A5064283689- Semantic Web and Ontologies
- Access Control and Trust
- Service-Oriented Architecture and Web Services
- Multi-Agent Systems and Negotiation
- Logic, Reasoning, and Knowledge
- Privacy-Preserving Technologies in Data
- Bayesian Modeling and Causal Inference
- Cryptography and Data Security
- Anomaly Detection Techniques and Applications
- Mobile Crowdsensing and Crowdsourcing
- Machine Learning and Data Classification
- Adversarial Robustness in Machine Learning
- Topic Modeling
- Explainable Artificial Intelligence (XAI)
- AI-based Problem Solving and Planning
- Spam and Phishing Detection
- Peer-to-Peer Network Technologies
- Privacy, Security, and Data Protection
- Recommender Systems and Techniques
- Data Quality and Management
- Advanced Database Systems and Queries
- Distributed Sensor Networks and Detection Algorithms
- Biomedical Text Mining and Ontologies
- Business Process Modeling and Analysis
- Sharing Economy and Platforms
Amazon (United Kingdom)
2022-2023
University of Aberdeen
2009-2021
Özyeğin University
2012-2021
Prism Clinical Research
2020
Cardiff University
2018
Boğaziçi University
2005-2009
Deterministic neural nets have been shown to learn effective predictors on a wide range of machine learning problems. However, as the standard approach is train network minimize prediction loss, resultant model remains ignorant its confidence. Orthogonally Bayesian that indirectly infer uncertainty through weight uncertainties, we propose explicit modeling same using theory subjective logic. By placing Dirichlet distribution class probabilities, treat predictions net opinions and function...
Deep neural networks have achieved near-human accuracy levels in various types of classification and prediction tasks including images, text, speech, video data. However, the continue to be treated mostly as black-box function approximators, mapping a given input output. The next step this human-machine evolutionary process - incorporating these into mission critical processes such medical diagnosis, planning control requires level trust association with machine Typically, statistical...
Deep neural networks are often ignorant about what they do not know and overconfident when make uninformed predictions. Some recent approaches quantify classification uncertainty directly by training the model to output high for data samples close class boundaries or from outside of distribution. These use an auxiliary set during represent out-of-distribution samples. However, selection creation such is non-trivial, especially dimensional as images. In this work we develop a novel network...
Large language models (LLMs) are increasingly used as automated judges to evaluate recommendation systems, search engines, and other subjective tasks, where relying on human evaluators can be costly, time-consuming, unscalable. LLMs offer an efficient solution for continuous, evaluation. However, since the systems that built improved with these judgments ultimately designed use, it is crucial LLM align closely ensure such remain human-centered. On hand, aligning challenging due individual...
The increasing number of service providers on the Web makes it challenging to select a provider for specific demand. Each consumer has different expectations given in contexts, so selection process should be consumer‐oriented and context‐dependent. Current approaches typically have consumers receive ratings from other consumers, where reflect consumers' overall subjective opinions. This may misleading if contexts satisfaction criteria. In this paper, we propose that objectively record their...
Selecting the right parties to interact with is a fundamental problem in open and dynamic environments. The amplified when number of interacting high, parties' reasons for selecting others vary. We examine service selection an e-commerce setting where consumer agents cooperate identify providers that would satisfy their needs most. Previous approaches are usually based on capturing exchanging ratings consumers providers. Rating-based have two major weaknesses. 1) given particular context....
As the semantic web expands, ontological data becomes distributed over a large network of sources on Web. Consequently, evaluating queries that aim to tap into this database necessitates ability consult multiple efficiently. In paper, we propose methods and heuristics efficiently query based series properties summarized data. our approach, each source summarizes its as another RDF graph, relevant section these summaries are merged analyzed at evaluation time. We show how analysis enables...
In information driven multi-agent systems, consumers collect about their environment from various sources such as sensors. Each source has its own limitations, capabilities, and goals. Therefore, there is no guarantee that a will provide the requested truthfully correctly. Even if provided only by trustworthy sources, it can contain conflicts hamper usability. this paper, we propose to exploit revise trust in information. This requires reasoning mechanism accommodate domain constraints,...
Stereotypical trust modeling can be adopted by a buyer to effectively evaluate trustworthiness of seller who has little or no past experience in e-marketplaces. The forms stereotypes based on her with other sellers. However, when the limited sellers, formed cannot accurately reflect evaluation towards To address this issue, we propose novel generalized stereotypical model. Specifically, first build semantic ontology represent hierarchical relationships among attribute values. We then fuzzy...
In this paper, we propose risk-calibrated evidential deep classifiers to reduce the costs associated with classification errors. We use two main approaches. The first is develop methods quantify uncertainty of a classifier's predictions and likelihood acting on erroneous predictions. second novel way train classifier such that classifications are biased towards less risky categories. combine these approaches in principled way. While doing this, extend learning pignistic probabilities, which...
When managing intelligence, surveillance, and reconnaissance (ISR) operations in a coalition context, assigning available sensing assets to mission tasks can be challenging. The authors' approach ISR asset assignment uses ontologies, allocation algorithms, service-oriented architecture.
The emergence of large scale, distributed, sensor-enabled, machine-to-machine pervasive applications necessitates engaging with providers information on demand to collect the information, varying quality levels, be used infer about state world and decide actions in response. In these highly fluid operational environments, involving consumers various degrees trust intentions, obfuscation is protect from misuses they share, while still providing benefits their consumers. this paper, we develop...
Bootstrapping trust assessment where there is little or no evidence regarding a subject significant challenge for existing and reputation systems. When direct indirect absent, approaches usually assume that all agents are equally trustworthy. This naive assumption makes vulnerable to attacks such as Sybil whitewashing . Inspired by real‐life scenarios, we argue malicious may share some common patterns complex features in their descriptions. If can be detected, they exploited bootstrap...
Selecting the right parties to interact with is a fundamental problem in open and dynamic environments. The exemplified when number of interacting high parties' reasons for selecting others vary. We examine service selection an e-commerce setting where consumer agents cooperate identify providers that would satisfy their needs most. Previous approaches are based on capturing exchanging ratings consumers providers. Contrary previous, rating-based selection, this paper advocates objective...
This paper argues the need for research to realize uncertainty-aware artificial intelligence and machine learning (AI\&ML) systems decision support by describing a number of motivating scenarios. Furthermore, defines uncertainty-awareness lays out challenges along with surveying some promising directions. A theoretical demonstration illustrates how two emerging ML AI technologies could be integrated value route planning operation.
When collaborating with an artificial intelligence (AI) system, we need to assess when trust its recommendations. Suppose mistakenly it in regions where is likely err. In that case, catastrophic failures may occur, hence the for Bayesian approaches reasoning and learning determine confidence (or epistemic uncertainty) probabilities of queried outcome. Pure methods, however, suffer from high computational costs. To overcome them, revert efficient effective approximations. this paper, focus on...