By: David A. Teich
Internet security was one of the first areas where machine learning (ML) was applied. Last year, I talked about how security in banking facilities is being enhanced with ML. Moving up to the wider issue of security in today’s world, it’s important to consider more robust security measures. Government buildings, airports, and other buildings at risk from terror attacks, whether from individuals or organizations, need more security that visual tracking of potential threats. In addressing these complex needs, there is an even greater opportunity for ML to help.
More and more facilities in the US are adding security checks to identify threats on entry. The tools and techniques are varied and must be integrated. Metal detectors, molecular “scent” detectors, visual movements, and even low powered radar are some of the techniques in use to improve security. There is a need, though, to push security out from the door. Terrorists don’t need to get into a building to cause loss of life. One example is the security at airports, where the checks are inside the building. The Brussels terror attacks in 2016 involved people who came into the unsecured areas. Unintrusive security tracking there and outside the buildings can identify risky behavior earlier and potentially save lives.
Patriot One Technologies is a company that is working to create systems to identify those risks. “While machine learning is key to different components of our systems, as in artificial intelligence (AI) vision,” said Peter Evans, CEO, Patriot One. “Integrating and understanding all the different data feeds is a powerful way ML is helping to alert security personnel to potential risks.” For instance, chemical sensing is pure data, and AI isn’t needed to recognize potential threats, but combining that with visual or magnetic resonance information can happen faster with AI in order to quickly identify what person or object is the potential threat.
In the real world, machine learning is messy. That is one of the things delaying automated vehicles. Seeing one face and trying to recognize it is one thing, but looking at the flow of people and vehicles outside of an airport terminal is far more complex. Identifying background noise is also a difficult problem in such busy environments.
The complexity is one of the reasons I’ve been referencing both artificial intelligence and machine learning, since they are overlapping concepts. Justin Granek, VP Operations for Xtract AI, a wholly owned subsidiary of Patriot One, is focused on using the right tools for each part of the system. “Deep learning (the primary AI machine learning tool) is great for understanding the vast amount of electromagnetic data,” stated Justin. “However, random forests are more useful to analyze other data sources that don’t require the heavy processing of neural networks. Well defined problems don’t even need AI or ML, but that analysis must still be integrated into a complete system.”
Similar to the financial institution example referenced above, one of the key challenges in these security solutions is in the rapid and accurate notification to security personnel in as quiet a way as possible. User interfaces matter, and that is still an area that is a gap between academic applications of AI/ML and real world usage of such systems. With as much data that is coming into the system, it’s important to distill it into information that can be quickly understood and is actionable for human response. There is an old joke that early weather prediction systems could accurately tell you what the weather will be two weeks ago. Security can’t be that slow. More importantly, depending on the risk, you need to quietly alert security or, in the case of an accident outside, one that’s not a threat, you might want a visual alert so that personnel know where to go and those involved know that help is coming. Again, strong ML systems are needed in order to identify and classify risks, so that both system and human actions are timely and appropriate.