Federated Learning for Accurate Object Classification

RT&L FOCUS AREA(S): Artificial Intelligence/ Machine Learning TECHNOLOGY AREA(S): Space Platform; Battlespace The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with section 3.5 of the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. OBJECTIVE: Design an Artificial Intelligence/Machine Learning (AI/ML) system that distributes across multiple sensors and systems terminating at the interceptor. DESCRIPTION: Federated Neural Nets (FNNs) train an algorithm across multiple, heterogeneous decentralized edge devices, such as sensors, or cell phones, where the assumption of independent, identically distributed data may not be valid. Google has explored this concept of keeping all the training data on the device, decoupling the ability to do machine learning from the need to store the data in the cloud, or a centralized server. FNNs have been utilized to learn a shared prediction model while components provide more secure transmission of information. For missile defense, the interceptor seeks determination of object class, as well as threat attributes to support handover applications that could benefit from artificial intelligence. However, the interceptor is separated from raw sensor data which could inform neural nets by layers of information processing. In a distributed system, generally, raw data is refined repeatedly through the system, and the terminal node then has very limited information available to resolve complex, dynamic situations. If the system is designed and trained, however, as one coherent network, the information transmitted for interceptor exploitation is maximized. In this case, layers may reside in a distributed manner and parameters are transmitted across the communications network, adding an extra layer of security. For a classical supervised machine learning system class labels are associated with input data, and this is the goal here, but the NN itself is distributed across multiple nodes. The proposed solution is not required to be a NN, or an FNN in particular, but needs to utilize elemental data in a manner that allows maximum flexibility for the terminal element. PHASE I: Design a system that distributes a machine learning application across multiple, disparate sensors. Analyze message size, processing loads at each node, and utility for terminal node. Explore factors that may inform feasibility, suitability, and utility in the missile defense system. Deliver a prototype system with test data and results. PHASE II: Expand and mature the prototype to the actual missile defense system with realistic data, transmission rates, and potential bandwidth limitations. Demonstrate performance for terminal node threat identification. Explore robustness across system uncertainties. PHASE III DUAL USE APPLICATIONS: Many data fusion applications where raw data can be fused at a remote node to predict a classification would benefit from this technology. This could include illness prediction from pharmacy ordering patterns and/or traffic prediction from cell phones. REFERENCES: 1. Communication-Efficient Learning of Deep Networks from Decentralized Data, H. Brendan McMahan, et al, Feb. 2017, arXiv: 1602.05629v3. ; 2. Federated Machine Learning: Concept and Applications, Qiang Yang et al, Feb. 2019, arXiv:1902.04885v1. ; 3. Performance Analysis of Distributed and Federated Learning Models on Private Data; International Conference on Recent Trends in Advanced Computing 2019. ; 4. Agnostic Federated Learning, Mehryar Mohri, et al, Proceedings of the 36th International Conference on Machine Learning, 2019.