[CUBIG DTS] Privacy-Safe Synthetic Data for Text, Tabular, and Images

DTS AI Module - Usage Instructions

Quick Start

Step 1: Run the Container

docker run -d
--name dts-ai-module
--gpus all
-p 8000:8000
-v /path/to/your/data:/data
-v /path/to/output:/results
709825985650.dkr.ecr.us-east-1.amazonaws.com/cubig-ai/dts-v1:1.0.2

Replace /path/to/your/data and /path/to/output with your actual host paths.

Step 2: Verify Health

Wait 30 seconds for service startup:

curl http://localhost:8000/health 

Expected response: { "status": "healthy", "service": "DTS AI Module", "version": "1.0.2" }

Step 3: Access Documentation

• Quick Start: http://localhost:8000/ 
• API Docs: http://localhost:8000/docs  (Swagger UI)
• User Guide: docker exec dts-ai-module cat /app/USER_GUIDE.md

API Usage Examples

Example 1: Image Synthetic Data

Generate 100 synthetic medical X-ray images:

curl -X POST http://localhost:8000/generate 
-H "Content-Type: application/json"
-d '{ "run_id": "medical_xray_001", "modality": "image", "private_data_path": "/data/xray_images", "num_samples": 100, "iteration": 2, "epsilon": 8.0, "delta": 1e-6, "positive_prompt": "a medical chest x-ray image, high quality", "image_width": 512, "image_height": 512, "gpu_id": "0" }'

Result: Synthetic images saved to /results/medical_xray_001/final/synthetic_final/

Example 2: Text Synthetic Data

Generate 1,000 synthetic medical texts:

curl -X POST http://localhost:8000/generate 
-H "Content-Type: application/json"
-d '{ "run_id": "medical_text_001", "modality": "text", "private_data_path": "/data/diagnosis_texts.csv", "num_samples": 1000, "iteration": 3, "epsilon": 8.0, "domain": "medical", "sub_domain": "diagnosis", "gpu_id": "0" }'

Result: Synthetic texts saved to /results/medical_text_001/final/synthetic_final.csv

Example 3: Tabular Synthetic Data

Generate 5,000 synthetic customer records:

curl -X POST http://localhost:8000/generate 
-H "Content-Type: application/json"
-d '{ "run_id": "customer_data_001", "modality": "tabular", "private_data_path": "/data/customer_dataset.csv", "num_samples": 5000, "iteration": 2, "epsilon": 8.0, "numerical_columns": ["age", "income", "credit_score"], "categorical_columns": ["gender", "education", "occupation"], "gpu_id": "0" }'

Result: Synthetic data saved to /results/customer_data_001/final/synthetic_final.csv

Privacy Parameters Guide

Epsilon - Privacy Budget

• Lower values: Stronger privacy protection, lower data utility
• Higher values: Weaker privacy protection, higher data utility
• Recommended: 8.0 for balanced privacy-utility tradeoff
• Range: 0.1 (very strong) to 10.0 (moderate)

Delta - Privacy Failure Probability

• Recommended: 1e-6 (should be less than 1/N, where N is number of samples)

Iteration - Quality Improvement

• 1: Fast, lower quality
• 2-3: Balanced (recommended)
• 4-5: Slower, higher quality

Troubleshooting

Issue: Connection refused to port 8000 Solution: Wait 30-60 seconds for service startup, then retry

Issue: GPU not found Solution: Verify NVIDIA Container Toolkit: docker run --gpus all nvidia/cuda:12.1.0-base-ubuntu22.04 nvidia-smi

Issue: private_data_path not found Solution: Use container internal path (/data/...), not host path

Issue: Out of memory Solution: Reduce num_samples, iteration, or image dimensions

Support

For technical support:

1. Check API documentation at http://localhost:8000/docs 
2. Review user guide: docker exec dts-ai-module cat /app/USER_GUIDE.md
3. Contact AWS Marketplace support