I recently examined the public presentations of teams participating in a NASA Frontier Development Lab exhibition. The teams were selected to present tech solutions for NASA's space exploration challenges. The FDL held its 2017 wrap-up event down at Intel's Santa Clara auditorium, so I drove on down and lucked out finding a parking space. I had to see what kinds of space-related innovations apply to private sector use.
The presenters were graduate-level statistical modelers and astrophysicists. Explaining comet tracking and solar storm prediction is beyond the scope of this blog. The salient lessons lie in the statistical models and data science methods the teams used that may be available to the rest of us. Teaching a neural network to recognize patterns requires feeding it large amounts of synthetic data "training sets." Experts design those networks, but the most user-friendly open source networks are available for business applications.
Deep learning methods in the teams' work applied long short-term memory, random forests, and convolutional neural networks to compare data sets derived from deep learning methods. They used t-distributed stochastic neighbor embedding (t-SNE) to reduce high-dimensional data into scatter plots and DBSCAN to separate data outliers from clusters. Experts apply principal component analysis to discover eigenvector centrality, useful in tracking clusters of nodes of influence. All of these advanced techniques are useful additions to more prosaic forecasting methods found in MBA curricula. Time series forecasting models, for example, are useful for establishing baseline trends in naturally occurring data, but they miss anomalies and outliers completely.
Those of us who don't have post-graduate data science credentials can still employ useful tools. Google's Kaggle platform enables crowdsourced data analysis solutions. The Keras API enables open source experimentation with deep neural networks, and now it extends the TensorFlow open source library. Google Search results for "open source deep learning platform" show a growing number of tools and libraries available for use.
The US government's technology outreach programs offer valid paths to bonanza. Space tech AI and machine learning designed for celestial data sets could easily adapt to commercial use in mining, energy, and environmental remediation. NASA FDL exhibitions attract very smart people whose advanced data science skills are very valuable to startups building commercial solutions that process vast amounts of volatile, unstructured data. Count on Alfidi Capital to discover tech expertise that is out of this world.