A Kaggle research prediction competition hosted by Google
Google - Fast or Slow? Predict AI Model Runtime
Train a model that predicts runtime rankings for different tile configurations when executing machine learning models on TPUs. Since a machine learning model can be represented as a graph, with nodes being operations and edges being tensors, a GNN model can be trained to predict the execution times of different configurations and rank them accordingly. A detailed description of the problem is described in the paper here.
A Jupyter Notebook that offers a solution to the above challenge via using an ensemble of Gradient Boosting Machines to predict future sales.
This notebook is also hosted on Kaggle, so you can run and or edit the code yourself for free.