BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning the Difference that Makes a Difference with Counterfactua
 lly Augmented Data - Zachary Lipton\, Carnegie Mellon University
DTSTART:20220210T150000Z
DTEND:20220210T160000Z
UID:TALK169808@talks.cam.ac.uk
CONTACT:Marinela Parovic
DESCRIPTION:Despite alarm over the reliance of machine learning systems on
  so-called spurious patterns\, the term lacks coherent meaning in standard
  statistical frameworks. However\, the language of causality offers clarit
 y: spurious associations are due to confounding (e.g.\, a common cause)\, 
 but not direct or indirect causal effects. Inspired by this literature (an
 d borrowing from it gesturally)\, we address natural language processing\,
  introducing methods and resources for training models less sensitive to s
 purious patterns. Given documents and their initial labels\, we task human
 s with revising each document so that it (i) accords with a counterfactual
  target label\; (ii) retains internal coherence\; and (iii) avoids unneces
 sary changes. Interestingly\, on sentiment analysis and natural language i
 nference tasks\, classifiers trained on original data fail on their counte
 rfactually-revised counterparts and vice versa. Classifiers trained on com
 bined datasets perform remarkably well\, just shy of those specialized to 
 either domain. I will discuss this method\, the early results\, some conce
 ptual underpinnings of the approach\, and some recent follow-up work.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR