Machine Learning on the stream is useful for a few important reasons: scenarios where we want to dramatically reduce the resource utilization while providing high fidelity results and in use cases where we need algorithms to adapt to changing patterns and drifts in distributions real time.In this talk, we will discuss ongoing work in the area of streaming machine learning and show how we leverage Flink and DSP to build real time machine learning systems that allow us to perform adaptive thresholding and anomaly detection online.As an application of these principles, we will showcase how real time machine learning is used to detect anomalies in DSP pipelines.The talk will introduce relevant background in streaming machine learning as well as the problem of anomaly detection on Kubernetes logs.
Slides PDF link - https://conf.splunk.com/files/2019/slides/DEV1139.pdf?podcast=1577146224