Traditional machine learning on tabular data from crop yields and soil chemistry to ecological monitoring often requires extensive feature engineering, bespoke hyperparameter tuning, and large training sets that simply don’t exist in emerging field studies.

This interactive, hands-on workshop will introduce you to Google’s TabFM, an open-source/open-weight tabular foundation model for zero-shot classification and regression. You will work with agricultural and environmental datasets to explore how TabFM can be applied to scientific prediction tasks, particularly in data-limited research settings.

The session will provide practical experience with TabFM and demonstrate its potential for accelerating data-driven research in agriculture and environmental science.

• Practical, Notebook-Driven Workflows: We will walk through executable Jupyter/Colab notebooks using real-world agricultural and environmental examples.

• No Software Engineering Background Required: The session focuses on applying pretrained models using standard Python data-science libraries (pandas, standard APIs).

• Benchmarking Against Domain Baselines: See how TabFM compares against traditional research staples like Random Forests, XGBoost, and standard linear models.

Who Should Join?

• University researchers, postdoctoral scholars, and graduate students.

• Industry R&D scientists.

• Applied data scientists collaborating with scientific domain specialists.

• Anyone using or making predictions with tabular data! Examples will be grounded in Agronomy, Ecology, Environmental Sciences, Geospatial Analytics, etc, but workflows can be adopted to any domain!

Prior familiarity with basic Python and tabular data manipulation (e.g., pandas DataFrames) is helpful, but no deep learning or engineering background is assumed.