How ml4sci/cms works
Main Components
a. Data Processing Pipeline
• Frame Producers (Input Layer)
• Various Taggers (Processing Layer)
• TensorFlow Models (Inference Layer)
b. Core Components
• Data Formats Module
• Frame Producers
• Multiple Specialized Taggers:
• EG Tagger
• QG Tagger
• Tau Tagger
• Top Tagger
System Architecture Representation
[Data Input Layer]
• ↓
[Frame Processing Layer]
• ↓
[Tagger Layer]
• ↓
[Model Inference Layer]
• ↓
[Output Layer]
Specific Elements
a. Data Flow Components:
• Input data sources
• Frame Collections
• Detector data streams
b. Processing Components:
• DetFrameProducer
• EGFrameProducer
• JetFrameProducer
• Frame Cropping/Striding utilities
c. ML Components:
• TensorFlow model integration
• ResNet models (show different channel versions)
• Inference paths
d. Tagger Components:
• EG, QG, Tau, and Top taggers
• Their respective configurations
• Inference paths
Key Architectural Elements to Highlight
• The modular nature of the taggers
• The shared frame production infrastructure
• The ML model integration points
• The configuration system
• The build system integration (BuildFile.xml locations)
Technical Details
• Interface boundaries between components
• Data format transformations
• Configuration injection points
• Resource requirements
• Performance-critical paths