Building Scalable End-To-End Deep Learning Pipeline in the Cloud (DeveloperWeek Global 2020)

Build a scalable end-to-end deep learning pipeline in the cloud by leveraging Cloud Native Orchestrators and a combination of AWS services, including Lambda, Batch, and SageMaker.

Key takeaways

The main challenge of GPUs is their high cost and the need to maintain a cluster.
Cloud Native Orchestrators provide a convenient way to build scalable end-to-end deep learning pipelines.
Batch processing is useful for processing large amounts of data, as it allows for parallel execution.
The speaker recommends using AWS Lambda for short processing tasks, AWS Batch for multiple tasks, and SageMaker for GPU training jobs.
Serverless infrastructure is a good option for inference, but it may not be the best choice for training due to the cold start problem.
The speaker suggests using a combination of services to build a scalable deep learning pipeline.
The cost of prediction is a key consideration, and the speaker recommends using Inferentia instances for inference.
The speaker recommends using a combination of services, including Lambda, Batch, and SageMaker, to build a scalable deep learning pipeline.
The speaker also suggests using a repository to store the model and its versions.
The speaker recommends implementing a feedback loop to improve the model over time.
The speaker suggests using a combination of services, including Lambda, Batch, and SageMaker, to build a scalable deep learning pipeline.
The speaker recommends using a combination of services to build a scalable deep learning pipeline.
The speaker suggests using a combination of services, including Lambda, Batch, and SageMaker, to build a scalable deep learning pipeline.
The speaker recommends using a combination of services to build a scalable deep learning pipeline.
The speaker suggests using a combination of services to build a scalable deep learning pipeline.
The speaker recommends using a combination of services to build a scalable deep learning pipeline.

Building Scalable End-To-End Deep Learning Pipeline in the Cloud (DeveloperWeek Global 2020)

More talks