Abhisek Keshari

My primary research interests lie at the intersection of Computer Vision, Reinforcement Learning, and Natural Language Processing, with a broader goal of developing intelligent systems that can perceive, reason, and act in complex real-world environments. I am particularly interested in learning representations from high-dimensional, multimodal data and designing models that generalize beyond narrow task settings. By combining perception and decision-making, I aim to contribute to the development of systems that move closer to human-like understanding and adaptability.

In Computer Vision, my interests include visual representation learning, object detection and segmentation, and vision-based reasoning. I am motivated by challenges such as learning from limited labeled data, robustness to distribution shifts, and integrating spatial and temporal information. I am especially interested in how self-supervised and multimodal learning approaches can improve visual understanding by leveraging auxiliary signals such as language or interaction with the environment.

In Reinforcement Learning and Natural Language Processing, I am drawn to problems involving sequential decision-making, language grounding, and human-AI interaction. I am interested in reinforcement learning for long-horizon tasks, sample-efficient learning, and policy generalization. In NLP, my focus includes representation learning, language understanding, and the use of large language models for reasoning and control. I am particularly excited about research that combines vision, language, and reinforcement learning—such as embodied AI, instruction-following agents, and multimodal decision-making systems—where agents learn from both perceptual inputs and natural language guidance.