Inference the simple perceptron onnx model

[chris-chris]

Goal

• Implement simple perceptron on mina-zkml library

Reference Code

Description

implement save & load function to save & load compiled model.

Task Process

1. Open PR for the issue and create branch.
2. Write down technical specification to accomplish the task.
3. Comment down your estimated hours to complete the task.
4. I'll comment the estimated hours as well
5. Start the task
6. Track your time consumption while working
7. Comment out your actual time consumed to complete the task

[sshivaditya2019]

Objectives

• Extend existing ONNX parser with Kimchi Library support for zkSNARK circuit generation
• Implement tensor operations (MatMul, ReLU) compatible with Kimchi constraints
• Add support for tensor datum types including Opaque type
• Create a working perceptron implementation using the extended system
• Write comprehensive tests for the new functionality

Key Components

Tensor Operations Extension

The following operations need Kimchi circuit implementations:

• MatMul operation
• ReLU activation
• Addition/Subtraction for bias
• Shape transformation operations

Datum Types

New tensor datum types to be supported:

• Opaque type for Kimchi-specific data
• Circuit-compatible numeric types
• Conversion utilities between existing and new types

SupportedOps Extension

Extend the existing SupportedOps enum with:

• Circuit-compatible Input operation
• Circuit-compatible MatMul
• Circuit-compatible ReLU
• Additional helper operations

Circuit Integration

Core Kimchi integration components:

• Constraint generation system
• Witness handling
• Proof generation interface
• Verification system

Implementation Steps

1. Core Circuit Infrastructure

• Tasks:
  • Create circuit generation primitives
  • Implement constraint system
  • Add witness handling
  • Set up proof generation framework
• Estimated Time: 25 hrs

2. Tensor Operations

• Tasks:
  • Extend existing tensor operations with circuit support
  • Implement MatMul circuit constraints
  • Add ReLU circuit implementation
  • Create shape transformation utilities
• Estimated Time: 30 hrs

3. Type System Extension

• Tasks:
  • Add Opaque type support
  • Implement conversion utilities
  • Create circuit-compatible numeric types
• Estimated Time: 15 hrs

4. SupportedOps Implementation

• Tasks:
  • Add circuit-compatible Input operation
  • Implement circuit MatMul operation
  • Add circuit ReLU operation
• Estimated Time: 20 hrs

5. Error Handling

• Tasks:
  • Extend error types for circuit operations
  • Add circuit-specific validation
  • Create debug utilities
• Estimated Time: 5 hrs

Testing Strategy

Unit Tests

• Test circuit generation for each operation
• Validate constraint satisfaction
• Verify proof generation and verification
• Estimated Time: 8 hrs

Integration Tests

• Test perceptron implementation
• Validate end-to-end proof generation
• Verify composite operations
• Estimated Time: 7 hrs

Time Estimate

Development Process

• Core Circuit Infrastructure: 25 hrs
• Tensor Operations: 30 hrs
• Type System Extension: 15 hrs
• SupportedOps Implementation: 20 hrs
• Error Handling: 5 hrs
• Total: 95 Hours

Testing

• Unit Tests: 8 hrs
• Integration Tests: 7 hrs
• Total: 15 hours

Total Project Time Estimate: 110 hours