This is a dedicated watch page for a single video.
A company is running deep neural networks on GPU-based instances for their product recommendation system. The system relies on an ML-workflow that fetches each dataset from an S3 bucket, loads it into a Keras model pulled from a local repository, then continuously delivers the results into a separate bucket. The current workflow is expensive to run and involves manual resource management. The ML-team that supervises the system wants to have the least cost and operating overhead. They’re looking for an automated solution with the following requirements: Given that a job requires several hours of on-going work, in case of a failure, a job should be able to retry automatically. Each job must be submitted to a central queue. The workflow will only run once a week, every Monday to Friday. How should the ML-team redesign the workflow?