Professional Documents
Culture Documents
OpenSAP Sustai1 Week 2 All Slides
OpenSAP Sustai1 Week 2 All Slides
Advanced analysis
Support decision making
Any means to make machines solve
Artificial Intelligence tasks like humans do Take action
Many applications
Learn through training
Machine Learning Computers can learn from examples
to solve tasks without being explicitly Statistical model
programmed to do so
Manual feature extraction
Artificial Neural Brain-inspired machine learning algorithms
Networks
Multiple layers
Handle complex data
Deep Subset of ML that uses deep artificial
neural networks Enable large datasets
Learning
Advanced applications
https://www.gartner.com/en/topics/artificial-intelligence
https://www.britannica.com/technology/artificial-intelligence
© 2022 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 3
SAP AI overview and introduction to SAP AI Core
Market overview: enterprise applications with embedded AI
197 +15.7%
AI Non-Centric 169
95 106 121 143
AI
Apps
AI Centric 12
14 17 20 25 30 +19.4%
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
SAP AI Core
Complementary AI runtimes training
training&&serving
servingAI scenarios©
AI scenarios
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
SAP AI Core
Complementary AI runtimes training & serving AI scenarios
open@sap.com
Follow all of SAP
www.sap.com/contactsap
Templates
▪ Templates are used to define the training pipelines and
how a model has to be deployed and served at the level
of the main tenant
Executable
▪ An executable is a template that is instantiated for a
purpose, such as training a model or creating a
deployment
Scenario
▪ A scenario is a group of related executables for a use
case within the user’s tenant
Execution
▪ The operation of executing an executable with a given
configuration
Artifact
▪ A reference to data or a file that is produced or
consumed by an execution or deployment which
typically resides on an object store
Configuration
▪ A collection of parameters, artifact references, and
executables that is used to run an execution or a
deployment
AI API
▪ AI API provides a standard way of managing your artifacts and workflows (such as training scripts, data, models, and model servers).
▪ Open API format, HTTP-based RESTful API definition.
Clients
open@sap.com
Follow all of SAP
www.sap.com/contactsap
Training part
Investigate the model architecture and design the training Serving part
pipeline
Design the web application that will use the trained model and
Write the code that executes all the needed steps that will serve the inference requests
Kubernetes
Docker
Container Lifecycle
Templates are used to define the training pipelines and how a model has to be deployed and served
▪ Workflow templates are used to: ▪ Serving templates are used to deploy one or
– train models more trained models on a model server and rely
– ingest data, pre-process and post-process on KServe
data ▪ KServe provides a standard model inference
– execute batch inference pipelines platform on Kubernetes, regardless of the
framework used to develop the model
▪ Workflow templates are based on the Argo
Workflows workflow engine ▪ KServe supports modern serverless inference
workload with autoscaling including Scale to
▪ Workflow templates instruct AI Core about how
Zero on GPU
to execute the Docker images, what inputs to
provide, what outputs to generate, etc. ▪ Serving templates instruct AI Core about the
container to execute, the model to use, etc.
Docker secret
Resource plan
Resource plan
Workflow and
serving templates
▪ A GitHub repository contains the declarative descriptions (templates) of the K8s infrastructure currently
desired in SAP AI Core.
▪ An automated process, ArgoCD, makes the underlying K8s infrastructure match the described state in the
repository.
▪ To create a new scenario/executable or update the existing ones, you only need to update the files in the
repository, then ArgoCD handles everything else.
Defect
Detection defect-det-app
Training template
Serving template
defect-det-scenario Training executable Serving executable
Defect Detection
Image Dataset
Docker image
open@sap.com
Follow all of SAP
www.sap.com/contactsap
Datasets
Purpose
▪ Hyperscale storage is a storage mechanism with the ability
to massively expand under a controlled and efficient
managed, software-defined framework.
PROs
▪ Cost-effective, robust, and durable, rich tooling.
▪ Data access is handled via the Argo artifacts feature which
copies data in/out of pipeline containers.
Current status
▪ AWS S3 and Azure Blob storage.
▪ Support for SAP HANA Cloud data lake files.
Future roadmap
▪ Adding other hyperscalers (like GCP).
© 2022 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 4
SAP AI Core deep-dive part 3: storage access and resource groups ups
Resource groups and multitenancy
▪ A resource group represents a virtual collection of related resources within the scope of one SAP AI
Core tenant.
▪ SAP AI Core tenants use resource groups to isolate related ML resources and workloads.
▪ Scenarios, executables, and Docker registry secrets are shared across all resource groups.
▪ Runtime entities such as executions, deployments, configurations, and artifacts belong to specific resource
groups and cannot be shared across resource groups.
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the
AI Core and Use
Data in S3 Get the Get
AI Submit for Submit for deployment
and create a trained deployment
Launchpad training deployment URL for
resource model URL
onboarding inferencing
group
open@sap.com
Follow all of SAP
www.sap.com/contactsap
IoT Gateway
x6
Computer vision
Computer vision is an interdisciplinary scientific field that deals with how computers can gain high-level
understanding from digital images or videos.
Industrial inspection
Image segmentation
Self-driving cars
Medical scans
Satellite imagery
▪ A convolutional neural network (CNN) is a deep learning algorithm which can take in an input image,
assign importance to various aspects/objects in the image and be able to differentiate one from the other.
▪ The architecture of a CNN is analogous to that of the connectivity pattern of neurons in the human brain and
was inspired by the organization of the visual cortex.
