Graham Bent

Distributed policy based access to networked heterogeneous ISR data sources

Within a coalition environment, ad hoc Communities of Interest (CoIs) come together, perhaps for only a short time, with different sensors, sensor platforms, data fusion elements, and networks to conduct a task (or set of tasks) with different coalition members taking different roles. In such a coalition, each organization will have its own inherent restrictions on how it will interact with the others. These are usually stated as a set of policies, including security and privacy policies. The capability that we want to enable for a coalition operation is to provide access to information from any coalition partner in conformance with the policies of all. One of the challenges in supporting such ad-hoc coalition operations is that of providing efficient access to distributed sources of data, where the applications requiring the data do not have knowledge of the location of the data within the network. To address this challenge the International Technology Alliance (ITA) program has been developing the concept of a Dynamic Distributed Federated Database (DDFD), also know as a Gaian Database. This type of database provides a means for accessing data across a network of distributed heterogeneous data sources where access to the information is controlled by a mixture of local and global policies. We describe how a network of disparate ISR elements can be expressed as a DDFD and how this approach enables sensor and other information sources to be discovered autonomously or semi-autonomously and/or combined, fused formally defined local and global policies.

Machine learning approaches for small data in sensor fusion applications

Machine learning approaches like deep neural networks have proven to be very successful in many domains. However, they require training on a huge volumes of data. While these approaches work very well in a few selected domains where a large corpus of training data exists, they shift the bottleneck in development of machine learning applications to the data acquisition phase and are difficult to use in domains where training data is hard to acquire. For sensor fusion applications in coalition operations, access to good training data that will be suitable for real-life applications is hard to get. The training data sets available are limited in size. For these domains, we need to explore approaches for machine learning which can work with small amounts of data. In this paper, we will look at the current and emerging approaches which allow us to build machine learning models when access to the training data is limited. The approaches examined include statistical machine learning, transfer learning, synthetic data generation, semi-supervised learning and one-shot learning.

Audio Analysis as a Control Knob for Social Sensing

While humans can act as effective sensors, human input is subject to a high degree of error and highly dependent on the context. Furthermore, extracting the signal from the noise for social sensing is a difficult challenge. One approach to improving the accuracy of social sensing is to use physical sensors as a control knob for social sensing algorithms. In this paper, we present an architecture for using audio sensors as a way to control an algorithm used for social sensing of interesting events. We present various use cases where the architecture is applicable, and go into the details of one specific use case, namely using crowd behavior in a golf-course to identify and control social media feeds related to the course.

Using machine learning to emulate human hearing for predictive maintenance of equipment

At the current time, interfaces between humans and machines use only a limited subset of senses that humans are capable of. The interaction among humans and computers can become much more intuitive and effective if we are able to use more senses, and create other modes of communicating between them. New machine learning technologies can make this type of interaction become a reality. In this paper, we present a framework for a holistic communication between humans and machines that uses all of the senses, and discuss how a subset of this capability can allow machines to talk to humans to indicate their health for various tasks such as predictive maintenance.

An efficient hypercube labeling schema for dynamic Peer-to-Peer networks

This paper addresses the general problem of reducing unnecessary message transmission thereby lowering overall bandwidth utilization in a Peer-to-Peer (P2P) network. In particular, we exploit the characteristics of a P2P network engineered to resemble a hypercube. The reason for doing this is to achieve constant computation time for inter-node distances that are needed in the process of query optimization. To realize such a hypercube-like engineered structure, we develop a new labeling scheme which assigns identifiers (labels) to each node and then uses these labels to determine inter-node distances as is done in a hypercube, thus eliminating the need to send out queries to find the distance from one node to another. The labels allow for creating a virtual overlay which resembles a hypercube. We prove that the labeling scheme does in fact allow for reducing bandwidth utilization in the network. To confirm our theoretical findings we conduct various experiments with randomly selected P2P networks of various sizes. Detailed statistics on the outcome of these experiments are provided which show clearly the practical utility of the labeling approach. We assign labels to each node in a P2P network to facilitate and improve communication.Labels create a virtual hypercube overlay.Known hypercube algorithms are adopted in this environment.Theoretical and experimental findings prove the feasibility and efficiency of this approach.

Computing on encrypted data and its applicability to a coalition operations environment

Coalition operations often invoke the sharing of information and IT infrastructure amongst partners. Whilst there may be a coalition ‘need to share’ data this is often tempered by a ‘need to know’ principle that often prevents valuable information from being exchanged, particularly with classified data. Ideally, coalition partners would wish to share data that can be used to compute specific results that are only relevant to a given operation, without revealing all of the shared information. In this paper we will present the concept of a secure coalition cloud architecture that is capable of storing encrypted data and of performing arbitrary computations on the encrypted data on behalf of users, without at any stage having to decrypt it. To do this we make use of a fully homomorphic encryption scheme using a novel approach for managing encryption and decryption keys in a public key infrastructure (PKI) setting.