Sci-Int

Status: Ongoing, Funded by: DARPA

The Scientific Integrity (Sci-Int) project is an endeavor funded by the Defense Advanced Research Projects Agency (DARPA), whose goal is to develop solutions to identify misconduct in scientific research through the detection of image tampering in scientific papers.

Working together with the Universities of Purdue, South California (USC), Campinas (Unicamp), Naples Federico II, and the Politécnico di Milano, our team leads the development of techniques to (1) rule out false positive cases of misidentified misconduct, and to (2) better report and document misconduct practices in retraction notices.

To accomplish item (1), we plan to perform semantic analyses of the suspect paper cases, in order to identify situations in which duplications or manipulations of content were made with fair scientific purpose (e.g., to visually highlight or to better explain previous results). To do that, we plan to use natural language processing to perform basic automated paper text and image caption interpretation.

To accomplish item (2), we plan to rely on techniques of Provenance Graph Construction to present better explanations of how problematic images relate to each other, and how they share content across different papers.

Research Team

• Prof. Edward Delp (Lead PI)
• Prof. Walter Scheirer (PI)
• Prof. Anderson Rocha (PI)
• Prof. Kevin Bowyer (PI)
• Prof. Daniel Moreira (PI)
• João P. Cardenuto (PhD Student)