Role of different actors in AI Value Chain
A fully automated AI procurement process
AI-App pipeline: integrating framework to bring data, algorithms, and deployment tools together.
How it works
BonsAPPs will allow users to develop AI solutions to solve real-world industry challenges at the deep edge that requires a number of AI artifacts and services produced by different stakeholders. AI developers and platform providers will be accessible on the BonsAPPs services layer to respond to these challenges, leveraging on the Bonseyes AI Marketplace to deliver end-to-end, containerized, ready-to-integrate and re-usable solutions that will also be integrated and interoperable with the AI4EU platform.
INNOVATOR
Defines challenge
CHALLENGE
WORKFLOW
CHALLENGE
DATA SCIENTIST
Solves challenge
TRAINING
TOOLS
AI ASSET
BONSEYES AI MARKETPLACE
DEVELOPER
Deploys AI APP
DEPLOYMENT
TOOLS
AI APP
INTEGRATOR
Integrates AI APP
BENCHMARK
TOOLS
AI SOLUTION
BONSEYES AI MARKETPLACE
INNOVATOR
Defines challenge
CHALLENGE
WORKFLOW
CHALLENGE
DATA SCIENTIST
Solves challenge
TRAINING
TOOLS
AI ASSET
DEVELOPER
Deploys AI APP
DEPLOYMENT
TOOLS
AI APP
INTEGRATOR
Integrates AI APP
BENCHMARK
TOOLS
AI SOLUTION
BONSEYES AI MARKETPLACE
BONSEYES AI MARKETPLACE
INNOVATOR
Defines challenge
CHALLENGE WORKFLOW
CHALLENGE
DATA SCIENTIST
Solves challenge
TRAINING
TOOLS
AI ASSET
DEVELOPER
Deploys AI APP
DEPLOYMENT
TOOLS
AI APP
INTEGRATOR
Integrates AI APP
BENCHMARK
TOOLS
AI SOLUTION
BONSEYES AI MARKETPLACE
AI SERVICE LAYER
A Scalable Services Layer
For Real-World User-Focused AI Solutions
The BonsAPPs services will be accessible on the Bonseyes Marketplace and interoperable with the AI4EU community to respond to AI Challenges fitting the end users needs.
Bonseyes Functionalities
Challenges
AI Assets
AI Apps
AI Solutions
Developer Platforms
Services
Experimentation with Ai apps, platforms & PoCs
Security & licensing
Model optimization
Deployment on platforms & Benchmarking
Infrastructure
Services
Validators/Users
Challenges
AI Assets
AI Apps
AI Solutions
Developer Platforms
Edge Intelligence enablement requires a service layer
Fully integrated training and deployment workflows for AI Apps
We propose an end-to-end AI pipeline to develop and deploy Deep Neural Network solutions on embedded devices. We propose a modular AI pipeline architecture which contains four modular main tasks:
DATA
Collection
Training
Model compression
Hardware-in-the-Loop Deployment
Model optimization, deployment and brenchmarking
Solution integration
Model verification
Data
Collection and labeling of data. Data augmentation and export of training and benchmarking datasets
Training
Model training and compression
Infrastructure
Optimization
Benchmarking
Code optimization on target hardware
Deployment
Test Protocol
Protocol and testing results
Test Bias / Sensitivity Analysis
Model analysis for bias and sensitivity to noise
Integration
AI App integrated into system
Verification
Correct performance is verified
Evaluation Report Feedback for next KPI target
Verification Report
Enabling shift from Cloud to the Deep Edge
Transition from Cloud Centric to the Deep Edge Centric Model for End-Users
Optimization tools and infrastructure required for deployment of AI to resource-constrained devices:
Reduction in development time compared to monolithic system design methods
Reduction in cost of ownership related to training of deep learning models comparing to current training approaches designed for the cloud
Re-usability of AI assets/apps on the marketplace
Reduction in computational and memory requirement with current existing solutions for deep learning models on embedded systems
