Reliability Society & Women in Engineering

This event will be hybrid, but we encourage in-person attendance. Registration is free, but is required so that we can plan pizza, salad, and beverages.

When arriving Wednesday evening, please be prepared to show a government-issued photo ID, such as a Massachusetts driver’s license, at the main gate at Wood Street. Tell the guard that you would like to attend the IEEE Reliability meeting in the Main Cafeteria.

After parking, walk towards the main building near the flagpole. Before entering, look left, and walk down the steps. At the bottom of the steps, turn right. Walk straight through the double set of double doors and straight into the Main Cafeteria.

Looking forward to seeing you in person!

Dan Weidman – Chair, IEEE Boston Reliability Chapter

Agenda:

5:00 pm doors open, for networking. Arriving earlier is ok.

5:30 pm: Pizza, salad, and beverages arrive, while networking continues.

6:00 pm: Introduction to the presentation, followed by the formal presentation.

About 7:30 pm: Formal presentation and formal Questions & Answers end.

About 8:00 pm or so: Informal Q&A and networking end.

Location: In-person attendance is encouraged.  MIT Lincoln Laboratory, 244 Wood St, Lexington, Massachusetts, Building: Main Cafeteria

In case you cannot attend in person, a Zoom link will be posted here shortly before the event.

Registration: 

Traditional statistical methods assume that physical and artificial systems exist in thermodynamic equilibrium, where uncertainty follows exponential distributions. However, real-world complex adaptive systems, such as hydro-meteorological hazards, biological population shifts, epidemiological outbreaks, and volatile financial markets—operate far from equilibrium. In these systems, heavy tails with extreme fluctuations often result in divergent variance. Artificial intelligence (AI) algorithms are trained to minimize variations on the cross-entropy between a learned distribution and a training set. In linear, non-fluctuating systems, the learning process is optimizing the configurational DOF (degrees of freedom); however, in nonlinear, fluctuating systems, the nonlinear DOF dominates.

The coupled entropy addresses this issue by separating linear scale (known unknowns) from non-linear tail shape (unknown unknowns). By taking mathematical expectations over a modified probability space called the independent-equals distribution, coupled entropy increases the penalty on rare outliers while dampening their sampling frequency. This keeps sampling variance finite during training while still capturing the dynamics of the original complex system. The coupled entropy has been proven to be the only function that reduces the nonlinear DOF to zero, when matched to the nonlinearity of the system.

Building directly on an empirical foundation, we propose the Active inference with Coupled free Energy (ACE). This work offers a framework for autonomous agents operating in extreme environments.

Kenric Nelson

Presented by Dr. Kenric Nelson, with Igor Oliveira and Amenah AL Najafi

Biography:

Dr. Kenric Nelson is the Founder and President of Photrek, Inc., which is developing novel approaches to Risk Governance, including risk-aware machine intelligence, particularly for environmental risks, and governance policies for decentralized communities. Nelson is the inventor of the Coupled Entropy and its application to learning extreme models. Prior to launching Photrek, Nelson was a Research Professor with Boston University’s Electrical & Computer Engineering Department (2014-2019) and a Sr. Principal Systems Engineer with Raytheon Company (2007-2019). His education in electrical engineering includes a B.S. degree Summa Cum Laude from Tulane University, an M.S. degree from Rensselaer Polytechnic Institute, and a Ph.D. degree from Boston University. His management education includes an Executive Certificate from MIT Sloan and participation in NSF’s I-Corps program.