Search Constraints

Start Over
Filtering by: Creator Yazidi, Anis Remove constraint Creator: Yazidi, Anis

Search Results

  • On the online classification of data streams using weak estimators

    Resource Type:
    Conference Proceeding
    Creator:
    Tavasoli, Hanane, Oommen, B. John, and Yazidi, Anis
    Abstract:
    In this paper, we propose a novel online classifier for complex data streams which are generated from non-stationary stochastic properties. Instead of using a single training model and counters to keep important data statistics, the introduced online classifier scheme provides a real-time self-adjusting learning model. The learning model utilizes the multiplication-based update algorithm of the Stochastic Learning Weak Estimator (SLWE) at each time instant as a new labeled instance arrives. In this way, the data statistics are updated every time a new element is inserted, without requiring that we have to rebuild its model when changes occur in the data distributions. Finally, and most importantly, the model operates with the understanding that the correct classes of previously-classified patterns become available at a later juncture subsequent to some time instances, thus requiring us to update the training set and the training model. The results obtained from rigorous empirical analysis on multinomial distributions, is remarkable. Indeed, it demonstrates the applicability of our method on synthetic datasets, and proves the advantages of the introduced scheme.
    Date Created:
    2016-01-01

  • A novel clustering algorithm based on a non-parametric “anti-bayesian” paradigm

    Resource Type:
    Conference Proceeding
    Creator:
    Yazidi, Anis, Oommen, B. John, and Hammer, Hugo Lewi
    Abstract:
    The problem of clustering, or unsupervised classification, has been solved by a myriad of techniques, all of which depend, either directly or implicitly, on the Bayesian principle of optimal classification. To be more specific, within a Bayesian paradigm, if one is to compare the testing sample with only a single point in the feature space from each class, the optimal Bayesian strategy would be to achieve this based on the distance from the corresponding means or central points in the respective distributions. When this principle is applied in clustering, one would assign an unassigned sample into the cluster whose mean is the closest, and this can be done in either a bottom-up or a top-down manner. This paper pioneers a clustering achieved in an “Anti-Bayesian” manner, and is based on the breakthrough classification paradigm pioneered by Oommen et al. The latter relies on a radically different approach for classifying data points based on the non-central quantiles of the distributions. Surprisingly and counter-intuitively, this turns out to work equally or close-to-equally well to an optimal supervised Bayesian scheme, which thus begs the natural extension to the unexplored arena of clustering. Our algorithm can be seen as the Anti-Bayesian counter-part of the wellknown k-means algorithm (The fundamental Anti-Bayesian paradigm need not just be used to the k-means principle. Rather, we hypothesize that it can be adapted to any of the scores of techniques that is indirectly based on the Bayesian paradigm.), where we assign points to clusters using quantiles rather than the clusters’ centroids. Extensive experimentation (This paper contains the prima facie results of experiments done on one and two-dimensional data. The extensions to multi-dimensional data are not included in the interest of space, and would use the corresponding multi-dimensional Anti-Na¨ıve-Bayes classification rules given in [1].) demonstrates that our Anti-Bayesian clustering converges fast and with precision results competitive to a k-means clustering.
    Date Created:
    2015-01-01