Conception d'expériences pour connaître les préférences visuelles personnalisées des occupants de bureaux privés.

Design of experiments for learning personalized visual preferences of occupants in private offices.

Numéro : pap. 3716

Auteurs : AWALGAONKAR N. M., XIONG J., BILIONIS I., et al.

Résumé

This paper presents an online data-driven methodology which actively queries a new occupant for learning their personalized visual preferences. Preference is governed by a latent preference relation equivalent to a scalar utility function (the higher the utility, the higher the preference for that state). Information about occupant’s preferences is available via pairwise - comparison queries (duels between two different states). We model our uncertainty about the utility via a Gaussian Process (GP) prior and the probability of the winner of each duel by means of a Bernoulli likelihood. This generalized preference model is then used in conjunction with pure exploration acquisition function to drive the elicitation process by actively selecting new queries to pose to the occupant. In this paper, an experimental framework is introduced which focusses on actively selecting new duels to learn the structure of utility everywhere in the state space with fewest possible queries. We illustrate the benefits of our framework by showing that these frameworks need drastically fewer duels for inferring the structure of underlying latent utility function as opposed to randomized data collection. In future, we are going to develop new frameworks which would help us in actively selecting new duels to infer the state with maximum utility function value (focusing on inferring the just the state with maximum utility function value rather than inferring it everywhere). We are also going to use these newly developed frameworks for sequential design of experiments to infer the preferences of new occupants working inside private offices.

Documents disponibles

Format PDF

Pages : 10

Disponible

  • Prix public

    20 €

  • Prix membre*

    15 €

* meilleur tarif applicable selon le type d'adhésion (voir le détail des avantages des adhésions individuelles et collectives)

Détails

  • Titre original : Design of experiments for learning personalized visual preferences of occupants in private offices.
  • Identifiant de la fiche : 30025076
  • Langues : Anglais
  • Source : 2018 Purdue Conferences. 5th International High Performance Buildings Conference at Purdue.
  • Date d'édition : 09/07/2018

Liens


Voir d'autres communications du même compte rendu (88)
Voir le compte rendu de la conférence