BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:The Extremes of Interpretability in Machine Learning: Sparse Decis
 ion Trees and Scoring Systems and Interpretable Neural Networks
DTSTART:20210507T161500
DTEND:20210507T173000
DTSTAMP:20260609T084011Z
UID:ae20debec9eab4586c593f54d36ed60a293dbf31327f6f9add5f8b2e
CATEGORIES:Conferences - Seminars
DESCRIPTION:Cynthia Rudin\, Duke\nWith widespread use of machine learning\
 , there have been serious societal consequences from using black box model
 s for high-stakes decisions\, including flawed bail and parole decisions i
 n criminal justice\, flawed models in healthcare\, and black box loan deci
 sions in finance. Transparency and interpretability of machine learning mo
 dels is critical in high stakes decisions. In this talk\, I will focus on 
 two of the most fundamental and important problems in the field of interpr
 etable machine learning: optimal sparse decision trees and optimal scoring
  systems. I will also briefly describe work on interpretable neural networ
 ks for computer vision.\n\nOptimal sparse decision trees: We want to find 
 trees that maximize accuracy and minimize the number of leaves in the tree
  (sparsity). This is an NP hard optimization problem with no polynomial ti
 me approximation. I will present the first practical algorithm for solving
  this problem\, which uses a highly customized dynamic-programming-with-bo
 unds procedure\, computational reuse\, specialized data structures\, analy
 tical bounds\, and bit-vector computations.\n\nOptimal scoring systems:  
 Scoring systems are sparse linear models with integer coefficients. Tradit
 ionally\, scoring systems have been designed using manual feature eliminat
 ion on logistic regression models\, with a post-processing step where coef
 ficients have been rounded. However\, this process can fail badly to produ
 ce optimal (or near optimal) solutions. I will present a novel cutting pla
 ne method for producing scoring systems from data. The solutions are globa
 lly optimal according to the logistic loss\, regularized by the number of 
 terms (sparsity)\, with coefficients constrained to be integers. Predictiv
 e models from our algorithm have been used for many medical and criminal j
 ustice applications\, including in intensive care units in hospitals.\n\nI
 nterpretable neural networks for computer vision: We have developed a neur
 al network that performs case-based reasoning. It aims to explains its rea
 soning process in a way that humans can understand\, even for complex clas
 sification tasks such as bird identification.\n\nPapers:\nJimmy Lin\, Chud
 i Zhong\, Diane Hu\, Cynthia Rudin\, Margo Seltzer\nGeneralized and Scalab
 le Optimal Sparse Decision Trees. ICML\, 2020.\n\nBerk Ustun and Cynthia R
 udin\nLearning Optimized Risk Scores. JMLR\, 2019. Shorter version at KDD 
 2017.\n\nStruck et al. Association of an Electroencephalography-Based Risk
  Score With Seizure Probability in Hospitalized Patients. JAMA Neurology\,
  2017.\n\nChaofan Chen\, Oscar Li\, Chaofan Tao\, Alina Barnett\, Jonathan
  Su\,\nCynthia Rudin This Looks Like That: Deep Learning for Interpretable
  Image Recognition. NeurIPS\, 2019.\n 
LOCATION:https://epfl.zoom.us/j/85679229059
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR