[Ontobras 2015] Keynote: Large-scale Semantic Web Reasoning - by Grigoris Antoniou

Large-scale Semantic Web Reasoning - by Grigoris Antoniou

1^st Keynote Ontobras

Intro: Big Data is commonly associated with Data Mining and Machine learning.

·      Uncover hidden patterns, thus new insights

·      Mostly statistical approaches

He claims semantics and reasonings are also relevant:

1. Semantic interoperability

2. Decision making

3. Data cleaning

4. Inferring high-level knowledge from data

1. Why do we need Semantic Interoperability?

• ·      To create added value through combination of different, independently maintained sources. E.g. Health (combine healthcare, social and economic data to better predict problems and to derive interventions)
• ·      Combine historic and actual data

Problems: complexity and dynamicity

2. Make sense of the huge amounts of data:

• -       turn it into action
• -       be able to explain decisions – transparency and increased confidence
• -       be able to deal with imperfect, missing or conflicting data
• -       all in the remit of Knowledge Representation (KR)
• -       E.g. alert of a possible dangerous situation for an elderly person when certain conditions are met.

Potential Domains: smart cities, intelligent environments, ambient assisted living, intelligent healthcare (including remote monitoring), disaster management

But can we deliver?

The problem:

• -       traditional approaches work in centralized memory
• -       But we cannot load big data (or the Web) on a centralized memory, nor are we expected to do so in the future.
• To the rescue: New computational paradigms
• -       Developed in the past decade as part of high-performance computing, cloud computing etc.
• -       Developed independently of the Semantic Web (SW) and KR, but we can use them.

What Follows:

•  Basic RDFS reasoning on Map Reduce
•  Computationally simple nonmonotonic reasoning on Map Reduce
•  Computationally complex ontology repair approach using Signal/Collect (can we apply these approaches to exponentially reasoning tasks?)

Problems:

• Load Balancing 
• High I/O Cost 
• Program Complexity

MapReduce:

• Introduced by Google in 2004
• Computation is expressed only with Maps and Reduce (popular implementation: Hadoop)

No restante da palestra, ele propos soluções utilizando MapReduce.

Still missing / Future Work:
- What pararelization architecture are more appropriate?
- There are no agreed benchmarks for large-scale reasoning
- More complex types of reasoning should be considered, such as spationtemporal reasoning and exponential reasoning.
- Extreme reasoning: make reasoning with big data work in real time (there are new systems, like Apache Storm)