A Multimodal Dynamic Bayesian Learning Framework for Complex Decision-making
Abstract
This project will develop a comprehensive machine learning framework to enhance the overall effectiveness of the decision-making process in complex military operations. The framework will build novel statistical learning models that leverage machine capabilities to transform large-scale complex and fast changing data of different forms to interpretable knowledge and inform complex decision-making. A human-in-the-loop paradigm will be seamlessly integrated into the framework to visualize the discovered knowledge and enable intuitive interactions with the model outcome to achieve effective human-machine collaboration. Our objectives are to (1) analyze and fuse large amounts of heterogeneous and dynamic data streams from multiple sources and provide interpretable decision recommendations, (2) conduct uncertainty analysis to establish proper confidence levels and perform loss estimation to mitigate potential risks in decision-making, (3) reduce cost for additional data acquisition on the battlefield to achieve model uncertain reduction, and (4) provide an intelligent interface to display decision recommendations and allow intuitive interactions to support human-machine collaborative learning that continuously refines model performance to maximize the overall effectiveness of decision-making. This pioneering research project builds a comprehensive framework to support complex decision-making by collectively addressing its multifaceted challenges.Our research will create an innovative model to simultaneously analyze multimodal heterogeneous data streams to extract high-level features and their temporal dependencies within and across different data modalities. A decision-aware prediction model will algorithmically fuse multimodal features through Bayesian multi-kernel learning to generate robust and interpretable prediction results, while accounting for uncertainty and loss analysis, is an integral component of the decision model outcome. When model uncertainty is high and additional data needs to be collected to improve the model confidence, an active learning model will provide key insight to achieve cost-effective information acquisition through a novel data sampling mechanism. Finally, the rich output from the decision model will be effectively communicated to a human decision team through a hierarchical visualization structure to organize information into different tiers. Such a visualization structure will facilitate both human-machine and human-human interactions to best support complex collaborative decision-making.This work will directly impact several priorities defined in the Naval Research and Development Framework, specifically ???Augmented Warfighter???, and ???Sensing & Sense-Making???. This research will also advance several fundamental areas in both AI and machine learning. These include dynamic and sequential data modeling, multimodal data fusion, Bayesian statistical learning, nonparametric modeling, and interactive machine learning. The project will have the capability of providing decision-support to numerous areas within the Naval domain ranging from humanitarian missions to both non-kinetic and kinetic conflicts. The proposed research is potentially transformative and broadly applicable to many other science and engineering domains (e.g., finance, business, medicine, and biology), where large volumes of complex data are collected in real time, which go beyond human capacity to fully comprehend and manage. Therefore, a systematic decision support system is in need that leverages machine intelligence to analyze large-scale complex and dynamic data for knowledge discovery and decision recommendation. It is also essential to allow humans to stay in the loop for interpreting the result and interacting with the system to achieve high-quality decisions through continuous human machine collaboration.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 17, 2018
- Source ID
- N000141812875
Entities
People
- Qi Yu
Organizations
- Office of Naval Research
- Rochester Institute of Technology
- United States Navy