Lecture Notes in Artificial Intelligence 14294
Lecture Notes in Artificial Intelligence 14294
Lecture Notes in Artificial Intelligence 14294
Series Editors
Randy Goebel, University of Alberta, Edmonton, Canada
Wolfgang Wahlster, DFKI, Berlin, Germany
Zhi-Hua Zhou, Nanjing University, Nanjing, China
The series Lecture Notes in Artificial Intelligence (LNAI) was established in 1988 as a
topical subseries of LNCS devoted to artificial intelligence.
The series publishes state-of-the-art research results at a high level. As with the LNCS
mother series, the mission of the series is to serve the international R & D community
by providing an invaluable service, mainly focused on the publication of conference and
workshop proceedings and postproceedings.
Zied Bouraoui · Srdjan Vesic
Editors
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2024
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors
or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Program Committee
Additional Reviewers
Jesse Davis
Tree ensembles such as (gradient) boosted trees and random forests are a popular class of
models that are often used in practice. Unfortunately, merely achieving good predictive
performance is insufficient for a deployed model because it is important to assess other
factors such as a model’s robustness and explainability. However, like other expressive
model classes (e.g., neural networks), it is challenging to learn robust models where
decisions can be explained. For example, it is often possible to flip an example’s predicted
label by applying a tiny, specifically constructed perturbation. This type of behavior is
undesirable because it degrades a model’s performance and erodes a user’s trust in the
model. This talk will argue that the solution to this problem is to develop techniques
that are able to reason about a learned model’s behavior. Moreover, I will advocate that
using such approaches is a key part of evaluating learning pipelines because it can help
debug learned models and the data used to train them. I will present two approaches for
gaining insight into how a model will behavior. First, I will discuss a generic approach for
verifying whether a learned tree ensemble exhibits a wide range of behaviors. Second,
I will describe an approach that identifies whether the tree ensemble is at a heightened
risk of making a misprediction in a post-deployment setting. Throughout the talk I will
use several illustrative examples from real-world applications, with an emphasis on
applications in professional soccer.
Mixing Time and Uncertainty. A Tale of Superpositions
Rafael Peñaloza
Formalisms capable of dealing with time and uncertainty are necessary for modelling
the existing knowledge of (business) processes which must interact with an unreliable
environment. Yet, combining time and uncertainty is far from trivial and can easily lead
to undecidability, making those formalisms useless in practice. A recent proposal for
probabilistic temporal logic uses the idea of quantum superposition, where an object
simultaneously has and does not have a property, until it is observed. We apply this
superposition semantics to Linear Temporal Logic, and show how it can be used for
Business Process Modelling tasks.
On Belief Update According to Katsuno & Mendelzon:
Novel Insights
Eduardo Fermé
The aim of Belief Change Theory is to provide a formal framework for understanding
how an agent’s beliefs evolve in response to new evidence. Over the past 35 years, various
operators have been proposed to handle different types of situations and evidence. The
core of this theory consists of belief revision operators, which are designed to update
an agent’s beliefs based on more reliable evidence. The standard model is the AGM
revision, proposed by Alchourrón, Gärdenfors and Makinson.
Another important class of operators are update operators proposed by Katsuno and
Mendelzon in 1991 (KM-update). The difference between revision and update operators
is that revision operators aim to correct an agent’s beliefs, whereas update operators aim
to incorporate the results of a change in the world, without presuming that the agent’s
previous beliefs were incorrect. This difference is often summarized as belief revision
being concerned with changing beliefs in a static world, while update is concerned with
the evolution of beliefs in a dynamic world.
In this presentation, we will showcase recent research that revolves around the KM-
update model of belief change.
1. The model’s efficacy in accurately capturing changes occurring in the world. We will
introduce some philosophical and technical aspects on this point. KM-update assumes
that any situation can be updated into one satisfying that input, which is unrealistic.
To solve this problem, we must relax either the success or the consistency principle.
We propose and characterize a model where not all the input are “reachable”.
2. The interconnection between KM update and AGM revision. We will examine the
relationship between these two approaches.
3. The iteration of update. We will explore the methodology of incorporating iterative
updates, drawing inspiration from the work of Darwiche and Pearl of iterated AGM
revision.
By delving into these areas, we aim to provide a comprehensive understanding of
KM-Update and its associated research developments.
Contents
Decision Theory
Argumentation Systems
Bayesian Networks