FDive: Learning Relevance Models Using Pattern-based Similarity Measures

Abstract

Figure 1: FDIVE learns to distinguish relevant from irrelevant data through an iteratively improving classification model by learning the best-fitting feature descriptor and distance function. (1) Users express their notion of relevance by labeling a set of query items, in this case, images. (2) These labels are used to rank all similarity measures by their ability to distinguish relevant from irrelevant data. (3) The system applies the selected similarity measure to learn a Self-Organizing Map (SOM)-based relevance model. Users explore and refine the model by supplying relevance labels in uncertain data regions, especially near the decision boundaries. ABSTRACTThe detection of interesting patterns in large high-dimensional datasets is difficult because of their dimensionality and pattern complexity. Therefore, analysts require automated support for the extraction of relevant patterns. In this paper, we present FDIVE, a visual active learning system that helps to create visually explorable relevance models, assisted by learning a pattern-based similarity. We use a small set of user-provided labels to rank similarity measures, consisting of feature descriptor and distance function combinations, by their ability to distinguish relevant from irrelevant data. Based on the best-ranked similarity measure, the system calculates an interactive Self-Organizing Map-based relevance model, which classifies data according to the cluster affiliation. It also automatically prompts further relevance feedback to improve its accuracy. Uncertain areas, especially near the decision boundaries, are highlighted and can be refined by the user. We evaluate our approach by comparison to state-of-the-art feature selection techniques and demonstrate the usefulness of our approach by a case study classifying electron microscopy images of brain cells. The results show that FDIVE enhances both the quality and understanding of relevance models and can thus lead to new insights for brain research.

Interactive Graph Construction for Graph-Based Semi-Supervised Learning

“…It not only enables users to query data for labeling via active learning, but also allows better understanding and refining of the classification model via visualizations. Dennig et al [23] provided FDIVE to detect the best-fitting features and distance functions based on labels provided by users. The features and distance functions are then used to train a SOM-based relevance model, which is visually explorable and can be further refined by providing more labels.…”

Peax Interactive Visual Pattern Search in Sequential Data Using Unsupervised Deep Representation Learning

“…A similar approach has been proposed by Behrisch et al [3] for scatter plots, where the user interactively trains a classifier based on the user's feedback to learn to capture the interestingness of scatter plot views. Recently, Dennig et al [15] proposed a system to interactively learn the best combination of feature descriptors and a distance function for pattern separability, assuming the existence and user's knowledge about the feature descriptors. In contrast, CueFlik [21] is a tool for interactive concept learning in image search, which allows the user to rank the results of a text-based image search using simple binary feedback.…”

PSEUDo: Interactive Pattern Search in Multivariate Time Series with Locality-Sensitive Hashing and Relevance Feedback

“…On the one hand, users need a mechanism to define a visual search pattern, for instance, through query-by-sketch or query-by-example [21], and on the other hand, a way to steer the retrieval process. One opportunity here is active learning-based approaches that rely on the user's explicit relevance feedback, such as in [9,23,50].…”

“…The user can not process more than maybe a few dozen without being stuck in a tedious labeling process. To follow the central idea of (visual) active learning [4,23,27], we choose representatives based on two concepts: exploitation and exploration. Given a query pattern and (initial) feedback from the user, PSEUDo will exploit this feedback to choose archetypal representatives.…”

“…The remainder is retrieving interesting patterns based on the user's notion of similarity. To define similarity, researchers have created various distance metrics [35] and, similar to our approach in PSEUDo, (adaptive) search and manifold learning strategies [12,23,59].…”