LarKC WP4 Teleconference, 2.00pm - 3.00pm, Jan 19, 2011
The skype ID of the chair: zhisheng_huang
Participants
Danica(Sheffield), Jose(MPG), Hans（MPG), Yi(WICI), Ivan（Ontotext), Mihai(IRF), Gaston（VUA), Zhisheng(VUA), Jun(VUA/NPU), Spyros(VUA), Reto (Innsbruck), Matthias(HLRS).
Objective
This is a pre-meeting for the joint WP2+WP4 workshop at the LarKC Lyon Meeting (Wednesday, 02-Feb-11, 14:00-17:00, Room Princess Takamatsu)
Agenda
- Tighter Integration of Selection and Reasoning: Discussion on the comments from the LarKC 2nd Year review
- Agenda of the joint workshop
Comments from the LarKC 2nd year review
- We note that some of the specific claims of the initial plan have not been addressed in great depth, e.g. using information retrieval as a selection step before symbolic reasoning, and implementing anytime behaviors.
- More integration of IR and reasoning
- the connection between reasoning and heuristics seems to have almost entirely disappeared from the project. Maybe there is a good reason for this but it is not one I understand at the moment. I would appreciate a short note of comment on this (an e-mail message is perfectly adequate).
- the connection between reasoning and economics (anytime query answering) also seems to have disappeared. In this case as well I would like to ask you to please carry out some actual tests in the last reporting period of the project to show what difference, if any, the application anytime answering methods make in the answering of queries, even queries that could comfortably be answered exhaustively on your current largest RDF stores.
- Application of anytime behaviours and stopping rules.
- Reasoning that cannot be reduced to query answering applied to large data. The intent is to show that tasks that are impossible or prohibitively complex in SQL are useful and feasible in LarKC. This might take the form of a comparison. Danica: the document made after the last WP2 teleconference:
https://docs.google.com/document/d/1tCaT4RVgWo8Ctj6f0NUmom2AsXGkaLmGeA9SijZ7OTs/edit?hl=en_GB&authkey=CMDRxZ0N# <https://docs.google.com/document/d/1tCaT4RVgWo8Ctj6f0NUmom2AsXGkaLmGeA9SijZ7OTs/edit?hl=en_GB&authkey=CMDRxZ0N#>
Reto: rule-based reasoning for the interleaving processing. Using IRIS could give some examples as said. IRIS is still rather limited in the number of facts that it can take into account. hence, it pays of to limit the size of the factbase.
Gaston: we are going to have another teleconference with WP6. We are going to see whether or not the interleaving scenario also occurs in the WP6 use case on urban computing and reasoning with noisy data.
Zhisheng: We have to design a heuristic langauge (which is based on RDF/OWL) for LarKC to specify heuristics to guide selection, reasoning, and other processing in LarKC. The heuristic information will be dynamic and be changed from time to time. The reasoner can update the heuristic information, so that the selector can improve the quality of selection, based on the heuristic knowledge/information.
Danica: Heuristic information can be context-dependent or context-independent. Maybe involved user-interests, like WICI's work.
Hans: We should make them simple and easy to be used at the beginning.
Zhisheng: Another issue: how to design a general evaluation framework, so that the reasoners can make the evaluation on the answers of reasoning and give a feedback to the selector, so that the selector can make smarter selection next time at the interleaving circle.
Workshop Agenda
Topics:
1. Scenarios: find some interleaving scenarios from the LarKC use cases (data, query, and workflow).
2. Heuristics: heuristic data format (query, preferred data sources, preference, uncertainty?): Gaston(?)
3. Evaluation: how to evaluate the answers automatically or semi-automatically? human evaluator? gold standards?
4. Dynamics of heuristics: what, how, and when to update heuristic knowledge/information?
5. Smart selection with heuristics: how to make a smart selection with heuristics?
6. Reasoner feedback on heuristics: how to give a feedback to the selector? or how to update heuristic information?
Each one is encouraged to pick up one or more issues above, and have some brain storming on it, and present some ideas in the workshop.