search

drcrook1 / AI_Accelerators_Quality

AI/ML Accelerator sample for training and inferencing very large quantities of similar-ish things in a cost effective manner. Based on experience for real time quality detection or produced things in the manufacturing space.
MIT License
4 stars 2 forks source link

Redesign Entire Solution #9

Open drcrook1 opened 4 years ago

drcrook1 commented 4 years ago

Current Solution solves only for streamed windowed telemetry. Need ability to swap algorithms out, add support for iot edge, include administrative capabilities which do not require development effort.

This means a full redesign is required. We will start with 1 click deploy of solution of the redesign with blank code deployed to the infrastructure.

Initial pass at a primarily infrastructure focused version of the redesign is attached ai_acc_architecture_image

Additional design requirements include:

  1. Interface for ALL ML models
  2. Up front considerations for multiple ml models & multiple versions of each model
  3. Multiple deployment locations for models including a diverse and distributed edge as well as centralized cloud.
  4. One click deploy strategy.
  5. Ability to leverage cognitive services & custom ML pipelines
  6. Ability to interchange transformation, models, ui and data capture modules.

V1 oob solution will be focused on edge module deployments leveraging cognitive services with chained models yeilding an object detection result to be displayed in the UI.

drcrook1 commented 4 years ago

meta data regarding class of edge device whether it is a single camera 1 compute or multi camera 1 compute or multi camera multi compute.

drcrook1 commented 4 years ago

Replace the Architecture Diagram Event Grid for Dapr

drcrook1 commented 4 years ago

image

Include concept for building out this as a pack definition inside of a larger ai platform.