GYAN: A methodology for rule extraction from artificial neural networks

Artificial neural network (ANN) learning methods provide a robust and non-linear approach to approximating the target function for many classification, regression and clustering problems. ANNs have demonstrated good predictive performance in a wide variety of practical problems. However, there are strong arguments as to why ANNs are not sufficient for the general representation of knowledge. The arguments are the poor comprehensibility of the learned ANN, and the inability to represent explanation structures. The overall objective of this thesis is to address these issues by: (1) explanation of the decision process in ANNs in the form of symbolic rules (predicate rules with variables); and (2) provision of explanatory capability by mapping the general conceptual knowledge that is learned by the neural networks into a knowledge base to be used in a rule-based reasoning system. A multi-stage methodology GYAN is developed and evaluated for the task of extracting knowledge from the trained ANNs. The extracted knowledge is represented in the form of restricted first-order logic rules, and subsequently allows user interaction by interfacing with a knowledge based reasoner. The performance of GYAN is demonstrated using a number of real world and artificial data sets. The empirical results demonstrate that: (1) an equivalent symbolic interpretation is derived describing the overall behaviour of the ANN with high accuracy and fidelity, and (2) a concise explanation is given (in terms of rules, facts and predicates activated in a reasoning episode) as to why a particular instance is being classified into a certain category.

Algebraic attacks on clock-controlled stream ciphers

Stream ciphers are encryption algorithms used for ensuring the privacy of digital telecommunications. They have been widely used for encrypting military communications, satellite communications, pay TV encryption and for voice encryption of both fixed lined and wireless networks. The current multi year European project eSTREAM, which aims to select stream ciphers suitable for widespread adoptation, reflects the importance of this area of research. Stream ciphers consist of a keystream generator and an output function. Keystream generators produce a sequence that appears to be random, which is combined with the plaintext message using the output function. Most commonly, the output function is binary addition modulo two. Cryptanalysis of these ciphers focuses largely on analysis of the keystream generators and of relationships between the generator and the keystream it produces. Linear feedback shift registers are widely used components in building keystream generators, as the sequences they produce are well understood. Many types of attack have been proposed for breaking various LFSR based stream ciphers. A recent attack type is known as an algebraic attack. Algebraic attacks transform the problem of recovering the key into a problem of solving multivariate system of equations, which eventually recover the internal state bits or the key bits. This type of attack has been shown to be effective on a number of regularly clocked LFSR based stream ciphers. In this thesis, algebraic attacks are extended to a number of well known stream ciphers where at least one LFSR in the system is irregularly clocked. Applying algebriac attacks to these ciphers has only been discussed previously in the open literature for LILI-128. In this thesis, algebraic attacks are first applied to keystream generators using stop-and go clocking. Four ciphers belonging to this group are investigated: the Beth-Piper stop-and-go generator, the alternating step generator, the Gollmann cascade generator and the eSTREAM candidate: the Pomaranch cipher. It is shown that algebraic attacks are very effective on the first three of these ciphers. Although no effective algebraic attack was found for Pomaranch, the algebraic analysis lead to some interesting findings including weaknesses that may be exploited in future attacks. Algebraic attacks are then applied to keystream generators using (p; q) clocking. Two well known examples of such ciphers, the step1/step2 generator and the self decimated generator are investigated. Algebraic attacks are shown to be very powerful attack in recovering the internal state of these generators. A more complex clocking mechanism than either stop-and-go or the (p; q) clocking keystream generators is known as mutual clock control. In mutual clock control generators, the LFSRs control the clocking of each other. Four well known stream ciphers belonging to this group are investigated with respect to algebraic attacks: the Bilateral-stop-and-go generator, A5/1 stream cipher, Alpha 1 stream cipher, and the more recent eSTREAM proposal, the MICKEY stream ciphers. Some theoretical results with regards to the complexity of algebraic attacks on these ciphers are presented. The algebraic analysis of these ciphers showed that generally, it is hard to generate the system of equations required for an algebraic attack on these ciphers. As the algebraic attack could not be applied directly on these ciphers, a different approach was used, namely guessing some bits of the internal state, in order to reduce the degree of the equations. Finally, an algebraic attack on Alpha 1 that requires only 128 bits of keystream to recover the 128 internal state bits is presented. An essential process associated with stream cipher proposals is key initialization. Many recently proposed stream ciphers use an algorithm to initialize the large internal state with a smaller key and possibly publicly known initialization vectors. The effect of key initialization on the performance of algebraic attacks is also investigated in this thesis. The relationships between the two have not been investigated before in the open literature. The investigation is conducted on Trivium and Grain-128, two eSTREAM ciphers. It is shown that the key initialization process has an effect on the success of algebraic attacks, unlike other conventional attacks. In particular, the key initialization process allows an attacker to firstly generate a small number of equations of low degree and then perform an algebraic attack using multiple keystreams. The effect of the number of iterations performed during key initialization is investigated. It is shown that both the number of iterations and the maximum number of initialization vectors to be used with one key should be carefully chosen. Some experimental results on Trivium and Grain-128 are then presented. Finally, the security with respect to algebraic attacks of the well known LILI family of stream ciphers, including the unbroken LILI-II, is investigated. These are irregularly clock- controlled nonlinear filtered generators. While the structure is defined for the LILI family, a particular paramater choice defines a specific instance. Two well known such instances are LILI-128 and LILI-II. The security of these and other instances is investigated to identify which instances are vulnerable to algebraic attacks. The feasibility of recovering the key bits using algebraic attacks is then investigated for both LILI- 128 and LILI-II. Algebraic attacks which recover the internal state with less effort than exhaustive key search are possible for LILI-128 but not for LILI-II. Given the internal state at some point in time, the feasibility of recovering the key bits is also investigated, showing that the parameters used in the key initialization process, if poorly chosen, can lead to a key recovery using algebraic attacks.

