NTCIR-19 ModelRetrieval¶

Pre-trained Model Retrieval Task
Advances in AI have led to a proliferation of pre-trained machine learning models across many domains. However, choosing the most suitable model for a new task remains time-consuming and costly.
This task defines a benchmark for efficiently retrieving the pre-trained model that best matches a user's problem, reducing selection overhead and accelerating downstream deployment.
Ready to participate? Start by registering:
Motivation¶
Practitioners currently navigate a vast space of models using manual trial-and-error, which incurs high computational and labor costs. This task aims to:
- Encourage development of retrieval algorithms that predict model performance without exhaustive experimentation.
- Provide a standardized benchmark for fair comparison.
- Lower the barrier to entry for researchers in this field.