AllThingsRTC - Finding Hidden Issues with Machine Learning

One of my clients is, the only major WebRTC network monitoring provider. I have been working with their analytics team and they asked me to present for them at the AllThingsRTC conference in San Francisco.

WebRTC produces mountains of data that can be used to optimize streams and debug problems—if you know where to look. I reviewed how uses unsupervised learning to discover non-obvious issues inside the vast amount of call quality data the company collects. ML methods reviewed (and ones I had do learn deeply enough to explain) include feature reduction with Principal Component Analysis (PCA), Clustering with Gaussian Mixture Model (GMM), and optimizing cluster sizes with Bayesian Inference Criterion (BIC).

Duplicating the process is a lot easier than understanding how it works!

Check out my talk at:

See the slides here: