From the course: Advanced Data Processing: Batch, Real-Time, and Cloud Architectures for AI
Join today to access over 26,000 courses taught by industry experts.
Architectures for AI
From the course: Advanced Data Processing: Batch, Real-Time, and Cloud Architectures for AI
Architectures for AI
- [Instructor] How are the architectures for AI impacted by the characteristics of the use case? In general, the type of architecture needed for AI is heavily impacted by two considerations, the latency constraints for the use case and the expected deployment mode for the use case. What is latency in ML? It can be considered as the time elapsed between when raw data is available for training or inference, and when the training or inference is successfully completed. This involves the entire ML pipeline, including feature engineering, multiple iterations of training and inference. Training and inference can have different latency requirements. This brings us to two modes of ML, namely batch and real-time. Batch processing involves processing data in batches. Typically, this can take a long time, greater than an hour. Batch jobs are run at regular intervals like hourly or weekly. ML training is usually a batch process. It involves multiple iterations of training until expected results…
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.