Inference and prediction will be as central to tomorrow’s applications as databases are to today’s.
Every application, every website, every device is already collecting more information than anyone knows how to effectively use. We need an inference solution — a simple way to make sense of the structure that underlies the data — that will scale everywhere. Developers and innovators want to exploit this information in their applications, but you can’t hire a data scientist for every app.
At Prior Knowledge, we’re building systems that are flexible enough, and easy enough to use, that they can be deployed anywhere. Our inference technology can incorporate your prior knowledge about the world, and effortlessly integrate new data. And it’ll be faster and cheaper too.
Soon every tool you use will be learning and adapting in real time to the data it observes. At Prior Knowledge, we’re building the infer-structure that will make this possible.
Bayesian Nonparametrics makes infer-structure possible.
The modeling primitives that underlie our products and services are based on a relatively new class of
Bayesian nonparametric objects. Bayesian nonparametrics goes well beyond the models that have seen such success in recent years in domains like spam detection. Those methods — Naive Bayes classifiers and Bayesian networks — aren’t nearly expressive or flexible enough for most real world use cases. Instead, we use more structured Bayesian priors to capture much richer hypothesis spaces. This lets us flexibly and robustly handle noisy, sparse, and heterogeneous data where other methods fail.
Massively scaled MCMC makes infer-structure practical.
But advanced modeling is only half of the story. Even the best model is useless if it can’t be matched with big, messy, real-world data. To make joint inference possible in production we use massively scaled Markov Chain Monte Carlo (MCMC) samplers. These inference engines — and we’re producing some of the most advanced ones in the world — give us the computational power to turn nonparametric Bayesian methods from an elegant irrelevance into a transformative force for data analytics.
