Machine Learning

Back to Data Science

It is no secret that machine learning has become a pillar of technological advancements in the present era. Enabling computer systems to learn on their own requires a balance between acquiring sufficient data, understanding the underlying problem, and choosing the best algorithm for the job. At PEC, we understand the complex world of machine learning in the context of engineering and science applications. We work with a broad range of approaches including neural networks, data classifiers, genetic programming, and symbolic regression. While we have experience in traditional data science algorithms, we are continually keeping our finger on the pulse of the machine learning industry.

We help our clients understand the machine learning landscape and demystify the allure and confusion often attached to the field. We evaluate each client’s needs on a case-by-case basis and deploy algorithms suited to the task at hand – whether that be a bleeding-edge technique or classical approach. Our philosophy is that good solutions should embody simplicity, extensibility, and fidelity to the underlying physics of the problem.

Core ML

PEC’s team employs machine learning algorithms and artificial intelligence frameworks to solve challenging engineering problems. We have experience handling unwieldy data that require dimensionality reduction, data structuring, and parallel processing. We’ve developed techniques for handling data sparsity that involve advanced regularization, scaling, and emulation methods. Our experience includes a broad spectrum of approaches, ranging from regression and classification to evolutionary and deep learning algorithms. We have also employed algorithms based on mathematical concepts that include graph theory, network analysis, convex optimization, and outlier detection. This knowledge base allows us to deploy the best ML approaches to fit our clients’ needs.

Genetic Programming

A shortcoming of complex machine learning algorithms is the “black-box” nature of their underlying mechanics. PEC has sought to counter this weakness by employing a different paradigm called genetic programming. This technique creates a virtual ecosystem of organisms, whose genes are comprised of mathematical expressions. The population of this ecosystem reproduces and mutates over many generations, evolving over time to produce elegant mathematical solutions. When applied to problems such as symbolic regression, a detailed and comprehensible solution is derived based both on accuracy and simplicity. PEC has developed a state-of-the-art framework to apply genetic programming and symbolic regression to complex engineering problems, known as EMBER. The solutions it generates can be clearly understood and interpreted by humans, which helps us verify that the models are not only correct, but that they are “right for the right reasons.”

Computer Vision

Our team develops and applies computer vision algorithms to a wide variety of problems related to engineering testing and analysis. We have experience leveraging the Open Source Computer Vision Library (OpenCV) to perform detailed analysis on static images and full motion video. We offer pragmatic solutions that bridge the gap between qualitative observations and highly-reliable sets of quantitative data. We have developed solutions for high speed video tracking of fragments generated during explosions, automated laser sighting systems, batch processing of tiled photographs, etc. Our algorithms are augmented by various machine learning and statistical techniques to improve performance and to stay on the bleeding edge. These solutions help automate research and testing efforts, reduce analyst burden, and produce non-subjective interpretations and measurements.

Related Resources

Fragment Field Modeling Using Digital Twins and Generative Adversarial Networks (GAN)

Pulsed Laser Measurement for Materials at Extreme Loading Rates

FragTrack and ParticleTrack: Computer Vision Applications to Track & Quantify Test Video Data


Ready to work with us?