diff --git a/_data/keywords.yaml b/_data/keywords.yaml index 6801a6b..5c24c53 100644 --- a/_data/keywords.yaml +++ b/_data/keywords.yaml @@ -1,21 +1,35 @@ +- keyword: Logics + count: 6 +- keyword: Ontologies + count: 6 - keyword: Knowledge Representation count: 4 - keyword: Knowledge-based Robotics count: 4 +- keyword: Reasoning + count: 4 - keyword: Computational argumentation count: 3 +- keyword: Description Logics + count: 3 - keyword: Hybrid Intelligence count: 3 - keyword: BERT models count: 2 +- keyword: Belief functions + count: 2 - keyword: Computer Vision count: 2 - keyword: Domain Coverage count: 2 +- keyword: Explainability + count: 2 - keyword: Human-in-the-Loop count: 2 - keyword: Implementation count: 2 +- keyword: Information fusion + count: 2 - keyword: Knowledge Graphs count: 2 - keyword: Large Language Models @@ -28,6 +42,8 @@ count: 2 - keyword: Reproducibility count: 2 +- keyword: Uncertainty + count: 2 - keyword: Argument Mining count: 1 - keyword: Deep Learning/Language Model @@ -44,12 +60,16 @@ count: 1 - keyword: Interaction count: 1 +- keyword: Knowledge Compilation + count: 1 - keyword: Knowledge Engineering count: 1 - keyword: LLMs count: 1 - keyword: Language Model count: 1 +- keyword: Learning + count: 1 - keyword: Nature Langue Processing count: 1 - keyword: Ontology Engineering diff --git a/_data/topics.yaml b/_data/topics.yaml index 636687e..339cbd6 100644 --- a/_data/topics.yaml +++ b/_data/topics.yaml @@ -3,6 +3,8 @@ description: 'Projects focus on how to combine KR with Large language Models, including BERT models and generative LLMs like LLaMA and GPT series.' - name: Bias detection in LLMs description: 'These projects focus on how to detect bias in LLMs to ensure fairness in society.' +- name: Uncertainty representation + description: 'TBD -- would be something for Daira to think about' # topics from 2023-2024 - name: Knowledge Extraction @@ -20,7 +22,7 @@ - name: Knowledge Graphs and Deep Learning description: 'These projects study how to deep learning models can capture different types of semantics represented in the data.' - name: Formal Logics, Modal Logics - description: 'Projects looking into the application of differe kind of logics in AI (multi-agent) systems.' + description: 'Projects looking into the application of different kinds of logics in AI (multi-agent) systems.' - name: Ontologies and Reasoning description: 'These projects focus on reasoning over OWL or description logic-based ontologies.' diff --git a/_theses_dir/DPP_KnowComp.md b/_theses_dir/DPP_KnowComp.md new file mode 100644 index 0000000..d8647f9 --- /dev/null +++ b/_theses_dir/DPP_KnowComp.md @@ -0,0 +1,36 @@ +--- +title: 'Knowledge Compilation for Forgetting Uncertain Evidence' +layout: default +description: 'In this project, students will become familiar with some knowledge compilation techniques and rules for combining uncertain evidence. In particular, they will explore the advantages and disadvantages of using knowledge compilation for forgetting uncertain evidence.' +topic: 'Formal Logics, Modal Logics' +keywords: + - 'Uncertainty' + - 'Knowledge Compilation' + - 'Information fusion' + - 'Belief functions' +supervisor: 'Daira Pinto Prieto' +contact: 'd.pintoprieto@uva.nl' +degree: 'M.Sc.' +description_link: '/theses_dir/DPP_KnowComp' +--- + +## {{page.title}} +*Supervisor: {{page.supervisor}} ({{page.contact}})* + +#### Background +Knowledge compilation is a collection of computational approaches that allows to break down some (computationally) hard problems into an offline and an online phase. If the online part can be computed in polynomial time, the problem is said to be compilable to P. In belief function theory there are some rules of combination of evidence whose computation is compilable to P. Therefore, we can think of real-world scenarios where uncertain evidence can be combined and decombined using these rules, overcoming the challenge of their computational complexity. + + +#### Potential projects: +- Following existing literature, implement a solution to compute the unnormalized and normalised rules of combination based on knowledge compilation. Study the advantages and disadvantages of using this solution compared to using other algorithms to decombine uncertain evidence. . +- Extend the literature on the computation of combination rules through knowledge compilation. This may include adapting propositional formulas to accept more general evidence, or defining new propositional formulas that can be used to implement other rules of combination via weighted model counting. + + +#### Literature +- Pierre Marquis. 2015. Compile! In Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI’15). AAAI Press, 4112-4118. + +- Daira Pinto Prieto. Combining Uncertain Evidence: Logic and Complexity. Chapter 6. PhD thesis, University of Amsterdam, 2024. ISBN 978-94-6473-618-2. +- Daira Pinto Prieto, Ronald de Haan, and Sébastien Destercke. 2024. How to efficiently decombine belief functions? In Proceedings of the 20th International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems (IPMU 2024). + + + diff --git a/_theses_dir/DPP_UncerEv.md b/_theses_dir/DPP_UncerEv.md new file mode 100644 index 0000000..96dac4d --- /dev/null +++ b/_theses_dir/DPP_UncerEv.md @@ -0,0 +1,30 @@ +--- +title: 'Uncertain Evidence in Artificial Intelligence' +layout: default +description: 'This project aims to address the problem of dealing with uncertain information in the context of artificial intelligence. Students can approach this question from a conceptual and/or experimental perspective.' +topic: 'Formal Logics, Modal Logics' +keywords: + - 'Uncertainty' + - 'Belief functions' + - 'Information fusion' +supervisor: 'Daira Pinto Prieto' +contact: 'd.pintoprieto@uva.nl' +degree: 'B.Sc.' +description_link: '/theses_dir/DPP_UncerEv' +--- + +## {{page.title}} +*Supervisor: {{page.supervisor}} ({{page.contact}})* + +#### Background +*Give some background information* + +Developing methods for aggregating uncertain information is a thriving area of research. Some of the questions we can investigate on this topic are: What are we uncertain about? How does uncertainty interact with other properties of information (such as consistency or relevance)? How do different aggregation methods behave in practice? + +#### Potential projects: +- Compare different evidence combination rules on a relevant dataset focusing on those cases with mutually contradictory evidence. +- A literature review about epistemic and aleatory uncertainty in artificial intelligence. This review could be complemented with illustrative examples built on real-life data and a (non-extensive) review on computational methods to deal with these kinds of uncertainty. + + +#### Literature +- Chaki, J. (2023). Handling uncertainty in artificial intelligence (1st ed.). Springer Singapore. ISBN 978-981-99-5333-2.