, robotics, and social welfare distribution planning.