Computing Science and Mathematics
Technical Reports from 2014

The following Departmental Technical Reports were published from 2014 onwards. There is another page for reports published before 2014.

Where the author of a report below is highlighted as a link, click on it to contact the author (via home web page or email). Some reports are available for downloading. Click on the symbol preceding each article for the following:

PDF

Title: Predicting Emotional Dysregulation
Author(s): Kenneth J. Turner, Brian O'Neill, Gary Cornelius, Evan H. Magill
Date: September 2017
Number: CSM-200
Keywords: Assisted Living, Emotional Dysregulation, Machine Learning, Physiological Data, Smartwatch, Wearable Device
Abstract:

The background is given to mental health problems and their manifestation as emotional dysregulation (inability to control emotion). This motivates the work in the paper to predict emotional episodes using physiological data collected from mass-market devices (smartwatches). The design of a system to achieve this is explained. Data is collected from two commercial smartwatches for analysis and training of classifiers. Using colour-coded indications, feedback is given to staff and patients about the risk of an emotional episode. Medical staff can view detailed patient charts in order to make a considered judgement about the risk and how best to mitigate this. An overview is given of how features are extracted from the physiological data and then fed into machine learning algorithms that predict the likelihood of an episode. Results are presented for a small-scale evaluation with a partner hospital that looks after patients with brain injuries. The approach achieves a typical prediction accuracy up to 82% (aggressive episodes) or up to 68% (normal behaviour), anticipating episodes up to 4 hours ahead (physical data) or up to 4 days ahead (sleep data). Pointers are given to future developments of the work.

Title: Handling Conflict-Prone Policies in Multiple Domains
Author(s): Kenneth J. Turner
Date: July 2014
Number: CSM-199
Keywords: Call Control, Conflict Detection, Feature Interaction, Home Automation, Policy
Abstract:

Features as self-contained units of functionality have been widely used in telephony and in many software systems. Feature interaction is a long-recognised and insidious problem whereby features can interfere with each other. Policies (user-defined rules) have been used as a flexible approach in system management, and have been extended to networked applications such as home automation and sensor networks. Policy conflict is the analogue of feature interaction. A new approach is described for identifying conflict-prone policies in multiple domains. This relies on domain knowledge that defines the abstract effects of policy actions. Conflict filtering is performed statically, but supports conflict detection and resolution dynamically. The technique has been implemented in the RECAP tool (Rigorously Evaluated Conflicts Among Policies). With limited user guidance, this tool automatically detects conflicting actions and automatically generates resolutions for these. The approach is generic, but is illustrated with policies for call control and home automation. The technique has improved the scalability of conflict handling, and has considerably reduced the manual effort previously required to deal with conflicts. The technique has also been evaluated in live practice.

Title: Cognitively Inspired Fuzzy Based Audiovisual Speech Filtering
Author(s): Andrew K. Abel and >Amir Hussain
Date: April 2014
Number: CSM-198
Abstract:

In recent years, the established link between the various human communication production domains has become more widely utilised in the field of speech processing. Work by the authors and others has demonstrated that intelligently integrated audio and visual information can have a vital role to play in speech enhancement.

Of particular interest to our work is the potential use of visual information in future designs of hearing aid and listening device technology. A novel two-stage speech enhancement system, making use of audio-only beamforming, automatic lip tracking, and visually derived speech filtering, was initially developed by the authors and its potential evaluated in a previous paper. This work found that the use of visual information was of benefit in some scenarios, but not all. In addition to the use of visual information based on the concept of lip-reading, there is also scope for the development of cognitively inspired speech processing approaches that function in a similar manner to the multimodal attention-switching nature of the human mind. One example of this is the use of the visual modality for speech filtering in only the most appropriate environments (such as when there is a lot of background noise, and when the visual information is of a suitable quality to be used). This cognitively inspired approach ensures that visual information is only used when it is expected to improve performance.

It is also worth considering the possibility of environments where multimodal information may be sporadic and of varying quality. One single speech filtering approach may produce inadequate results when applied to a wide range of environments. To alleviate this, we present a cognitively inspired, fuzzy logic based, multimodal speech filtering system that considers audio noise level (using a similar manner to level detectors used in conventional hearing aids) and evaluates the visual signal quality in order to carry out more intelligent automated speech filtering. These detectors are used as part of a fuzzy logic based system to determine the optimal speech filtering solution for each frame of speech. When tested with a wide variety of challenging data, the results show that a nuanced approach is capable of automatically switching between approaches when considered appropriate. The proposed approach is intended to be a cognitively inspired, scalable and adaptable framework, with promising initial results.

Title: Augmenting Metaheuristics with Rewriting Systems
Author(s): Jerry Swan, Martin Edjvet and Ender Özcan
Date: January 2014
Number: CSM-197
Abstract:

We describe the use of a rewriting system to determine equivalence classes over the search-space of optimisation problems. These equivalence classes may be used to derive redundant subsequences in the search-space for incorporation into metaheuristics such as backtracking, genetic algorithms and tabu-search. We use this approach as the basis for a new Tabu search variant (Equational TS) and apply it to the Quadratic Assignment Problem, yielding significant results in terms of the number of iterations to convergence.

Title: The Accent Policy System for Home Care
Author(s): Kenneth J. Turner
Date: April 2014 (Revised)
Number: CSM-188
Keywords: ACCENT (Advanced Component Control Enhancing Network Technologies), APPEL policy language (ACCENT Project Policy Environment/Language), Goal, OSGi (Open Systems Gateway initiative), Policy
Abstract:

This report describes the architecture, installation and configuration of the ACCENT policy system. It is seen that virtually all the ACCENT components are bundles deployed on an OSGi system. These bundles communicate using the OSGi event service. The details are given of how to set up and configure each of the bundles.

Title: The Accent Policy Wizard
Author(s): Kenneth J. Turner
Date: April 2014 (Revised)
Number: CSM-166
Abstract:

The Accent project (Advanced Component Control Enhancing Network Technologies) developed a practical and comprehensive policy system for call control/Internet telephony. The policy system has subsequently been extended for management of sensor networks/wind farms on the Prosen project (Proactive Control of Sensor Networks). The policy system has also been extended for management of home care/telecare on the Match project (Mobilising Advanced Technologies for Care at Home).

This report focuses on a web-based policy wizard that acts as the primary interface between end users and the policy system. The policy wizard has an intimate knowledge of the Appel policy language (Adaptable and Programmable Policy Environment and Language). The wizard makes use of domain-specific ontologies so that it can be used in any application. The wizard allows end users to create policies using near-natural language without knowing or seeing XML, and to upload them to the policy system. The wizard also provides a number of convenience functions such as predefined policy templates, editing and activating existing policies, and defining policy variables. Besides regular policies, the wizard supports resolution policies for handling conflicts among policy actions.

Title: Appel: An Adaptable and Programmable Policy Environment and Language
Author(s): Kenneth J. Turner, Stephan Reiff-Marganiec, Lynne Blair, Gavin A. Campbell and Feng Wang
Date: April 2014 (Revised)
Number: CSM-161
Abstract:

The Accent project (Advanced Component Control Enhancing Network Technologies) developed a practical and comprehensive policy system for call control/Internet telephony. The policy system has subsequently been extended for management of sensor networks/wind farms and of home care/telecare.

This report focuses on Appel (Adaptable and Programmable Policy Environment and Language). It provides an overview of the language, and presents the language in XML schema form. The core language has been instantiated for call control, for sensor networks, and for home care. Sample goals and policies of different kinds are provided to illustrate these applications.

Up one level to Computing Science/Mathematics Reports