ALOOF: Autonomous Learning of the Meaning of Objects
CREOL @ JOWO 2017
Valerio Basile
23/9/2017
CHIST-ERA (EU) project 2014-2017
- Sapienza University: Computer Vision
- TU Wien: Perception
- University of Birmingham: Robotics
- Inria Sophia Antipolis: Knowledge Management
Inria@ALOOF
- Natural Language Processing
- Knowledge Representation
- Semantic Web
Lifelong Learning Loop
- Robot deploy
- Perception (segmentation, object recognition, scene detection, ...)
- Knowledge processing
- Human validation
- Robot deploy
- ...
J. Young, L. Kunze, V. Basile, E. Cabrio, N. Hawes, B. Caputo
Semantic Web-Mining and Deep Vision for Lifelong Object DiscoveryICRA 2017
DeKO: Default Knowledge about Objects
A large-scale, RDF knowledge base of prototypical knowledge about objects.
- Large-scale: learn from data
- RDF: don't reinvent the linked data wheel
- Prototypical: beliefs, not facts
- Objects: know your boundaries
Default knowledge
- Classification
- Prototypical locations
- Prototypical actions
- Frame instances
T-box
- DBpedia
- WordNet/BabelNet
- FrameBase/Framester
- DeKO Ontology*
Knowledge sources
- Crowdsourcing
- Manual annotation
- Relation Extraction (EKAW 2016, SWJ -- under revision)
- Keyword linking (AAAI 2017 -- under revision)
- Frame instance parsing and clustering (AnSWeR 2017)
Output
DeKO is coming soon... In the meantine:
- KNEWS (Knowledge Extraction with Semantics)
https://github.com/valeriobasile/learningbyreading
- Data sets
https://project.inria.fr/aloof/data/
- Open position
* Representing default knowledge is an open problem