BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:A Sea of Words: An In-Depth Analysis of Anchors for Text Data
DTSTART:20250624T120000Z
DTEND:20250624T130000Z
DTSTAMP:20260509T204500Z
UID:indico-event-14399@indico.math.cnrs.fr
DESCRIPTION:Speakers: Damien Garreau (Julius-Maximilians-Universität Wür
 zburg)\n\nAnchors is a post-hoc\, rule-based interpretability method. For 
 text data\, it proposes to explain a decision by highlighting a small set 
 of words (an anchor) such that the model to explain has similar outputs wh
 en they are present in a document. In this paper\, we present the first th
 eoretical analysis of Anchors\, considering that the search for the best a
 nchor is exhaustive. After formalizing the algorithm for text classificati
 on\, we present explicit results on different classes of models when the v
 ectorization step is TF-IDF\, and words are replaced by a fixed out-of-dic
 tionary token when removed. Our inquiry covers models such as elementary i
 f-then rules and linear classifiers. We then leverage this analysis to gai
 n insights on the behavior of Anchors for any differentiable classifiers. 
 For neural networks\, we empirically show that the words corresponding to 
 the highest partial derivatives of the model with respect to the input\, r
 eweighted by the inverse document frequencies\, are selected by Anchors.\n
  \nPaper: https://proceedings.mlr.press/v206/lopardo23a.html\n\nhttps://
 indico.math.cnrs.fr/event/14399/
URL:https://indico.math.cnrs.fr/event/14399/
END:VEVENT
END:VCALENDAR