Continual Learning in Practice
arXiv:1903.05202
Abstract
This paper describes a reference architecture for self-maintaining systems that can learn continually, as data arrives. In environments where data evolves, we need architectures that manage Machine Learning (ML) models in production, adapt to shifting data distributions, cope with outliers, retrain when necessary, and adapt to new tasks. This represents continual AutoML or Automatically Adaptive Machine Learning. We describe the challenges and proposes a reference architecture.
Presented at the NeurIPS 2018 workshop on Continual Learning https://sites.google.com/view/continual2018/home
References in corpus (6)
- Practical Bayesian Optimization of Machine Learning Algorithms
- Bayesian Online Changepoint Detection
- Change-Point Detection in Time-Series Data by Relative Density-Ratio Estimation
- Experiments with Random Projection
- Computing Extremely Accurate Quantiles Using t-Digests
- Multiple Adaptive Bayesian Linear Regression for Scalable Bayesian Optimization with Warm Start
Cited by in corpus (10)
- Challenges in Deploying Machine Learning: a Survey of Case Studies
- DualNet: Continual Learning, Fast and Slow
- Monitoring and explainability of models in production
- Modyn: Data-Centric Machine Learning Pipeline Orchestration
- Efficient Architecture Search for Continual Learning
- Towards better data discovery and collection with flow-based programming
- HyperStream: a Workflow Engine for Streaming Data
- Knowledge-Adaptation Priors
- Online Continual Learning in Image Classification: An Empirical Survey
- Interpretable Anomaly Detection with Mondrian P{ó}lya Forests on Data Streams