UiB, UiT Machine Learning Group and CAIR with papers accepted to AAAI 2021
The AAAI (Association for the Advancement of Artificial Intelligence) 2021 Conference had 9034 submissions. Only 21% of the papers were accepted. Among them were papers from NORA partners UiB, UiT and CAIR (UiA).
The paper Measuring Dependence with Matrix-based Entropy Functional by the UiT Machine Learning Group, in collaboration with NEC Labs Europe and the University of Florida, was one of the papers that made the cut.
- This work proposes entirely new ways to quantify dependencies among random variables, says Professor and co-author of the paper Robert Jenssen at the UiT Machine Learning Group, and he adds - This is important for instance for advancing deep learning research. I am very happy that we as a Norwegian machine learning research group can contribute at the highest international level within our field.
The Centre for Artificial Intelligence Research (CAIR) paper Human-Level Interpretable Learning for Aspect-Based Sentiment Analysis was also accepted at AAAI 2021. The paper proposes a human-interpretable learning approach for aspect-based sentiment analysis (ABSA), employing the recently introduced Tsetlin Machines (TMs). The interpretability is attained by converting the intricate position-dependent textual semantics into binary form, mapping all the features into bag-of-words (BOWs). The experiments show how each relevant feature takes part in conjunctive clauses that contain the context information for the corresponding aspect word, demonstrating human-level interpretability. At the same time, the obtained accuracy is on par with existing neural network models, reaching 78.02% on Restaurant 14 and 73.51% on Laptop 14. Congratulations to UiAs Rohan Yadav and supervisor Lei Jiao, with co-authors Morten Goodwin.
UiB had no less than three papers accepted for the conference. These were:
Living Without Beth and Craig: Explicit Definitions and Interpolants in Description Logics with Nominals, authored by Alessandro Artale, Jean Jung, Andrea Mazzullo, Ana Ozaki and Frank Wolter
- The Craig interpolation property (CIP) states that an interpolant for an implication exists iff it is valid. The Beth definability property (BDP) states that an explicit definition exists iff a formula stating implicit definability is valid. Thus, they transform potentially hard existence problems into deduction problems in the underlying logic. Description Logics with nominals do not have the CIP nor BDP, but in particular, deciding and computing explicit definitions of concepts or individuals has many potential applications in ontology engineering and ontology-based data management. In this article we show two main results: even without Craig and Beth, the existence of interpolants and explicit definitions is decidable in the description logics with nominals ALCO and ALCIO. However, living without Craig and Beth makes this problem harder than deduction: we prove that the existence problems become 2ExpTime-complete, thus one exponential harder than validity.
Mining EL Bases with Adaptable Role Depth, authored by Ricardo Guimarães, Ana Ozaki, Cosimo Persia and Baris Sertkaya
- Description Logic (DL) knowledge bases are one of the most prominent ways to formalise and share knowledge without ambiguity. They have been particularly successful in fields rich in terminological knowledge such as Biology, Medicine, and Manufacturing. Although the tools supporting ontology creation and maintenance have evolved over the years, building ontologies is still a demanding task. In particular, it involves not only ontology engineers, but domain experts as well. Moreover, the process of modelling knowledge, even with modern methodologies remains time-consuming. In some cases, it is possible to build an ontology from a structure (e.g. a database or knowledge graph) automatically. By adapting notions from Formal Concept Analysis to DLs one extract rules in the form of concept inclusions. In DLs, these concept inclusions can involve arbitrarily large expressions via nesting. Thus, it is not clear whether a finite base exists and, if so, how large concept expressions may need to be. We first revisit results in the literature for mining ontologies from finite interpretations in the description logic EL. Those mainly focus on finding a finite base while fixing the role depth but potentially losing some concept inclusions (with larger concepts) that hold in the interpretation. Then, we present a new strategy for mining EL bases that is adaptable in the sense that it can bound the role depth of concepts depending on the local structure of the interpretation without losing EL concept inclusions that hold in the interpretation.
Present-Biased Optimization, authored by Fedor Fomin, Pierre Fraigniaud and Petr Golovach.
- This paper explores the behavior of present-biased agents, that is, agents who erroneously anticipate the costs of future actions compared to their real costs. Specifically, the paper extends the original framework proposed by Akerlof (1991) for studying various aspects of human behavior related to time-inconsistent planning, including procrastination, and abandonment, as well as the elegant graph-theoretic model encapsulating this framework recently proposed by Kleinberg and Oren (2014). The benefit of this extension is twofold. First, it enables to perform fine grained analysis of the behavior of present-biased agents depending on the optimisation task they have to perform. In particular, we study covering tasks vs. hitting tasks, and show that the ratio between the cost of the solutions computed by present-biased agents and the cost of the optimal solutions may differ significantly depending on the problem constraints. Second, our extension enables to study not only underestimation of future costs, coupled with minimization problems, but also all combinations of minimization/maximization, and underestimation/overestimation. We study the four scenarios, and we establish upper bounds on the cost ratio for three of them (the cost ratio for the original scenario was known to be unbounded), providing a complete global picture of the behavior of present-biased agents, as far as optimisation tasks are concerned.
For more information, follow this link: https://aaai.org/Conferences/AAAI-21/