**Optimizing Efficient Knowledge Graph Inference with Tempor

Source: Dev.to

Challenge Statement

Design a Temporal Graph Neural Network (T‑GNN) architecture that can efficiently process large‑scale knowledge graphs with millions of entities and relationships while incorporating temporal relationships between edges. The network should optimize a loss function that balances the accuracy of predicting temporal edge probabilities and the computational efficiency of inference.

Specific Constraints

• Knowledge graph size
  • 10 million entities
  • 100 million edges
  • 50 million temporal relationships (timestamps of edge creations or updates)
• Each node may have up to 50 edges, with varying degrees of node‑ and edge‑regularization.
• Model inference time must be < 10 minutes for a batch size of 1 024 samples.
• The network should capture both local and global graph patterns to improve temporal edge prediction accuracy.
• Use a combination of sparse matrix operations and Graph Attention Networks (GATs) to optimize computation and memory usage.

Evaluation Metrics

• Temporal edge prediction accuracy (e.g., AUC‑ROC)
• Model inference time (milliseconds per sample)
• Model complexity (number of parameters and FLOPS)
• Robustness to graph perturbations (e.g., node/edge removals)

Submission Requirements

• Provide a T‑GNN implementation in a popular deep learning framework (e.g., PyTorch, TensorFlow).
• Include a clear, reproducible experiment setup and report the performance metrics listed above.
• Be prepared to discuss design trade‑offs in the network architecture and evaluation strategy.

Submission Deadline: March 1st, 2026.