Learning from Past

The PRE-AMBLE:

It has been said many times by scholars, historians and wise men, that "History has a way of repeating itself". A more nuanced and updated version to that statement is, "At macroscopic societal level, there are threads of group behaviors that seems to form a pattern. These pattern seems to repeat under similar conditions."

Well... the second statement is a soothing music (or Hard core rock band , depends...) to a Machine Learning technologist. If there is a pattern and it tends to repeat then, by capturing right data, those patterns can be learnt (Modeled) and under similar conditions could be predicted (Inferred).

The MID-AMBLE:

This is a very simplistic yet fundamental view of application of ML/AI paradigm to solve real world problems. At Rekor Systems (www.Rekor.ai), I and my team have been focusing on solving issues relating to Urban infrastructure, Traffic management, safety and Urban planning. Almost all of these topics are macroscopic societal behavior patterns. These are impacted by external decisions such as urban planning decisions, roads, highway and pedestrian infrastructure, and external events such as weather conditions, large scale social events and other predictable periodic events (Super Bowl). In turn these macroscopic behavioral pattern affect the very same decisions and events in future in subtle and some time not so subtle fashion.

In essence there are underlying dependencies, patterns and events that form a loosely coupled, partial feedback system(s), that have a macroscopic system response, both in steady state and for an impulse impetus. The steady condition refers to changes in the events and inputs that are within a certain margin and somewhat expected. The examples could be morning rush hour traffic on a normal Monday as well as traffic density observed on a Friday before a long weekend. These could be modeled and under certain constraints be predicted.

The response of the "Partial feedback system" to impulse input is far more interesting and challenging. In our scenario and impulse input to the system could be a larger scale weather system moving through a geographical region, or a security threat/event that has a cumulative as well as domino affect on the surrounding pattern of behavior.

The curious and technically challenging aspect is that , both of the above scenarios could be modeled in a mathematical and by extension (with some assumptions) a practical system composed of historical data, distributed sensors, ML Models, algorithms and a processing infrastructure.

CONCLUSION: I and my team at Rekor Systems are working on exactly this scenario. We are incorporating various sensor such, video cameras to monitor vehicles and pedestrians, connected vehicle data, historical traffic pattern data and geographical location based information. This is a complex, vast and time consuming undertaking, but the resulting ecosystem has tremendous value in commercial as well it's impact on improving macroscopic behavior from planning and safety perspective.

What Next: This will be an ongoing post and as I and my team makes progress , I will share some real world examples and results. Till then ...be curious and always ask questions.