MQP Topics

• Adaptive systems and visualization

  The field of machine learning deals with systems that adapt to a particular environment or task. The adaptation process may involve feedback from a "teacher" (supervised learning) or it may occur without such direct feedback (unsupervised learning, self-organization). Visualization techniques have the potential to provide important insights into the dynamics of adaptation. In turn, adaptive systems may be used to develop spatial/visual representations of multivariate datasets. Our MQP's in this area explore such bidirectional connections between adaptive systems and visualization techniques.
  • "Visualization of Neural Network Learning"
    Student: Matthew Streeter
    Coadvisors: Professor Matt Ward

  ---

• Combination of relevance rankings

  The following is a seemingly simple abstract question that is of paramount importance in achieving effective collaboration among web search engines, in integrating heterogeneous databases, and in dealing with uncertainty in knowledge-based systems:

  Given the opinions of several "judges" in the form of rankings of various candidate options together with priority values, how can one produce combined priority values for the various candidates and an associated consensus ranking?

  Several MQP's will investigate approaches to addressing the above question; each MQP will include the implementation of a system that incorporates some of the approaches in one of the applications mentioned above. Two distinct sources of relevant techniques are mathematical voting theory and the transformation-based approach recently developed by Professor Alvarez.
  • "Combination of Relevance Rankings for Web Metasearch"
    Student: Anna Novikov
    Coadvisors: Professor Carolina Ruiz

  ---

• AI and data mining

  The ability to extract concise descriptions from raw data is very important in a variety of activities, including scientific research, analysis of stock prices, and online commerce. This extraction process is known as data mining. Our MQP's in data mining combine techniques from artificial intelligence (AI) and other areas to produce systems capable of learning useful descriptions from large databases.
  • "TinFOIL: A System for Learning Rules from Examples"
    Winner, Provost's MQP Award for Computer Science, 1999
    Student: Matthew L. Berube
    Coadvisors: Professor Carolina Ruiz

    "Mining Patterns from Sequence Data"

    Student: Tara Halwes

    Coadvisors: Professor Carolina Ruiz

  ---

• Recommender systems

  The advent of Internet commerce provides a fertile medium for the development of marketing tools that adapt to individual potential customers. Recommender systems in particular provide user-specific views of available item collections based on profiles of user preferences and past experiences involving user behavior. Fields such as Artificial Intelligence (AI) and Data Mining have developed techniques that should have a significant impact on recommender system design. A series of MQP's in this area will explore a variety of approaches to recommender systems and collaborative filtering.
  • "Collaborative and Content-Based Filtering using Neural Networks"
    Students: Amy Kao, Dana Rock
    Coadvisors: Professor Carolina Ruiz