BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:RO-FIGS: Efficient and Expressive Tree-Based Ensembles for Tabular
  Data - Urška Matjašec (University of Cambridge)
DTSTART:20250306T130000Z
DTEND:20250306T140000Z
UID:TALK228847@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:Tree-based models are often robust to uninformative features a
 nd can accurately capture non-smooth\, complex decision boundaries. Conseq
 uently\, they often outperform neural network-based models on tabular data
 sets at a significantly lower computational cost. Nevertheless\, the capab
 ility of traditional tree-based ensembles to express complex relationships
  efficiently is limited by using a single feature to make splits. To impro
 ve the efficiency and expressiveness of tree-based methods\, we propose Ra
 ndom Oblique Fast Interpretable Greedy-Tree Sums (RO-FIGS). RO-FIGS builds
  on Fast Interpretable Greedy-Tree Sums\, and extends it by learning trees
  with oblique or multivariate splits\, where each split consists of a line
 ar combination learnt from random subsets of features. This helps uncover 
 interactions between features and improves performance. The proposed metho
 d is suitable for tabular datasets with both numerical and categorical fea
 tures. We evaluate RO-FIGS on 22 real-world tabular datasets\, demonstrati
 ng superior performance and much smaller models over other tree- and neura
 l network-based methods. Additionally\, we analyse their splits to reveal 
 valuable insights into feature interactions\, enriching the information le
 arnt from SHAP summary plots\, and thereby demonstrating the enhanced inte
 rpretability of RO-FIGS models. The proposed method is well-suited for app
 lications\, where balance between accuracy and interpretability is essenti
 al.\n\n"You can also join us on Zoom":https://cam-ac-uk.zoom.us/j/83400335
 522?pwd=LkjYvMOvVpMbabOV1MVTm8QU6DrGN7.1
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR