Using active Learning for Selection
Participants:
- Danica (USFD)
- Yi (Siemens)
Agenda:
- joint experiment to test whether subsetting based on IR and ML can reduce the problem space for reasoning
to agree on the design of the experiment in: https://docs.google.com/document/d/1IYKUbO95VivimsaVTUjgtIh0Lag4i9HZ_0SC8qNkDXE/edit?hl=en_GB&authkey=CPnY--IP
Minutes:
- design of the experiment ok
- Actions:
- danica: send virtual molecules and example of subset to Yi (done)
- danica to ask jose about evaluation software as it would be nice not to implement that but use theirs (done)
- yi to figure out whether virtual documents can be used for datapoints or any other way how active learning can be used for selection
- for now we both do proof of concept with geography ontology so that we better understand whether the whole experiment makes sense
- in two weeks we both need to have this small experiment finished; how do we know if we have finished it?
- given the two sample sparql queries we need to generate relevant subsets (set of statements), load them into a repository (Virtuoso, Sesame, or any other) and get some results (we might also get 0 results depending on how well our method works for the sample queries)
- email communication for problems, and then organize a telco in 2 weeks to check the status
- maybe: after we finish proof of concept think about inviting others (Ivan for help with owlim or Amsterdam people for the reasoning part? involve MPG as well if they are interested)