Mihail Popescu, University of Missouri

Abstract: Zadeh proposed the idea of computing with words (CWW) in 1996 (“fuzzy logic=computing with words”). However, in the last decade another paradigm of computing with words has emerged, based on embedding words in some Rn space. Word embedding has sprouted the current AI revolution based on a variety of transformer models, including large language models (LLMs). A classic example of computing of words equation using word embeddings is: king – man + woman ~ queen (note the soft equal,~, between the concepts). In this talk we will argue that using LLMs is computing with words. Moreover, we will also propose the concept of “computing with tokens” which is a generalization of “computing with words”. Tokes represent pieces of images or of signals obtained by tokenization (chucking) which are typically used in visual language models. Moreover, employing LLMs in medicine presents a series of challenges such that: privacy (data can’t be exposed to closed models), accuracy (high risk of “hallucinations”) and injection of specialized knowledge such as ontologies or knowledge graphs. We present three medical examples of LLMs and embedding-based soft computing that propose some answers to previous challenges.

Bio

Mihail Popescu is a Professor of Biomedical Informatics at the University of Missouri at Columbia. He received a BS in Nuclear Engineering from the Bucharest Polytechnic Institute (IPB) in 1987, an MS in Medical Physics in 1995, an MS in Electrical Engineering in 1997 and a PhD in Computer Science in 2003 from the University of Missouri at Columbia. Dr. Popescu has published over 200 refereed papers, 2 books and he holds 5 US patents. His research focus has been on medical intelligent systems including machine learning, image processing, and eldercare technologies.

Jozo Dujmovic, San Francisco State University

Abstract. Graded Logic (GL) is a propositional logic of human commonsense reasoning and decision making. GL is fully continuum-valued, i.e., everything is a matter of degree. It is based on continuum-valued logic variables (graded truth), continuum-valued simultaneity (graded conjunction), continuum-valued substitutability (graded disjunction), and continuum-valued importance of logic variables. The graded conjunction and the graded disjunction are dualized, complementary, and unified in a single continuum-valued, andness-directed, importance-weighted, idempotence-selectable, and annihilator-selectable fundamental logic function called Graded Conjunction/Disjunction (GCD). These properties are necessary for generating explainable results of the GL decision models.

A primary distinctive feature of GCD is its parameterized continuous transition that spans the full range, from the drastic conjunction (a model of ultimate simultaneity) to the drastic disjunction (a model of ultimate substitutability). To adjust the properties of graded conjunction and graded disjunction, GL employs an adjustable conjunction degree (“andness”) and an adjustable disjunction degree (“orness”). These concepts, introduced in 1973, mark the beginning of GL's development. This presentation reflects a half century of GL’s development and applications, coinciding with the publication of the presenter’s new book Graded Logic, Intelligent Systems Reference Library, Volume 273, Springer, August 2025.

Based on the strict use of continuum-valued concepts, Graded Logic can be situated within the broader context of soft computing, fuzzy systems, and computational intelligence. In soft propositional calculus, graded truth may be interpreted both as the degree of truth of a specific proposition and as the degree of membership in a fuzzy set defined by that proposition. Since fuzzy logic models are not andness-directed and typically rely on logic aggregators based on t-norms and t-conorms, Graded Logic can be regarded as an extension and generalization of fuzzy logic and fuzzy decision models across the full range of andness.

Graded logic is the necessary mathematical infrastructure of the Logic Scoring of Preference (LSP) decision method which has applicability in solving a variety of complex evaluation, and decision problems. Decision-makers evaluate competitive alternatives through perceived gradations of truth, importance, suitability, simultaneity, and substitutability. Each variable has a semantic identity, reflecting its role and meaning tied to the goals and interests of a specific decision-maker. LSP method is primarily used in professional decision-making problems which require explainability, optimization, and confidence analysis. Such decision problems are frequently encountered in healthcare, ecology, urban development, personal recommender systems, agriculture, engineering, and many other areas. This presentation will include both the theoretical concepts of Graded Logic and sample professional applications of the GL-based LSP method.

Bio

Prof. Dr. Jozo Dujmović is a Professor of Computer Science at San Francisco State University, where he has taught since 1994 and served as department chair from 1998–2002. He earned his Dipl. Ing. (1964) and M.Sc. (1973) and Sc.D. (1976) degrees in electronic and computer engineering from the University of Belgrade. His research focuses on soft computing, software metrics, and computer performance evaluation. In 1973, he introduced the concepts of andness and orness and developed the Logic Scoring of Preference (LSP) method for evaluating and optimizing complex systems. He has authored over 200 refereed publications, including 23 books and book chapters. 

Dr. Dujmović has held academic positions at the University of Belgrade, University of Florida, University of Texas at Dallas, Worcester Polytechnic Institute, and taught graduate courses in Argentina. He has also worked in industry and consulting, and founded SEAS, a soft computing decision-modeling company, in 1997, where he currently serves as Principal. He is a Life Senior Member of IEEE, recipient of four best paper awards, and has served as General Chair for major IEEE and ACM conferences.

Nigel G. Ward, University of Texas at El Paso

Abstract: While amazing AI agents are being developed, turning them into useful interactive collaborators for people remains a challenge.  Interaction through spoken language has promise, but today this is often on the clunky side, unlike from the swift and effortless nature of day-to-day human-human interaction.  One common cause is the implicit assumption that every referent and every fact is either known to the user or not.  In contrast, humans constantly probe and confirm what each other knows.  For example, I can say “let’s meet at EPCC” using uptalk-style prosody in English, and thereby acknowledging that you may not know what/where EPCC is, and implicitly  inviting you to request elaboration if needed.  To build systems that can do this, we need to model the likelihood of the user knowing certain things, of needing to know them, and of being able to figure them out from a brief mention.  These problems may not be amenable to the usual machine-learning techniques  that have succeeded so well for large language models.  We are exploring these challenges in the course of building a game-playing agent able to interact deftly with a human co-player in the face of novel items and obstacles and time pressure.  

Bio

Nigel G. Ward is Professor of Computer Science at the University of Texas at El Paso.  He received his Ph.D. from the University of California at Berkeley in 1991, with Lotfi A. Zadeh as one of his committee members.  He served on the faculty of the University of Tokyo for ten years, and 2015-2016 was a Fulbright Scholar and Visiting Professor at Kyoto University. He chaired the Speech Prosody Special Interest Group from 2018 to 2024. He is co-creator of the Prosody Tutorial video series and similar presentations for the ACL, Interspeech, and the LSA. and is the author of Prosodic Patterns in English Conversation.

Ward's research areas are at the intersection of spoken language and human-computer interaction. Current topics include the subtle non-lexical and prosodic signals that enable inference of a dialog partner's needs, intentions, and feelings at the sub-second level; and ways to model and exploit these phenomena to improve speech-to-speech translation systems, dialog systems, and language assessment and teaching. These projects apply multiple methods: statistical, qualitative, experimental, corpus-based modeling, and systems building.

Larry Lesser, University of Texas at El Paso

UTEP Distinguished Teaching Professor Dr. Larry Lesser will share ideas and examples of mathematical poetry