MobileNetV2 pix2pix
Load surface images (OpenCV) Pre-processing (OpenCV) Build U-Net (TensorFlow) Train the model
Modified U-Net
Normal surface
MobileNet
V2
Defected surface
pix2pix
upsample
block
Mask/ground truth
open@sap.com
Follow all of SAP
www.sap.com/contactsap
Training part
Investigate the model architecture and design the training Serving part
pipeline
Design the web application that will use the trained model and
Write the code that executes all the needed steps that will serve the inference requests
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the Submit for Get the trained
Use
AI Core and data in S3 training (using model Get
Submit for deployment
AI Launchpad and create a customer’s (registered as deployment
deployment URL for
onboarding resource training artifact URL
inferencing
group scripts) automatically)
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the Submit for Get the trained
Use
AI Core and data in S3 training (using model Get
Submit for deployment
AI Launchpad and create a customer’s (registered as deployment
deployment URL for
onboarding resource training artifact URL
inferencing
group scripts) automatically)
open@sap.com
Follow all of SAP
www.sap.com/contactsap
Training part
Investigate the model architecture and design the training Serving part
pipeline
Design the web application that will use the trained model and
Write the code that executes all the needed steps that will serve the inference requests
© 2022 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC https://flask.palletsprojects.com/en/2.1.x/ 4
Defect detection serving with SAP AI Core SDK
End-to-end ML workflow in SAP AI Core and SAP AI Launchpad
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the Submit for Get the trained
Use
AI Core and data in S3 training (using model Get
Submit for deployment
AI Launchpad and create a customer’s (registered as deployment
deployment URL for
onboarding resource training artifact URL
inferencing
group scripts) automatically)
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the Submit for Get the trained
Use
AI Core and data in S3 training (using model Get
Submit for deployment
AI Launchpad and create a customer’s (registered as deployment
deployment URL for
onboarding resource training artifact URL
inferencing
group scripts) automatically)
0.1 SAP BTP: Create AI Core and AI 2.2 Git: Create a training workflow 4.1 Docker: Create own Docker image
Launchpad instances 2.3 AI Core: Choose the scenario 4.2 Git: Create a serving template
0.2 AI Launchpad: Connect the AI Core 2.4 AI Core: Register input dataset as an 4.3 AI Core: Choose the scenario
runtime to AI Launchpad artifact
0.3 4.4 AI Core: Create a configuration
AI Core: create an AI API client 2.5 AI Core: Create a configuration
4.5 AI Core: Trigger the serving execution
0.4 AI Core: Connect a GitHub repo 2.6 AI Core: Trigger the workflow
execution 4.6 AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment, check the logs
2.7 AI Launchpad: Check the status of
execution, check the logs
0 1 2 3 4 5 6
Upload the Submit for Get the trained
Use
AI Core and data in S3 training (using model Get
Submit for deployment
AI Launchpad and create a customer’s (registered as deployment
deployment URL for
onboarding resource training artifact URL
inferencing
group scripts) automatically)
4.5 AI Core:
AI Core:Trigger
Triggerthe
the Serving
serving Execution
execution
0.4 AI Core: Connect a GitHub repo
4.6 AI Launchpad: Check the status of
AI Launchpad: Check the status of
0.5 AI Core: Connect a Docker repo deployment,check
deployment, checkthe
the logs
logs
0 2 3 4 5 6
Get the pre-
Use
AI Core and Create a trained model Get
Submit for deployment
AI Launchpad resource from open deployment
deployment URL for
onboarding group source URL
inferencing
framework
and register it
as artifact
AI Core one-time initial configuration ML model Inferencing
SAP-specific
▪ AI Core help page
▪ AI Launchpad help page
▪ SAP BTP help page
▪ SAP AI Portfolio blog post
Technical Links
▪ VS code plugin for AI Core
▪ SAP Computer Vision Package
▪ Metaflow library for SAP AI Core
Tutorials
▪ SAP HANA Academy video series about the AI Core and AI
Launchpad onboarding
▪ SAP HANA Academy video series about Defect Detection with
TensorFlow in AI Core
▪ Create Your First Machine Learning Project using SAP AI Core
▪ Use Pre-Trained TensorFlow Models with a GPU in SAP AI Core
▪ Create an Object Detection Project using Computer Vision Package
and SAP AI Core
© 2022 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 16
Defect detection serving with SAP AI Core SDK
Useful links
Data science-specific
▪ https://www.britannica.com/technology/artificial-intelligence
▪ https://ml-ops.org/content/end-to-end-ml-workflow
▪ https://www.v7labs.com/blog/image-segmentation-guide
▪ https://towardsdatascience.com/a-comprehensive-guide-to-
convolutional-neural-networks-the-eli5-way-3bd2b1164a53
▪ https://www.v7labs.com/blog/data-preprocessing-guide
▪ https://resources.github.com/ci-cd/
▪ https://ubuntu.com/blog/kubernetes-versus-docker
▪ https://www.v7labs.com/blog/what-is-computer-vision
▪ https://neptune.ai/blog/performance-metrics-in-machine-learning-
complete-guide
▪ https://www.analyticsvidhya.com/blog/2017/09/common-machine-
learning-algorithms/
© 2022 SAP SE or an SAP affiliate company. All rights reserved. ǀ PUBLIC 17
Thank you.
Contact information:
open@sap.com
Follow all of SAP
www.sap.com/contactsap