Cevher V.,University of Maryland University College |
Indyk P.,University of Warsaw |
Carin L.,Signal Innovations Group, Inc. |
Baraniuk R.,Rice University
IEEE Signal Processing Magazine | Year: 2010
Many applications in digital signal processing, machine learning, and communications feature a linear regression problem in which unknown data points, hidden variables, or code words are projected into a lower dimensional space via © 2006 IEEE. Source
Agency: Department of Defense | Branch: Air Force | Program: SBIR | Phase: Phase I | Award Amount: 97.31K | Year: 2011
The proposed Phase I program will develop a Bayesian framework for modeling and predicting track uncertainty and performance as a function of OCs and integrate this framework with algorithms for tracking vehicles in WAMI. The proposed Performance and Uncertainty Modeling and Prediction for Tracking (PUMP-T) framework is comprised of the following key elements: (1) explicit extraction of operating conditions (OCs), (2) calculation of information-theoretic metrics for quantifying uncertainty in track posterior probability density functions (PDFs), (3) sparse Bayesian regression algorithms for modeling track uncertainty as a function of OCs, (4) Bayesian semi-supervised regression/classification algorithms for modeling track performance as a function of OCs, (5) an offline process for training the uncertainty and performance models, and (6) an online process for predicting track uncertainty and performance and directly validating the models in the context of the observed data. The underlying prediction models driving the PUMP-T framework will be founded on posterior PDFs over track states inferred within a rigorous Bayesian framework. In Phase I, the PUMP-T framework will be designed to be agnostic to the specific tracking algorithm employed such that alternate tracking algorithms can be integrated and evaluated within the PUMP-T framework. BENEFIT: While the PUMP-T framework will be leveraged for video tracking applications in the proposed program, the framework easily generalizes to other exploitation algorithms (e.g. detection and classification) and other sensor modes. Medical device manufacturers are developing cutting edge sensors leading to revolutionary advances in diagnosis of a variety of diseases, especially cancer. These devices have the ability to gather a significant amount of data, much more than a physician or technician can handle alone. Therefore, automated processing, including rigorous understanding of uncertainty and performance predictions, is necessary for realizing the full potential of these devices. While the video exploitation software leveraged under this SBIR offer significant impact on a broad spectrum of DoD and IC applications today and in the future, the exploitation software also offers significant benefit to several existing and new commercial markets, including municipal public safety, traffic analysis and flow optimization, mining of video for targeted advertising, corporate security, and mining of surveillance cameras for retail applications. SIG has had discussions with companies and venture capital interests that participate in such markets and believes that the exploitation products of this SBIR would benefit these video content extraction markets.
Agency: Department of Defense | Branch: Navy | Program: SBIR | Phase: Phase II | Award Amount: 499.99K | Year: 2014
Signal Innovations Group (SIG) proposes a Phase II program that addresses three primary capabilities for applications involving high-dimensional observed data: reversible nonlinear data dimensionality reduction, static and dynamic synthesis of data in the observed space based on a low-dimensional latent space, and intuitive and meaningful human interaction with the data in both the observed and latent spaces. Each of these three capabilities will be addressed in the Phase II Base effort through the development of models and algorithms, visualization techniques, and a prototype user interface. Additionally, proposed Phase II Options 1 and 2 will focus on maturing and optimizing the technology and software, ultimately leading to transition and integration with a targeted program of record.
Agency: Department of Defense | Branch: Air Force | Program: SBIR | Phase: Phase II | Award Amount: 1.05M | Year: 2013
ABSTRACT: This program will develop salient feature analysis and saliency cueing capabilities for SAR-based exploitation and demonstrate this technology with operationally relevant data and classifiers. This saliency technology will immediately enable the development of exploitable, robust, compact feature databases for sustainable radar CID. This new SAR-based saliency capability will complement existing HRR-based saliency technology by providing either independent or joint discovery of sparse sets of robust, exploitable features across these two common radar sensing modalities, in support of air-to-air or air-to-ground moving and stationary CID missions. This framework provides a balance between a compact target representation and target generalization across in-class variations. The underlying probabilistic model captures the uncertainty inherent to the training process due to limited or noisy data, and propagates these probabilities in a mathematically rigorous manner to downstream processes. The proposed work for this program will focus on near-term capabilities of sustainable radar CID database development and will lay the fundamental groundwork for potential fusion of radar and EO features based upon saliency and target geometry. BENEFIT: Phase II will result in a combined salient feature analysis, saliency cueing, and classification performance analysis capability for both HRR and SAR radar modalities integrated into POSSIBLE. This capability will enable the automatic identification of compact exploitable radar signature features over fewer target aspect states for more efficient and sustainable CID databases. While the salient feature analysis framework will be leveraged on the proposed program for classification with radar sensor modalities, the framework easily generalizes to other data modalities and exploitation applications, such as detection, recognition, identification, and general data categorization. Medical device manufacturers are developing cutting-edge sensors and equipment that is leading to revolutionary advances in non-invasive diagnosis of a variety of diseases, especially cancer. These devices have the ability to gather a significant amount of data, much more than a physician or technician can handle alone. Additionally, commercial satellite imagery providers are developing sensors that collect extremely high-resolution imagery. While more pixels generally provide more information, large amounts of data also lead to increased burden on image analysis, storage, and query. Therefore, automated processing, including rigorous understanding of salient features, is necessary for realizing the full potential of these new medical devices and satellite imagery. SIG is actively engaged with sensor and data providers in each of these industries and will leverage these relationships to commercialize the products of this Phase II SBIR.
Agency: Department of Defense | Branch: Navy | Program: SBIR | Phase: Phase II | Award Amount: 734.27K | Year: 2010
In the proposed Phase II program, the methods developed and implemented during Phase I research will be fully integrated within a common Bayesian in situ learning framework. We have developed several Bayesian classifiers, to which we will apply label acquisition and label confidence techniques. Additionally, we will extend the in situ learning framework to include multi-task learning. Previously collected sensing data are often available from different sensors or environments. Not all data are related, however the potential exists to share information between related tasks and exploit the contextual information of previous tasks. The current in situ learning process is inherently myopic; the algorithm identifies the single most-informative data sample. The ability to select multiple samples without relearning the classifier can increase computational efficiency and maximize analyst workload. Based on the theory of submodular functions, non-myopic in situ learning techniques for subset selection will be developed and integrated into the Bayesian framework. Finally, new statistical embedding technology will be investigated that allows an analyst to synthesize data for training and to augment the label acquisition process. A low-dimensional embedded space may be visualized, and any location on the manifold can be recreated in the original high-dimensional space.