Minimally supervised anomaly detection for imaging data
A novel approach to anomaly detection using minimally supervised learning, capable of distinguishing normal from abnormal imaging data without extensive human annotation, and incorporating sparse human input when available.
Overview
This solution offers a cutting-edge approach to anomaly detection in imaging data by leveraging minimally supervised learning techniques. It enables the differentiation between normal and abnormal data without the need for extensive human annotation. This technology also allows for the integration of sparse human annotations, enhancing its accuracy and adaptability in various imaging contexts. Developed by the University of Michigan, this method builds on proven expertise in unsupervised, few-shot, and semi-supervised representation learning.
Technical specifications
- Minimally supervised learning: Utilizes minimal human input, reducing the need for labor-intensive data labeling.
- Representation learning: Effectively separates normal from abnormal data within large imaging datasets.
- Human-in-the-loop: Incorporates sparse human annotations to refine and improve anomaly detection accuracy.
- Versatile applications: Applicable to various types of imaging data, allowing for broad use across industries.
Technology readiness level
Currently at TRL 3, this technology has been validated in controlled environments and builds on prior successful implementations of unsupervised anomaly detection. Further development and testing are required to transition to higher TRLs and full-scale deployment.
About University of Michigan
The University of Michigan is a comprehensive public research university based in Ann Arbor with additional campuses in Dearborn and Flint, known for a broad, interdisciplinary research enterprise and a major academic health system. Industry partners engage through co-located facilities, including a large north campus research complex with shared labs and incubator space, plus on-campus testbeds for rapid prototyping and validation. Proximity to Detroit’s mobility and manufacturing base, plus dedicated business engagement teams, streamlines sponsored research and access to talent. Research is supported by competitive federal funding from agencies such as NIH, NSF, DOE, and DoD, alongside state and industry sponsorship. A centralized tech transfer office manages IP, licensing, and startup support, with corporate memberships and flexible agreements.
delivered to your inbox.