Information and Decision Recommender Navy STTR FY2014.A Sol No.: Navy STTR FY2014.A Topic No.: N14A-T024 Topic Title: Information and Decision Recommender Proposal No.: N14A-024-0221 Firm: Archarithms, Inc. 2904 Westcorp Boulevard Suite 101 Huntsville, Alabama 35805-6436 Contact: Mark Lambrecht Phone: (256) 527-9360 Web Site: www.archarithms.com Abstract: The objective of this proposal is to develop an advanced Course of Action (COA) recommender that supports decision making. It is impossible for Warfighters to manually utilize the vast quantities of data at their disposal. To this end, advanced algorithms and data visualization are required to aid in analyzing the data and making informed decisions. The final output of such a system is actionable COA's along with supporting data that gives an indication as to why the recommendation was made. The proposed Archarithms - Auburn University (AU) team solution will use cutting edge machine learning algorithms to sift through the preponderance of data and transform it into actionable suggestions for the Warfighter. Deep Learning algorithms show significant promise with their ability to accurately map hidden, underlying relationships in data. Coupled with their ability to change and adapt as new data is presented, they represent a powerful solution to the "data problem." Through advanced Natural Language Processing techniques, human entered COA's will map into a Deep Belief Network. This network will analyze the underlying relationships between the entered COA's and a set of historical data. The product being an actionable recommendation that maps back to a key set of features in the training corpus. The final challenge for such a system is the ability to present and visualize the data. For this, we propose coupling our prototype with an advanced visualization system that quickly and accurately provides recommendations, allows access to key data-features and does not overwhelm the Warfighter. Benefits: Intuitive data mapping for enhanced COA understanding. Large amounts of open-source historical data for accurate high-level reasoning. Parallel processing of massive data sets for scalability and speed to provide quick answers. Leverages open source tools to reduce cost. Extract features from large data sets to reduce the feature dimensions to leverage multiple data sources to provide for more accurate recommendations. Parallel processing to perform feature reduction in real-time. Reveals hidden connections in the data for robust reasoning. Novel distributed training technique for scalability and speed to provide quick answers Return