W3C

DPVCG Meeting Call

10 APR 2025

Attendees

Present
beatrizEsteves, delaramGolpayegani, julioHernandez, markLizar, mayraRusso, paulRyan
Regrets
georgKrog, harshPandit
Chair
delaramGolpayegani
Scribe
beatrizEsteves, delaramGolpayegani

Meeting minutes

Repository: w3c/dpv

Agenda: https://www.w3.org/events/meetings/178d1c71-a92d-4da7-a196-6a89d0fe2277/202500410T133000/

Meeting minutes: https://w3id.org/dpv/meetings

Persistent ID for current minutes: https://w3id.org/dpv/meetings/meeting-2025-04-10

DPV v2.1.1 hotfixes

<ghurlbot> Issue 279 [FIX]: DPV v2.1.1 Hotfixes (by coolharsh55)

group: Ok to be published

Subjective Locations

<ghurlbot> Issue 209 [Concept]: `AtX` subjective Location concepts (by coolharsh55)

if the framework (the abstract concepts) are okay, to add them, and we will next discuss the specific taxonomy and how it should be associated with personal data e.g. home is location but that location is also personal data

group: Fine to integrate; later we should discuss how to connect with existing concepts re. personal data

Non-X Inverted Locations

<ghurlbot> Issue 208 [Concept]: Add Non-X (X = specific location) to represent locations that are not X (by coolharsh55)

note from harsh: The algorithm for this is: 1) if it is a country, then select all countries that are not this country and add skos:narrower/broader to the concept; 2) if it is a supranationalunion (EU) then select all countries that are not members and add skos:narrower/broader. If this is fine, I will implement it in RDF and next week we look at the completed HTML.

group: Okay to implement

AI Training

<ghurlbot> Issue 82 Provide vocabulary to specify purposes and permissions related to AI training (by scottkellum)

note from harsh: I am not satisfied with these. My issues are 1) how to link training to data processing in DPV; 2) how to separate techniques that may be used in training but are also applicable outside of training; and 3) how to link training to generation of the model.

note from harsh: The current thought process I have goes like this: ai:Train extends dpv:Use to state training is a type of processing (it always includes data, and train can refer to model or system). Then do not have any further distinction between training techniques and AI techniques at this point -- we need an "AI expert" to go over this list and classify them. I find there is a lot of confusion and misinterpretation of "using technique" to actually mean "using model trained using technique". So we don't do this classification at this point -- our technique taxonomy is decent as it is. To state how training was done, a process can state ai:hasTechnique xyz (this was Mayra's point I think). To state data is being used in training dpv:hasProcessing ai:Train is nice and simple.

group: Have ai:Train extending dpv:Use sounds good, we should also consider other processing operation like (model) validating and so on.

Comment from harsh after meeting: validation is a broader process that includes things other than operation on data, therefore it should not be of type dpv:Processing. Training is about manipulation of data specifically, so by definition it fits the processing concept.

group: Should we separate AI and training techniques? At this point probably it is not needed since we don’t have enough expertise yet to make the distinction.

DocGen improvement

<ghurlbot> Issue 262 Refactor documentation generator to be modular (by coolharsh55)

note from harsh: at the moment, we generate a lot of files (XML, JSON-LD, etc.) -- if the group agrees, we will only generate the minimum amount of files that don't show up as changes in both RDF and HTML, and only generate the other files when we are preparing a release (if Beatriz or Art are there, they can explain this)

group: Ok to only generate the minimum needed

AI Act Risk Level

<ghurlbot> Issue 231 Represent Risk Level classifications for AI/Systems in AI Act (by coolharsh55)

group: Proposed taxonomy seems ok to add to AI Act extension; The AI Act terms should be preferred. there was also agreement on representing specific instances has processes

Datasheets and ModelCards

<ghurlbot> Issue 94 Represent Datasheets and Model Cards with DPV (by coolharsh55)

delaramGolpayegani: Delaram: FYI, AI Cards can be already implemented DPV - if agreed to be add to DPV, we need guiding material on how to generate it using DPV.

No updates

<ghurlbot> Issue 278 [NEW]: Misc concepts based on GDPR CoC and Certifications (by coolharsh55)

<ghurlbot> Issue 238 Update EHDS extension with practical concepts (by coolharsh55)

topics were discussed, there were no updates

Next Meeting

The next meeting will be on APR-17 Thursday 13:30WET/14:30CET

Agenda will be continuing on topics started under v2.2, and any other updates that come up.

Minutes manually created (not a transcript), formatted by scribe.perl version 217 (Fri Apr 7 17:23:01 2023 UTC).