Dr. Ying Zhao is a research professor at the Naval Postgraduate School in Monterey, California. Zhao's research focuses on data sciences, machine learning, and artificial intelligence methods, including lexical link analysis (LLA), collaborative learning agents (CLA), and reinforcement learning for search, visualization, and analysis, for defense military applications in the areas of semantic and social networks, common tactical air pictures, combat identification, logistics, wargaming, and mission planning.
Since joining NPS, Zhao has been a principal investigator of many awarded DOD research projects. She is a co-author of four U.S. patents in knowledge pattern search from networked agents, data fusion, and visualization for multiple anomaly detection systems. She received her doctorate in mathematics from MIT and is the co-founder of Quantum Intelligence, Inc. Zhao is currently an ESEP scholar (Engineers and Scientists Exchange Program of DOD) at the Defence Science and Technology Lab, UK, through Nov. 30, 2022.
NPS has the best military experts, strategists, and policy influencers. NPS connects to war colleges and DOD research labs. NPS also has secure facilities. Our students are experienced warfighters and know the requirements; we can contribute a lot to U.S. national security priorities. The students are eager to learn the technologies and apply them to military applications. They are future leaders and can deliver strategies to guide and influence future warfare.
NPS Faculty are very motivated and well-funded for research. The DOD engineers and scientists have always been at the cutting edge of innovation. They have been developing and delivering military applications and beyond for generations, and nothing is entirely new. However, due to globalization, open society, new threats in addition to technological breakthroughs, a cultural shift may require us to adapt now.
Defense applications are uniquely challenging for data sciences, AI/ML technologies, and digital modernization because sometimes there is overwhelming big data such as in the logistics, sensor, and system engineering areas. Sometimes there is little data or no data such as emerging behavior and intention of red forces or unknown situations. To generate synthetic data, simulating the environment and adversaries is important. This is related to AI's cutting-edge AI GANS (generative adversarial networks), GPT3 (Generative Pre-trained Transformer 3), and AGI (Artificial General Intelligence) assistant models – "a knowledge system that is always around us, learns and help us learn" (Eric Schmidt). This is because the AI breakthroughs (e.g., AlphaGo, AlphaZero, AlphaFold among others) show that data sciences, AI/ML and tremendous computing power can create AI assistants to discover and "see" things that human cognition and eyes cannot see. For example, the computer found new moves in the strategy board game "Go" that Asian people who played "Go" for thousands of years did not discover. An equipped virtual agent should also aid human warfighters in a very revolutionary and profound way. This aligns with Eric Schmidt's, who chaired the U.S. National Security Commission on AI, view on how AGI would be used in the future.
For military applications, such as AGI, should include sets of AI/ML tools and simulation models, which are consistent, explainable, no black boxes, and can be used to test theories for a range of users in a wide range of applications such as campaign/mission planning, future warfighting concepts designing and simulation, warfighter training, etc., allow different questions to be asked easily. The proposed end user areas are very human-machine interactive. Therefore, there is a lot of room to compare machine intelligence and human cognition, which in my view should be complementary. For example, synthetic data and AI recommendations may not be accurate, valid, or feasible, therefore, need human collaboration to check the feasibility and validity of the synthetic data.
We need to invest in the AGI type of AI assistant for warfighters because a demonstration of a defense AI product in a specific area is difficult to implement in a short period with respect to the strict DOD data and application security. The need is to develop an AGI that does not need to re-program heavily toward plug-and-play with data models for specific processes and areas.
The ML/AI community is driven heavily by open-source software and algorithms. Data and business are private and unique to the industries and applications. Competitors may have been leveraging the open sources very much, and we should do that too. For example, the book "AI Superpowers" by Kai-Fu Lee discussed how China used open sources to speed up applied AI technologies dramatically.
In the past, I've been funded by the NAVAIR China Lake & ONR FNC, NPS NRP, and ONR NEPTUNE programs with various projects related to the framework I called "leverage AI to learn, optimize, and wargame (LAILOW) for defense applications." I have collaborated with the MIT AF AI Accelerator and CSAIL to apply LAILOW to cross-domain use cases. The NPS students are very experienced warfighters and know the requirements. Our students may not code like AI professionals and computer scientists, but they should be able to map the requirements to potential solutions and tools developed by the AI/ML community. NPS is also close to Silicon Valley and first-class AI universities like Stanford and Berkeley. There are tremendous opportunities and needs for the NPS to build the MIT AF AI accelerator-like platforms to communicate the defense application requirements to the academic communities, connect requirements with the advanced research results and tools, and build DOD AI application pipelines. The related funding platforms NavalX, DIU and Tech Bridges have been focusing on the startups, which is good. However, we can do more about education and students' thesis projects to reach out and leverage their resources.
The logistics and supply chain enterprise has a tremendous amount of data and probably is the unique place that can have a digital twin; therefore, "learn" refers to applying AI/ML to learn and discover patterns from historical data and then use the patterns for prediction and forecast. The logistics and supply chain enterprise is a very complicated business enterprise and needs constant decision and action optimization, which the operation research has traditionally addressed. "Win" refers to the wargame perspective that I have introduced in this project to divide the business process into the problem areas (attackers) and solutions areas (defenders); the wargame is between the problems and solutions. The wargame serves as a what-if analysis to analyze and discover risk areas where the enterprise can (resilience) or cannot (vulnerability) handle in a simulation and synthetic environment. LAILOW is an AGI framework and the USMC supply chain provides a good use case for the framework, the framework is important not for just logistic planning, but all planning areas.
NPS has planned to send students to the MIT AF AI Accelerator, I hope to continue working with the NPS students and the Accelerator to advance the LAILOW framework. The other proof-of-the-concept projects of LAILOW are shown below:
I am working on a Defence Science and Technology Lab project to design a virtual red AI agent for the integrated air missile defence, IADS. The vision of the red AI agent will be beyond IADS. I also work with the Dstl Porton group in the areas of AI for Radio Frequency, patterns of life used for modeling civilian populations, and influence operations using open source data and tools.
My personal interests also include quantum intelligence game and machine learning, quasicrystal theory as language models related to the origin of consciousness, causality, mathematical models, and human cognition.
