ID: Password: [Register]  
Home | Research | Publications | People | Courses | Resources | Board KOREAN  
 
  • Overview
  • Agent
  • Biometrics
  • Smart Phone
  • Ubiquitous
  • Robot
  • Projects

  •   Agent

    Artificial Assistant

    Team Manager : Jin-Hyuk Hong (hjinh_at_cs.yonsei.ac.kr)

    Team Members: Sung-Soo Lee (lss_at_sclab.yonsei.ac.kr), Joo-Huem Baek (bjh_at_sclab.yonsei.ac.kr), In-Jee Song (schunya_at_sclab.yonsei.ac.kr)

    1. Introduction

       We have researched an artificial secretary using the integrated model of human functionality that is based on the neural information processing. It infers the context and user intention from the environment by using the informal inference techniques, and improves itself to adapt to the target application gradually.

    2. Methods

    • Informal cognitive inference: Understanding the user's intention and situation is required for providing a proper service. In this research, we propose a cognitive inference technique using informal Bayesian network to manage the compound information. It not only manages uncertain problems probabilistically but also infers the user's intention actively by the mixed initiative interaction of the hierarchical modeling. We have confirmed the usefulness of the proposed informal Bayesian network by applying it to managing the user's schedules, solving the ambiguous situation of the service robot, etc.    Since some problems are hard to solve with a single model in real applications, a technique that combines multiple models might be useful. We also present informal combination techniques. A fuzzy integral of multiple SASOM is used to infer the preference of the user on web contents [1], while a set of various and reliable Bayesian networks is generated by the speciated genetic algorithm.
    • Incremental learning model: Template-based knowledge acquisition and sentence generation using genetic programming are developed for the incremental improvement of the proposed interactive artificial secretary.
      • Knowledge acquisition ─ When a query inputs, it analyzes the speech-act by using the finite state machine. It actively collects additional information and constructs the pattern-response pair by using a template corresponding to the speech-act [2].
      • Sentence learning ─ A sentence is represented by a sentence plan tree that is composed of leaf nodes and joint nodes. Sentences are changed to be adaptive to the domain by genetic programming, where the fitness of a sentence is given by the user.

    Figure 1. An integrated model of human functionality for the artificial secretary

    3. Experimental Results

    Fig. 2 shows the interactive artificial secretary that is applied to the schedule management application. It recommends most convenient schedule by analyzing the user's preference and situation through the informal cognitive inference.

    Figure 2. Schedule management by the interactive artificial secretary

    4. Publications

    [1] K.-J. Kim and S.-B. Cho, "Fuzzy integration of structure adaptive SOM's for web content mining," Fuzzy Sets and Systems, vol. 148, no. 1, pp. 43-60, 2004.

    [2] J.-H. Hong and S.-B. Cho, "Domain-adaptive conversational agent with two-stage dialogue management," Lecture Notes in Computer Science, vol. 3339, pp. 1147-1153, 2004.

    [3] J.-H. Hong, Y.-S. Song, and S.-B. Cho, "Mixed-initiative human-robot interaction using a hierarchical Bayesian network," ICRA, 2005.

    [4] J.-H. Hong, and S.-B. Cho, "Context management of conversational agent using two-stage Bayesian network," J. Korea Information Science Society: Practice of Computing, vol.10, no. 1, pp. 89-98, Feb 2004.

     


      Last update : 2015.04.06 @ Softcomputing lab info : webmaster at sclab.yonsei.ac.kr

      Soft Computing Laboratory,Dept. of Computer Science, Yonsei University,
      134 Shinchon-Dong, Seodaemoon-Gu, Seoul, 120-749, Korea
      Telephone : +82-2-2123-3877 or 4803