We are pleased to announce the upcoming AI So-Go-Chi Research Seminar, which will feature an invited talk by Dr. Nicolas Schwind (National Institute of Advanced Industrial Science and Technology, Japan). The seminar will be held in a hybrid format on October 30 (Wednesday), from 13:00 to 15:00.

We warmly welcome all interested participants and encourage you to share this information with colleagues who may also be interested.

Date & Time: October 30, 13:00–15:00
Venue: Hybrid (On-site and Online)

• On-site: Room M153, Main Building,
  Research Institute of Electrical Communication (RIEC), Tohoku University
• Online (Zoom)
  • Meeting ID: 946 6723 1022
  • Passcode: 923253

Program

Dr. Nicolas Schwind (National Institute of Advanced Industrial Science and Technology, Japan)
Title: From Editing to Learning Boolean Classifiers: A Belief Change Perspective
Abstract:
This talk explores principled ways to modify and learn Boolean classifiers through the lens of Belief Change Theory. Both tasks share the same overarching goal: establishing rational foundations for classifier design.
The first part addresses the problem of editing Boolean classifiers, i.e., determining how a classifier should be modified when new pieces of evidence arise. We introduce rationality postulates inspired by belief revision and present families of edit operators that satisfy these principles. The second part turns to learning from iterated belief change. We show how improvement operators, a broad class of iterated belief change operators, can define a learning model for binary classification. We present learning and inference algorithms tailored to this framework, along with empirical results that highlight two main insights: iterated belief change provides an effective form of online learning, and the axiomatic foundations of belief change operators open promising perspectives for the study of classification.

We look forward to your participation.