Skip to content

NTCIR-19 ModelRetrieval

NTCIR-19 ModelRetrieval Task Overview

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:

Register for NTCIR-19

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.