Sustainability Data Science Conference Awards

We're absolutely thrilled to celebrate the recipients of the Best Presentation Awards at the Sustainability Data Science Conference! 

• Anna Lukkarinen—Best Postdoc Presentation (top photo)

• Mohammad Aljubran—Best Student Presentation (middle photo)

• Andy Huynh—Best Student Presentation-2nd place (bottom photo)

Their passion for sustainability and data science is truly inspiring and we wholeheartedly appreciate their dedication and are eagerly anticipating their future accomplishments.

Anna's Presentation Title: Investor Reactions to Firm Sustainability—Evidence from a Randomized Field Experiment

Presentation Summary (Takeaways):

• Question: How do investors react to firms’ sustainability cues?

• Method: Randomized field experiment

• Result: Reactions overall negative but vary by investor segment

• Implications: Firms can mitigate negative reactions

Mohammad's Presentation Title: Thermal Earth Model for the Continental United States Using an Interpolative Physics-Informed Graph Neural Network (InterPIGNN)

Presentation Summary (Takeaways):

• We aggregated a dataset with over 400,000 temperature-at-depth measurements and over 20 other physical quantities spanning the continental United States.

• We developed a novel interpolative graph neural network (InterNet) for point cloud interpolation tasks, which we used to train a physics-informed model (InterPIGNN).

• We achieved state-of-the-art performance in predicting temperature-at-depth with mean absolute error of 4.8 °C, ten folds less than the best model available in the literature.

Andy's Presentation Title: Contrastive ground-level image and remote sensing pre-training improves representation learning for natural world imagery

Presentation Summary (Takeaways):

• We leverage the relationship between species’ characteristics and their environment to build a new pre-training task for natural world imagery

• CRISP multi-view pre-training improves performance on fine-grained visual classification tasks, especially rare species and undersampled regions

• The benefit of our multi-view pre-training approach extends to ambient data collections available across the world

Photo credits: David Gonzales Photos

This article is a part of the Stanford Data Science legacy publication. Read more about the HAI and Stanford Data Science merger.