Publications
Explorations
Memories in Adaptive Resonance Theory (ART) networks are based on matched patterns
that focus attention on those portions of bottom-up inputs that match active top-down expectations.
While this learning strategy has proved successful for both brain models and applications,
computational examples show that attention to early critical features may later distort
memory representations during online fast learning. For supervised learning, biased ARTMAP (bARTMAP)
solves the problem of over-emphasis on early critical features by directing attention away
from previously attended features after the system makes a predictive error. Small-scale,
hand-computed analog and binary examples illustrate key model dynamics.
Two-dimensional simulation examples demonstrate the evolution of bARTMAP
memories as they are learned online. Benchmark simulations show that featural biasing
also improves performance on large-scale examples. One example, which predicts movie
genres and is based, in part, on the Netflix Prize database, was developed for this project.
Both first principles and consistent performance improvements on all simulation studies
suggest that featural biasing should be incorporated by default in all ARTMAP systems.
Benchmark datasets and bARTMAP code are available here.
The green arrows indicate improvements, relative to fuzzy ARTMAP, in per-genre predictions made by biased ARTMAP.
The genre prediction benchmark dataset and generating code are available here.
CONFIGR (CONtour FIgure GRound) is a computational model based on principles of biological vision that completes sparse and noisy image figures. Within an integrated vision/recognition system, CONFIGR posits an initial recognition stage which identifies figure pixels from spatially local input information. The resulting, and typically incomplete, figure is fed back to the ''early vision'' stage for long-range completion via filling-in. The reconstructed image is then re-presented to the recognition system for global functions such as object recognition.
In the CONFIGR algorithm, the smallest independent image unit is the visible pixel, whose size defines a computational spatial scale. Once the pixel size is fixed, the entire algorithm is fully determined, with no additional parameter choices. Multi-scale simulations illustrate the vision/recognition system. Open-source CONFIGR code is available online, but all examples can be derived analytically, and the design principles applied at each step are transparent. The model balances filling-in as figure against complementary filling-in as ground, which blocks spurious figure completions. Lobe computations occur on a subpixel spatial scale. Originally designed to fill-in missing contours in an incomplete image such as a dashed line, the same CONFIGR system connects and segments sparse dots, and unifies occluded objects from pieces locally identified as figure in the initial recognition stage. The model self-scales its completion distances, filling-in across gaps of any length, where unimpeded, while limiting connections among dense image-figure pixel groups that already have intrinsic form. Long-range image completion promises to play an important role in adaptive processors that reconstruct images from highly compressed video and still camera images.
Image fusion has been defined as the acquisition, processing and synergistic combination of information provided by various sensors or by the same sensor in many measuring contexts. (Simone et al., 2002) When multiple sources provide inconsistent data, such methods are called upon to select the accurate information components. As quoted by the International Society of Information Fusion: Evaluating the reliability of different information sources is crucial when the received data reveal some inconsistencies and we have to choose among various options. For example, independent sources might label an object beach or road or river. A fusion method could address this problem by weighing the confidence and reliability of each source, merging complementary information, or gathering more data. In any case, at most one of these answers is correct.
Tagging, the process of user annotation of data, is a contemporary version of the indexing schemes of the early web. Users organize tags according to a folksonomy, a term that invokes both the self-organizing nature of this enterprise and the perceived failure of connectivity-based search engines. Folksonomy is generally defined as an internet-based methodology for information retrieval that is based on the set of collaboratively generated user tags.
We have undertaken a pilot study examining the application of information fusion methods to text analysis with an initial goal of identifying synonyms. With the freely chosen vocabulary of the tagging culture, users are apt to miss cross-references when choosing different terms for similar items. Special cases of synonyms include misspelled words and terms whose identity the user wishes to conceal. Two words with the same meaning may rarely co-occur: their synonymous nature is inferred by their appearance in various shared contexts. Learning of contexts is thus the proposed research strategy for identifying synonyms. Different terms that appear in related contexts during training are identified as synonyms during testing. The computational basis for the project's synonym identification procedure is the distributed representation of learned codes, a method which was developed for the image-based neural fusion applications.
Biased ART:
solves the problem of over-emphasis on early critical features by directing attention away
from previously attended features.
CONFIGR:
is a first-generation neural model for contour interpolation.
Knowledge Discovery from Labeled Web Documents:
is an attempt to sythesize disparate semantic information from user tags.