computer science feature

Mining Data to Fuel Discovery

By Richard Westlund

Mining Data to Fuel Discovery

By Richard Westlund
Signal processing study promises to spur vital insights

Just as fossil fuels propelled growth in the 20th century, data is the driving force in the 21st-century economy.

“Data is the new oil,” said Mitsunora Ogihara, professor of computer science and director of education for the University of Miami Institute for Data Science and Computing (IDSC).

“Understanding data patterns can lead to new discoveries in many disciplines, and powerful measurement instruments have redefined the thought processes and workflows in many modern engineering fields. These data-oriented workflows accelerate product conceptualization, feasibility research, product design, validation, manufacturing, and iteration."

Comprehensive analyses of troves of complex data also offer the potential to enhance real-time predictions of adversarial events. Toward that end, Ogihara is using artificial intelligence (AI) algorithms, deep learning tools, and other powerful computational resources to study long-duration signaling patterns. The collaborative research project is supported by a $50,000 grant from Keysight Technologies, Inc., which provides hardware and software solutions for signals captured from electronic devices.

Ogihara’s study, “Navigate the Ocean of Massive-Data-Scale Waveforms: Very-High-Dimensional Spectral Pattern Recognition, Inference and Interpretation,” draws on the University of Miami’s leading-edge technology resources to find insights hidden within vast seas of data by sampling long-duration waveforms millions of times per second.

Unlike a traffic light giving motorists three types of data—stop, go, or proceed with caution—long-duration electronic signals may contain tens of millions of data points amid a mixture of underlying patterns.

When analyzing long-duration electronic signals, “Sophisticated signal processing and pattern recognition algorithms are necessary to fully explore and utilize the data,” Ogihara explained. “Important patterns may be hidden in the massive quantity of data, well beyond the needle-in-a-haystack depth.”

Ogihara and his team are developing ways to characterize signals emerging under different conditions “by splitting sequences into chunks that are small enough to accommodate dynamic data organization, grouping them into similar patterns, and assessing long-term and short-term trends.”

Digital signals are often difficult to analyze with traditional methodologies, which typically focus on either time or frequency. “Recent engineering research suggests that most useful signal patterns reside in the dynamic spectra involving both time and frequency,” Ogihara said, “making this a timely area for exploration.”

Due to the complexity of long-duration signaling data, it is impossible for human engineers to “eyeball” useful patterns. “Since time-frequency analysis transforms waveforms into images,” Ogihara noted, “we plan to implement both manually crafted image features and deep-learning-based AI frameworks into a visual analysis module.”

By providing tools like pattern recognition and interpretation to navigate waveform data at a massive scale and at very high dimensions, Ogihara’s research stands to play a vital role in advancing the engineering discipline of signal processing.

“Our goal,” he said, “is to help realize the potential offered by massive waveform datasets and analytics and facilitate the development of new measurement instruments.”