Feature Request: `Implement Threshold-Consistent Margin Loss for Open-World Deep Metric Learning in TF-GNN`

I propose adding the Threshold-Consistent Margin Loss (TCM) function to the TF-GNN library. TCM is a novel loss function specifically designed for open-world deep metric learning, which has shown significant improvements in handling unseen classes and imbalanced data compared to traditional loss functions.

Motivation:

Open-world scenarios: Many real-world applications involve open-world scenarios where new classes can emerge over time. TCM is well-suited for these challenges.
Improved performance: TCM has demonstrated superior performance in terms of accuracy and robustness compared to other loss functions in open-world settings.
Community benefit: Incorporating TCM into TF-GNN will benefit the broader machine learning community by providing a powerful tool for addressing open-world problems.
Implementation details:

Function definition: Implement the TCM loss function as a TensorFlow operation.
Hyperparameters: Allow users to configure TCM hyperparameters (e.g., margin, temperature) to fine-tune the loss.
Integration: Integrate TCM with existing TF-GNN components for seamless usage.
Documentation: Provide clear documentation and examples to guide users in using TCM effectively.
Additional notes:

Consider providing pre-trained models or transfer learning options to accelerate development.
Explore opportunities for optimization and performance improvements.

By incorporating TCM into TF-GNN, we can significantly enhance the library's capabilities for open-world deep metric learning and empower researchers and developers to tackle challenging real-world problems.

Paper

No files, tests, or entry points are named. Start by reading the linked paper and locating TF-GNN's existing loss-function and component-integration APIs; done means a TCM operation with configurable margin and temperature, integrated into TF-GNN and accompanied by documentation and examples.