Amazon’s Architecture for Continual Learning
The research proposes a novel AutoML-based architecture to solve one of the toughest challenges in ML.
I recently started an AI-focused educational newsletter, that already has over 150,000 subscribers. TheSequence is a no-BS (meaning no hype, no news etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. Please give it a try by subscribing below:
The best source to stay up-to-date with the developments in the machine learning, artificial intelligence, and data…
Continual learning is one of the most monumental challenges in modern machine learning(ML). The traditional paradigm in ML evolves around training( or pretraining these days) models with human intervention. From a practical standpoint, this paradigm causes models to be “stuck on time” in terms of knowledge at any given time. Continual learning is an emerging paradigm to build ML models that can incrementally improve their knowledge. One of the most innovative work in this space came from Amazon Science( Amazon’s research division) published in a 2019 paper proposing a reference architecture for continual learning which combines novel AutoML techniques.
Titled Continual Learning in Practice the paper expands on some of the principles of AutoML to allow models deployed in productions to adapt to changes in data distributions. Amazon Research proposes the notion of Auto-Adaptive Machine Learning(ML), to enable ML systems that can adapt to changing conditions in production without the need of human maintenance. You can think about Auto-Adaptive ML as AutoML for deployed systems. While AutoML focuses on streamlining the creation of ML models, Auto-Adaptive ML tries to enable the maintenance of production systems using a zero-touch approach.
Amazon Research’s Auto-Adaptive ML architecture relies on the notion of continual learning to enable systems that can evolve even after they are deployed on production. The Auto-Adaptive ML architecture includes building blocks for the self-diagnosis of errors…