Auto-adaptive machine learning
How to use continual learning for your ML models
Yochay Ettun, CEO
@yochze
yochze@cnvrg.io
whoami
• Developer/Data scientist => CEO
• cnvrg.io = built by data scientists, for data scientists to help teams:
• Get from data to models to production in the most efficient and fast way
• bridge science and engineering
agenda
• Introduction to continual learning
• Why do we need continual learning
• ML pipeline with continual learning
• AutoML
• Deployment
• Monitoring
• Summary
Introduction to Continual Learning
• Unified analytics engine for large-scale data processing
• Faster processing speed of applications due to In-memory cluster computing (100x
improvement)
• Support different workloads – batch, iterative, streaming, interactive SQL etc.
• Support multiple languages and different environments
ML pipeline with Continual Learning
Why do we need continual learning?
Why do we need continual learning?
ML pipeline with Continual Learning
AutoML and Hyperparamaters Optimization
• Data is changing à Models are changing
• Train your model on different algorithms space:
• Tune each algorithm with hyperparameters
• Choose best model performance result
AutoML and Hyperparamaters Optimization
• Train Models at scale using kubernetes
• Track your models results
• Compare between them
Automatic & safe model deployments
• Deploy new version automatically but safely
• Run tests during updating your production models
• Use Canary deployment technique:
• Reduce the risk of deployment
• Rolling updates slowly using validation in each step
Automatic & safe model deployments
• Use Kubernetes
• Use Istio
• Read our guide
Monitoring your models
• Use Prometheus, kubernetes & alert manager
• Monitor input data:
• Search for unexpected data
• Measure the correlation of production data to train data
• Monitor prediction:
• Model confidence, model bias, and more
• Add validations on the fly
Monitoring your models
Trigger retraining
• Trigger retrain based on:
• Periodically (Once in a day/week/month)
• New training data that comes in
• model decay / model bias / alerts in production
• Make sure to track and monitor alerts
Summary
https://d39w7f4ix9f5s9.cloudfront.net/cf/6d/f484e8e5445cba
91a63eddd2b1c8/continual-learning-in-practice.pdf
Thanks!
https://cnvrg.io
info@cnvrg.io
+972-506-660186
How To Build Auto-Adaptive Machine Learning Models with Kubernetes

How To Build Auto-Adaptive Machine Learning Models with Kubernetes