WP5 sessions in Stuttgart meeting, 13-15 October 2008
WP5 plenary
20081013_WP5_Plenary_FINAL_PRESENTED.ppt
WP5+6+7a+7b session - Some notes
Review of wp7b storyboard (D7b1.1a.pdf), comments from WP7a and WP6.
AP: Georgina to organize a telco with WP6 and WP7a and WP7b (dedicates telcos) to go through the stories.
WP7b
1. Offline process. It can be realized inside larck, it will likely have different parameters as the online ones.
· Max time parameters will not be so relevant, not tight impositions in this sense.
· Desired n° results: as much as we can get
· Type of data set to identify: among them, text is required.
A requirement from the Urban computing mainly would be to consider the lifetime of data: some data may be outdated after some time. This must be part of the query: e.g. i want to identify only the data generated after 15:00h. This will be translated, when calling the identify plug-in in the necessary query/constraints according to the input required for that plug-in.
WP6 use case has strict requirements on real-time.
WP7a and b also require real-time response in the interaction with the user steps.
Saltlux: We need a way to specify the quality of the answer the user needs. This only can be done though measurable parameters (E.g. max time of response, min n° answers,...) or a combination of various measurable parameters.
It is agreed that we need an interface for the “configuration designer” that allows him to construct the pipeline and setup parameters individually for every plugin. Also this interface must allow to either: specify concrete plugins i want to run or specify type of plugins (and the D should find the most appropriate ones among the available ones of that type)
Review urban comp. use case. They have some extra req. such as unique name assumption
1. Wp7b. 1.
a) Identify
b) Transform (from one type to another, it can be specify which types, can be different)
c) Transform (using reasoning with rules provided by the user)
d) ok
e) Annotate texts using LLD (retrieval of new knowledge): Open issue for WP5. For the moment we leave it out of the scope of wp5 prototype, but we must think about it.
2. WP7b – Section 5.2.2. Point 1 is ok (same as previous). Point 2:
a) Query transform. How queries are distributed? This will be a perfect application for the thinking@home approach. We can consider different cases: e.g. 500 queries against 1 data set; e.g. 2 same query against different data sets.
Pending issue for WP5 (from use cases): storing data back.
WP6: several repositories with different kind of data, almost all related to time tables.
WP6 not implementing reasoner to find optimal path. Where shall we take it from? WP5 is already wrapping some existing plugins. We can do the same for other required external plugins.
WP5+2+3+4 - Some notes
Terminology: Rename of plugins agreed.
WP2: what was retrieval, now it is in some cases identify and in others transform
See presentation from Hamish
- Overlap with transform stream in wp3?
- Parallel tasks transform and select?
- Selection close to the source (in case of streaming)
- Input must be always rdf?
- Fixed pipeline integrated in 1 plugin. How to manage it?
- Granularity of data, how to split (or has it sense to split it) the streaming in terms of performance?
- Context: static info about plugins, dynamic info about the plugins, info about user. Stateful vs stateless.
o Pass it as a parameter between plugins
o Keep it in the data layer
o In KB of the platform
- Performance: wp2 have some ideas on how to measure, qualitative and quantitative
- Computational resources,... . Platform keep details. Logging library. Training period to stimate first qos performance parameters (Performance indicators)
WP4
See ppt from Zhisheng
- Library naming according to the plugin naming (in the repository)
- Memory mgmt loading? Who is deciding when to dereference the pointers to data and pass the data itself. Where to store data when it is a large amount of it? Related with storing data back we were discussing in use cases joint session.
- Indexing and hashing
- Similarity and relevance measurement and selection
QoS parameters:
Stefano (CEFRIEL): Consider the performance parameter different if it is data driven (all queries over the complete data) or same query over different pieces of data...
For the select component (and others?) Consider a parameter called: Time-out (or interruptable? -> better time-out)
AP: WP5 to group parameters in qos, functional, contract
CEFrIEL: how to certify that the qos parameters are kept? We are currently in wp5 analysing solutions such as slas
LUNCH MEETING: Frank, Cyc, UIBK
- Cyc: remove proprietary code,
- work on thread-safety, migrate current DECIDEr to V0.2 API
- Innsbruck:
- - complete current API - integrate ORDI as data-layer API - test previous item by moving to another demo-scenario,
- using a larger dataset (to be provided by Amsterdam)
- USTUTT:
- - set up issue tracking system, and fill with current issues (from earlier WP5 meetings and from the Stuttgart consultation sessions)
later:
- write a number of example "stub"-plugins that enable remote invocation
- (as web-service, as HTTP, other?). This is probably our quickest-win scenario to get remote execution going before the M12-M14 release
- design declarative representation for pipeline configurations