Biologically-Inspired Machine Intelligence for NASA's Missions, Systems, and Facilities

New advances in computing hardware architectures and algorithm development derived from high fidelity modeling of the mammalian cerebral cortex have the potential to provide revolutionary advances in machine intelligence capabilities. The Cortical Learning Algorithm (CLA) represents such an algorithm. It was recently open sourced in the summer of 2013 as part of the Numenta Platform for Intelligent Computing (NuPIC). We propose to assess the current capability of the CLA in the domain of anomaly detection, a domain which relates to a large set of NASA problems. We choose to focus on two machine learning (ML) techniques: regression and adverse event prediction. Our goal is to study how well the neural processing based mechanism of CLA performs on these two tasks. Both will be tested using a well understood scenario from a NASA Aviation Safety dataset. For the assessment we will make use of ACCEPT, an in-house developed Matlab-based framework geared towards the prediction of adverse events. ACCEPT contains a variety of state-of-the-art algorithms that can be used as a baseline to compare the performance of the CLA. We will assess the performance of all algorithmic approaches using metrics such as NMSE (Normalized Mean Square Error), false positive and missed detection rates, and detection time. Topics and assumptions associated with this research are: (1) assessment of CLA applicability towards the ML techniques described above is part of the assessment, (2) the CLA and NuPIC are in an early stage of development; the CLA mapping to neuroscience continues to evolve. More »