A5021557939- Multi-Agent Systems and Negotiation
- Logic, Reasoning, and Knowledge
- Business Process Modeling and Analysis
- Service-Oriented Architecture and Web Services
- Semantic Web and Ontologies
- Advanced Software Engineering Methodologies
- AI-based Problem Solving and Planning
- Bayesian Modeling and Causal Inference
- Natural Language Processing Techniques
- Software Engineering Techniques and Practices
- Advanced Database Systems and Queries
- Topic Modeling
- Complex Network Analysis Techniques
- Opinion Dynamics and Social Influence
- Multimodal Machine Learning Applications
- Model-Driven Software Engineering Techniques
- Data Management and Algorithms
- Distributed systems and fault tolerance
- Formal Methods in Verification
- Advanced Neural Network Applications
- Information Technology Governance and Strategy
- Speech Recognition and Synthesis
- Access Control and Trust
- Logic, programming, and type systems
- Game Theory and Applications
Google (United States)
2020
Brain Innovation (Netherlands)
2020
University of Luxembourg
2013-2017
Google (Switzerland)
2017
Utrecht University
2013
State-of-the-art machine learning methods exhibit limited compositional generalization. At the same time, there is a lack of realistic benchmarks that comprehensively measure this ability, which makes it challenging to find and evaluate improvements. We introduce novel method systematically construct such by maximizing compound divergence while guaranteeing small atom between train test sets, we quantitatively compare other approaches for creating generalization benchmarks. present large...
While mainstream machine learning methods are known to have limited ability compositionally generalize, new architectures and techniques continue be proposed address this limitation. We investigate state-of-the-art in order assess their effectiveness improving compositional generalization semantic parsing tasks based on the SCAN CFQ datasets. show that masked language model (MLM) pre-training rivals SCAN-inspired primitive holdout splits. On a more complex task, we leads significant...
Recent neural network-based language models have benefited greatly from scaling up the size of training datasets and number parameters in themselves. Scaling can be complicated due to various factors including need distribute computation on supercomputer clusters (e.g., TPUs), prevent bottlenecks when infeeding data, ensure reproducible results. In this work, we present two software libraries that ease these issues: $\texttt{t5x}$ simplifies process building large at scale while maintaining...
We apply an existing formal framework for practical reasoning with arguments and evidence to the Goal-oriented Requirements Language (GRL), which is part of User Notation (URN). This serves as a rationalization elements in GRL model: using attack relations between we can automatically compute acceptability status model, based on their underlying evidence. integrate into metamodel set out research further develop this framework.
The aim of this paper is to introduce and validate a logic-based framework that serves as the underlying model for recently introduced formalism capturing enterprise architecture design decisions by Plataniotis et al. Our working hypothesis knowledge in terms will enable consistency checks rationales advanced impact/what-if analysis when confronted with changes. We formalize set integrity constraints, which allow guidance decision during creation provide means perform checks. apply our...
In many logic-based BDI agent programming languages, plan selection involves inferencing over some underlying knowledge representation. While context-sensitive facilitates the development of flexible, declarative programs, overhead evaluating repeated queries to agent's beliefs and goals can result in poor run time performance. this paper we present an approach multi-cycle query caching for languages. We extend abstract performance model presented (Alechina et al. 2012) quantify costs...
Goal-oriented requirements modeling approaches aim to capture the intentions of stakeholders involved in development an information system as goals and tasks. The process constructing such goal models usually involves discussions between a engineer group stakeholders. Not all arguments can be captured or tasks: e.g., discussion whether accept reject certain rationale for acceptance rejection cannot models. In this paper, we apply techniques from computational argumentation approach by using...
Is an actor typically considered a human being? What about autonomous entity? We investigate the typical feature structure of common modeling concepts in order to create empirically grounded description semantic that people implicitly use while reasoning about, and with such concepts. Apart from insights into concept this work presents, consequences for quality models languages are discussed. finally discuss more detail how process modeling, especially when it involves multiple different...