search

ML-Schema / core

📚 CORE ontology of ML-Schema and mapping to other machine learning vocabularies and ontologies (DMOP, Exposé, OntoDM, and MEX)
http://purl.org/mls
26 stars 7 forks source link

Vocabularies x Ontologies #4

Open diegoesteves opened 8 years ago

diegoesteves commented 8 years ago

How broad should it be? What is the aim? ML or DM descriptions? Interoperability (engineering) or Meta-learning strategies (science)? As as much as we make it more complex, more and more people will avoid the usage. I suggest a vocabulary for engineering purposes (tools integrations e description of ML executions) and an ontology for more complex analysis (pre processing, statistical methods and related DM tasks). We should remember we have to provide ways for people understanding and using the methodologies. People from mathematics, statistics, electrical engineering, for instance, are not (commonly) aware of linked data methods and tools. Besides, what’s the sense of describing an ontology for ML if, in the end, we use it for schema representation only, instead of benefit meta-learning approaches, for instance? The benefit of providing this kind of metadata is not clear for everybody in the short term and is pretty hard to convince people to represent their data using a schema. "what is the current volume of (real) metadata we have?" and "what kind of information has been discovered so far?" are examples of questions we should highlight.

larisa-soldatova commented 8 years ago

Diego,

I completely agree with you. We should target: a simple vocabulary (accessible to everyone), + a suite of ontololgies for specific use cases, based on the vocabulary + really good guidelines on the usage of the vocabulary, extensions + good examples.

Larisa

On Tue, Oct 27, 2015 at 8:24 PM, Diego Esteves notifications@github.com wrote:

How broad should it be? What is the aim? ML or DM descriptions? Interoperability (engineering) or Meta-learning strategies (science)? As as much as we make it more complex, more and more people will avoid the usage. I suggest a vocabulary for engineering purposes (tools integrations e description of ML executions) and an ontology for more complex analysis (pre processing, statistical methods and related DM tasks). We should remember we have to provide ways for people understanding and using the methodologies. People from mathematics, statistics, electrical engineering, for instance, are not (commonly) aware of linked data methods and tools. Besides, what’s the sense of describing an ontology for ML if, in the end, we use it for schema representation only, instead of benefit meta-learning approaches, for instance? The benefit of providing this kind of metadata is not clear for everybody in the short term and is pretty hard to convince people to represent their data using a schema. "what is the current volume of (real) metadata we have?" and "what kind of information has been discovered so far?" are examples of questions we should highlight.

— Reply to this email directly or view it on GitHub https://github.com/ML-Schema/core/issues/4.

joaquinvanschoren commented 8 years ago

Agreed. What I feel we sort of agree on is that we need a commonly agreed set of concepts and definitions, as a core vocabulary. Then, existing and new ontologies could adapt to that core vocabulary and add interesting information. Agreeing on this core vocabulary is already a nice step forward.

I do believe some basic structure is useful (e.g. a classification algorithm reads data and produces models), and that can be expressed in a schema (e.g. in TTL syntax). I would not be too afraid of including such basic structure since it gives meaning to the concepts.

diegoesteves commented 8 years ago

Great! This issue is dependent on the issue #2 and the generated matrix of correlation based on existing vocabularies and ontologies.