ARiA applies broad interdisciplinary experience in acoustics, signal processing, and cognitive science toward innovative basic and applied research.
We specialize in developing
- real-time physics-based modeling algorithms for simulation and training
- machine-learning algorithms for automated pattern recognition
Our areas of competency in Acoustical Modeling & Simulation include
- Physical Models for Propagation, Reverberation, and Scattering in the Ocean and in Air
- Fidelity Assessment for Simulation-Based Training
- Scale-Model Validation Studies
By understanding the needs of the end user, whether a tactical-decision aid or real-time training system, ARiA is able to develop models and evaluate their fidelity with scientifically rigorous and tactically meaningfully metrics. Fidelity in modeling is a means to end, not a goal in itself.
Advances in modeling & simulation now enable simulation-based training to augment live training in many applications. But the viability of simulation-based training rests on understanding the relationship between training goals and fidelity and ensuring that the computational load and instructor load resulting from scenario design are optimized jointly with training goals. ARiA is developing new algorithms that automate the process of scenario design that work interactively with instructors to ensure that fidelity is sufficient to meet training goals while balancing restrictions due to computational load.
Our areas of competency in Signal Processing include
- Detection and Classification for Passive and Active Sonar Systems
- Perceptual and Cognitive Signal Processing for Pattern Recognition
- Supervised, Unsupervised, Semi-supervised, Active and On-Line Learning
- Model-Based and Physics-Based Signal-Processing Techniques
ARiA recognizes that “the best pattern recognition system that we know of is our auditory system” and has developed biomimetic signal-processing algorithms for detection and classification and machine-learning algorithms that learn from listeners how to best classify and identify complex signals. We also recognize that physics-based approaches can give unique insight to complex problems of detection and classification. At the same time, we create models that are data-driven and learn from the data themselves in order to avoid errors and brittleness that can arise from model-data mismatches.