Attendees

• wai-project@helsinki.fi

Goal of the meeting

• Introduce ideas for implementing the concepts outlined in the kick-off of the last week.

Discussion items

In meeting

1. Juho Piironen (for Antti) presented the first potential application of WAI: Model selection for linear models.
   The presented example consisted in trying to model some outcome about diabetes as a linear combination of several explanatory variable (age, gender, bmi, ...).
   Two potential main steps necessitating WAI are the variable selection process and the posterior predicting checking.
   There is also a further possibility: suitably visualising what's the best (most significant) reduction of the parametric space for explaining y.
   • interactive variable selection tools and example: https://github.com/paasim/shinyproj
     
     
2. Tomi presented few ideas on using WAI as elicitation of non-linearity.
   He suggested how working in particular with Gaussian Processes can open several opportunities for making successfully interacting the user with the AI.
   
   
3. Pedram introduced an application on Task-Intent Modelling. 
   A nice example is a user searching for some information (e.g. when googling). 
   The user has an idea of what he/she wants to find and in his/her mind has a list of items ordered by relevance (i.e. the first items are in his/her opinion the ones will provide easier the document he/she has been searching for). There are at least couple of situations requiring WAI: first, throughout the research process the user can learn about the topic he/she is searching for. This means that the order by relevance can evolve as more information joins the research. As second, we can argue that the list of items ordered by relevance that the user has in mind strongly depends on how clear is to the user what he/she is searching for. In the user has an understanding of what he/she wants but nothing more specific, not only the considered items could be related to uninformative related relevance but relevant items could be left out of the analysis (because the user is not aware of their existence).
   • Ulpu mentioned a rational information search model proposed in: https://helda.helsinki.fi/handle/10138/167284 (section 5)
     
     
4. Umberto presented the basic idea and concepts behind the Radial Velocity method, which is an indirect technique (not only) used for detecting Extrasolar Planets. This technique relies on both physical and geometrical conditions and one of its downsize is the impact that stellar activity has on resulting stellar spectrum. Each stellar spectrum consists in a collection of hundred of thousands of wavelength-
   WAI could be used for disentangling the impact of active regions (i.e. spots and faculae) from the pure Doppler shift caused by an Exoplanet is key for detecting ``Earth-like´´ planets. 
   Experts are needed for providing feedback on the stellar spectrum (quality of the absorption lines) and interpretation of the shape of the absorption lines.

Action items for the next meeting

Next Meeting in A346 Otaniemi + skype on February 14th 9:45 am - 11 am

• Keep presenting potential scenario in which WAI is needed
  
  • Umberto: updates on idea for applying WAI for detecting exoplanets using the radial velocity technique
  • Tomi: updates on his idea for applying WAI for elicitation of non-linearity
  • Pedram: updates on his idea for applying WAI for intent modelling