<koalie> koalie has changed the topic to: https://w3c.github.io/tpac-breakouts/sessions.html
Dave shows some slides.
dsr: First part is non-technical, about the motivation.
… 2nd part is about making RDF easier, and about machine learning
… Digital Transformation is a hot topic, at least in Europe. This work is supported by a large EU project, called Boost.
… DT is widely applicable.
… Modern car production is an example where it is already applied. E.g., individually adapted products, rather than mass production
… Requires digital integration, from office floor to shop floor to manufacturing, via board room.
… Also integration in time, keeping history of a product, use and lifecycle.
… Hurdles: many companies still use paper forms, knowledge is in people's heads, (paper) documents are outdated...
joemedley: Today's digital processes are often just the old paper processes done digitally.
dsr: Diff. orgs. have different histories and different ways of talking about things.
… Need an incremental approach, bring people along, capture knowledge.
… DAMA Intl. developed a "data governance wheel" with things to consider. Data governance and data management not necessarily clearly differentiated.
… Graphs are a flexible way to represent knowledge. More flexible than a SQL database.
… Knowledge Graphs (aka ontologies).
… There are many graph database vendors and the number is growing.
… Sentient Web is a watu to bring things together: sensors + graphs + rules for reasoning (AI, Machine Learning)
… No large org. has a single database, it is distributed.
Phil: True, but you'd be surprised how many are trying to make a "single source of truth" (without success).
dsr: RDF foundation for data.
… W3C Graph Data workshop last March.
… Writing white paper.
… Expecting standardization work at W3C.
… W3C Business Group.
dezell: A list of companies interested in the Bus. Group?
dsr: We have been slow in talking to companies after the workshop. But let me know if you're interested.
Phil: Interesting topic, but not sure I'll be able to join.
dsr: Flip side of digital transformation is need for good security. But many orgs. still working with old computers.
… UK's NHS, e.g., very much based on paper processes.
Phil: NHSX is trying to help them...
dsr: At least there is a common medical vocabulary.
dsr: Discussion topics:
… We can share use cases, collect ideas.
… 2nd part is technical details.
… Need ways to derive ontologies automatically, rather than by hand.
… "Easier RDF" project by David Booth.
… I did some hacking in August (during what W3C team calls "Geek Week").
… Simpler query languges.
gkellogg: Did you look at GraphQL?
dsr: Don't think I did, but the list on my slide is not meant to be complete.
Dave shows different queries (using Turtle syntax).
Implementation runs in a browser.
It fetches data and applies a query. Allows to step through the execution.
dsr: Expressing both the rules and the data in RDF is meant to make machine learning easier.
Phil: Is this really easier than Sparql?
dsr: Don't know yet, needs experimenting.
… Sparql, like SQL, looks for patterns.
… The RDF approach allows building a state machine.
… Shacl could, with this approach, become a machine that does validation.
… RDF/Turtle not ideal. There may be easier ways.
… Looked at research in cognitive science, brain research.
… Psychology uses the word "chunk" for a set of properties.
… A link between chunks is also a (simple) chunk.
… But have to deal with noise, imprecise data. Need to combine symbolic and statistical approaches.
… Data is accessed via a small "buffer". Looks limiting, but allows scaling.
Dave shows examples. One example is a model of how people count. You need rules that say which number follows which other number, it is not not like a calculator.
The demo shows how the rule that says 4 follows 3 is found and applied, then the next, etc.
Another demo shows a (simulated) car following a road.
dsr: Autonomous driving has rules that are spatial and temporal.
… There is a context as well. E.g., a book can describe a world that is like the real world, except for some rules.
… Imitiating human memory recall, finding probably useful information, dealing with noise.
… We can borrow from cognitive science, which has studied this since some time.
… Learning and problem solving typically starts quite wild and gets more focused with experience.
… Use inspiration from brain, conscious and unconscious functions.
… Cognitive Agent could get info from distributed sources.
… Next steps? Collect use cases, data sets. Also figure out the economics, cost of collecting a data set.
Phil: Research organizations working together could do it.
… Sounds very advanced.
dsr: Using the insights from research in other areas simplifies the problem.
s/Ditial/Digital
Succeeded: s/Required/Requires
Succeeded: s/make/to make
Succeeded: s/My org/NHSX/
Succeeded: s/not/is not/
Succeeded: s/Ditial/Digital/
Failed: s/Ditial/Digital
Succeeded: s/: rrsagent, make minutes//
Succeeded: s/th elist/the list
Maybe present: dsr, gkellogg, Phil