Small data. Strong noise. Great insights.
We tackle problems holistically.
We provide code-first, math-second courses with plenty hands-on experience and interactive lecture materials. LEARN MORE ►
We focus on what we do the best.
Probabilistic Programming allows the data scientist to focus on model design, i.e. on the qualitative structure of the model that demands domain knowledge, while a universal inference engine fills in the quantitative details to obtain insights from noisy data.
Overfitting is one of the main problems in the analysis of small data sets. We employ novel approximations of the marginal likelihood as well as state-of-the-art techniques such as LOO-PIT to objectively compare and combine candidate models of varying complexity.
We help you navigate the non-linearities and emergent effects that are ubiquitous in complex systems and make standard analysis approaches fail, such as time-varying correlations, fat-tailed distributions, structural breaks, or regime-switches.