Unpredictable predictions of self-driving cars AI. Handling inference in anomalous ever changing environment.

No matter how good your Machine Learning model is trained, the inference output space leaves a wide range for appearing irrelevant and unexpected results when real world gives a model an unforeseen challenge. Those error inferences may lead to accidental outcomes, there are notorious cases we all know. For business to rely on AI/ML such an outcomes are unacceptable. The solution is robust monitoring for the edge cases and implementing the Active Learning concept into businesses' AI/ML operations for those cases to be handled and learned. The talk will be dedicated to the problems of including edge cases into self-driving cars AI inference space, practical solutions and their implementation into business operations. Schematically and generally the process of implementing of ML into business processes clear and simple: data scientists prepare initial data set to build and train a model, then that model is deployed to serve in computer application to make inferences and predictions based on a real data fed to model’s inputs. While model works making inferences it is being monitored for accuracy and supposed to get re-trained in case of particular degree of discrepancy showed and then re-deployed into production. Train, deploy, monitor, over and again, and that is it. That is not, of course. Taking a self-driving cars case - the AI is prone not only to deliberate adversarial attacks - messing or vandalizing road signs or putting a mirror in front of AI-sight camera - but a natural environmental edge cases - pedestrian wearing a mascot costume or non-standard type of a road light - make it deliver decisions we call dumb and in some cases dangerous. No matter how good your Machine Learning model is trained, the inference output space leaves a wide range for appearing irrelevant and unexpected results when real world gives a model an unforeseen challenge. Those error inferences may lead to error spreading further to a business process of even to an accidental outcomes, there are notorious cases we all know. For business to rely on AI/ML such an outcomes are unacceptable. The talk will be dedicated to the problems of including edge cases into self-driving cars AI inference space, practical solutions and their implementation into business operations. We will discuss a robust monitoring and retraining solution for the edge cases and implementing the Active Learning concept into businesses' AI/ML operations for those cases to be handled and learned. In-depth technical topics to be covered in this talk are: Generative Adversarial Networks for concept drift detection Density based clustering algorithms for concept drift monitoring Masked AutoEncoder for Density Estimation (MADE) for edge concepts detection Active Learning optimisation