Ai Research & Innovation

This research helps machines better understand images and language by automatically finding where they make mistakes. Instead of relying on people to spot these errors, it uses smart algorithms that learn from the machines' answers. As a result, it identifies new areas where VLMs struggle, improving their overall performance.

This research investigates how emotions in user questions affect the responses of AI language models. It finds that while emotional prompts often have a small impact on performance, using an adaptive method can improve results. Essentially, tailoring the emotional tone of queries can lead to better AI interactions, especially in social contexts.

This research explores how advanced AI can spot misleading graphics and understand why they might be created. It uses a collection of tweets and examples from a visualization event to test various AI models. The goal is to improve how we identify and interpret deceptive visual information.

This research provides a smarter way for AI to find the best answers when the information is unclear or contradictory. Instead of just picking the most relevant documents, it measures uncertainty to decide which evidence is most helpful. This approach helps make AI more reliable in complex situations.

This research creates an AI that can help medical researchers generate ideas, conduct experiments, and write papers more effectively. It combines insights from medical literature with expert input to ensure that the research is relevant and evidence-based. The AI operates in different modes to suit various research needs and levels of autonomy.