Improving Artificial Intelligence to Recognize Shapes | CPT PPP Coverage

Cryptopolytech (CPT) Public Press Pass (PPP)
News of the Day COVERAGE

200000048 – World Newser
•| #World |•| #Online |•| #Media |•| #Outlet |

View more Headlines & Breaking News here, as covered by cryptopolytech.com

Artificial intelligence (AI) has the potential to accomplish incredible tasks, but it has a major limitation – it can’t recognize all shapes. Miaomiao Zhang, an assistant professor at the University of Virginia, is determined to address this blind spot in AI.

The current machine-learning models are not capable of analyzing and quantifying the shape of objects in images with complex structures and variations, especially in the context of medical imaging. These models are biased towards recognizing image textures and struggle to perceive an object’s global structures. For example, existing neural networks can be deceived into misidentifying an image of a cat with an elephant’s skin as an elephant, focusing only on local image details rather than recognizing the global shape.

This limitation is particularly important for Zhang, who specializes in developing AI for image processing and analysis in medical applications. AI tools, such as machine learning and deep neural networks, play a critical role in diagnosing progressive diseases like heart failure and Alzheimer’s. If AI cannot accurately recognize and quantify shape changes in medical images, it becomes challenging to detect abnormalities or subtle changes caused by diseases over time.

Zhang’s project, titled “Deep Neural Networks That Can See Shape From Images: Models, Algorithms, and Applications,” aims to create a new family of AI tools that can analyze and quantify shape information from images. The goal is to improve image analysis and machine learning in general, unlocking the full potential of AI in medical imaging. This will enable more accurate and reliable diagnoses, empowering healthcare professionals with enhanced insights for improved patient care.

To address AI’s limitations in shape analysis, Zhang has been granted a prestigious CAREER Award from the National Science Foundation. This award recognizes her potential for leadership in research and education. Zhang’s work has also been recognized by the National Institutes of Health and the Medical Image Computing and Computer Assisted Intervention Society.

Zhang’s research focuses on developing deep shape-based neural networks (DSNNs) that can effectively analyze shapes. This requires shifting how researchers teach the networks to see and analyze shapes, similar to teaching a child new words. The research will involve training the networks to analyze deformed variants of simple geometric figures using toy datasets. As the model gains intelligence, more complex images, including thousands of MRI and computed tomography images from clinical colleagues and research databases, will be incorporated.

This research will have a wide-ranging impact beyond medical imaging. Over the course of the project, Zhang’s team will develop software that enhances AI’s ability to recognize and understand shapes in various applications.