Secure Modeling and Intelligent Learning in Engineering Systems.



The use of digital multimedia (audio, video, and images) as evidence in every sector of litigation and criminal justice proceedings is becoming the norm. For digital media to be admitted as evidence into a court of law, its authenticity and integrity must be verified. In this context, our digital audio forensics framework aims to determine the underlying facts about an evidentiary recording and to provide authoritative answers to following questions: I)Does evidentiary recording subject to anti-forensic attack? II)What is the performance of a given audio forensic method in the present of anti-forensic attack? For details see SaTC: CORE: Small: Collaborative: ForensicExaminer: Testbed for Benchmarking Digital Audio Forensic Algorithms

Deep Forgery Detector, Funded by Michigan Translational Research and Commercialization

There does not exist any explainable AI (XAI) and Fairness in AI (FAI)-enabled Forensic examiner that can combat the audio and visual forgeries including deepfakes. We aim to develop a unified audio and visual forgeries detection tool which the multimedia forensics community desperately needs.

NeuroAssist: An Intelligent Secure Decision Support System for Prediction of Brain Aneurysm Rupture, Funded by Brain Aneurysm Foundation

There are three main goals of this work:

Goal I) Prediction of Intracranial aneurysm rupture: Neurosurgeons are in dire need to have decision support system to infer whether aneurysm be operated immediately or not. Our work aims to save lives from sudden death by predicting the rupture of aneurysm using extracting and integrating data features from structured and unstructured EMRs, WBS, MRI, CT Scan and MRA. This project is funded by Brain Aneurysm Foundation and Oakland University.The collaborators include Neuroscience Insitute Henry Ford Health System Michigan, and University of Michigan.
Goal II) To automatically generate machine understandable aneurysm knowledge required for Neuroassist framework. we will use natural language processing (NLP), morphological analysis of brain images, knowledge-based information extraction along with statistical analysis operations on biomedical literature, clinical guidelines, and clinical narratives.
Goal III) To develop Neuroassist chatbot that will attend meeting of neurosurgons giving its own opinion about prediction of any aneurysm rupture.

Towards predicting emergence of infectious diseases and pandemics using multimodal data

Artificial intelligence techniques can play crucial roles in predicting emerging infectious disease and in detection of pandemic outbreaks in earlier stages. The proposed project aims to tackle the problem by developing an intelligent and interactive surveillance system for classification, perception and prediction of various stages of pandemics using hybrid Knowledge and deep learning-based techniques, particularly Convolutional Neural Network (CNN), Graph Neural Network (GNN), and other text mining methods using data from following data sources: a) internet search engine keyword usage history, b) social media content, c) online news and media content, d) IoT (Internet of Things) sensor , and e) textual component of patients’ medical reports. The proposed system will continuously monitor these data sources and use them as input for discovering hidden knowledge in order to accomplish the task of surveillance of existing pandemic prone diseases and any new disease that may result in pandemic outbreak. Thus, the system will support epidemiologists and public health authorities to predict an outbreak and will also provide relevant information including risk and impact assessment of potential pandemics and possible strategies to combat the threat.

Domain specific Automated Knowledge Graph Generation

The automated curation of knowledge graph from voluminous neurological unstructured data can extract actionable information which is machine understandable and can potentially help knowledge discovery from clinical data, assisting clinical decision making and personalized treatment recommendation, to understand about unknown risk factors, natural history of complex diseases and effectiveness of different treatments. This project involve a novel framework to organize and integrate the clinical data into conceptual knowledge graphs by employing semantic and statistical analysis on the clinical data. The generated graphs are later used for knowledge-infused learning. Likewise, we are working on developing knowledge graph for Composite and Hybrid Material Interfacing and Pandemic managment.