Large area noise evaluation

A software tool (DRONE) has been developed to evaluate road traffic noise in a large area with the consideration of network dynamic traffic flow and the buildings. For more precise estimation of noise in urban network where vehicles are mainly in stop and go running conditions, vehicle sound power level (for acceleration/deceleration cruising and ideal vehicle) is incorporated in DRONE. The calculation performance of DRONE is increased by evaluating the noise in two steps of first estimating the unit noise database and then integrating it with traffic simulation. Details of the process from traffic simulation to contour maps are discussed in the paper and the implementation of DRONE on Tsukuba city is presented

Analytical modeling and sensitivity analysis for travel time estimation on signalized urban networks

This paper presents a model for estimation of average travel time and its variability on signalized urban networks using cumulative plots. The plots are generated based on the availability of data: a) case-D, for detector data only; b) case-DS, for detector data and signal timings; and c) case-DSS, for detector data, signal timings and saturation flow rate. The performance of the model for different degrees of saturation and different detector detection intervals is consistent for case-DSS and case-DS whereas, for case-D the performance is inconsistent. The sensitivity analysis of the model for case-D indicates that it is sensitive to detection interval and signal timings within the interval. When detection interval is integral multiple of signal cycle then it has low accuracy and low reliability. Whereas, for detection interval around 1.5 times signal cycle both accuracy and reliability are high.

Estimation of travel time on urban networks with midlink sources and sinks

This paper presents a methodology for estimation of average travel time on signalized urban networks by integrating cumulative plots and probe data. This integration aims to reduce the relative deviations in the cumulative plots due to midlink sources and sinks. During undersaturated traffic conditions, the concept of a virtual probe is introduced, and therefore, accurate travel time can be obtained when a real probe is unavailable. For oversaturated traffic conditions, only one probe per travel time estimation interval—360 s or 3% of vehicles traversing the link as a probe—has the potential to provide accurate travel time.

Integration of a road traffic noise model (ASJ) and traffic simulation (AVENUE) for a built-up area

A road traffic noise prediction model (ASJ MODEL-1998) has been integrated with a road traffic simulator (AVENUE) to produce the Dynamic areawide Road traffic NoisE simulator-DRONE. This traffic-noise-GIS based integrated tool is upgraded to predict noise levels in built-up areas. The integration of traffic simulation with a noise model provides dynamic access to traffic flow characteristics and hence automated and detailed predictions of traffic noise. The prediction is not only on the spatial scale but also on temporal scale. The linkage with GIS gives a visual representation to noise pollution in the form of dynamic areawide traffic noise contour maps. The application of DRONE on a real world built-up area is also presented.

Voltage unbalance reduction in low voltage distribution networks with rooftop PVs

Voltage Unbalance (VU) is a power quality issue arising within the low voltage residential distribution networks due to the random location and rating of single-phase rooftop photovoltaic cells (PVs). In this paper, an analysis has been carried out to investigate how PV installations, their random location and power generation capacity can cause an increase in VU. Several efficient practical methods are discussed for VU reduction. Based on this analysis, it has been shown that the installation of a DSTATCOM can reduce VU. In this paper, the best possible location for DSTATCOM and its efficient control method to reduce VU will be presented. The results are verified through PSCAD/EMTDC and Monte Carlo simulations.

Securing IEEE 802.11 wireless LANs

As the acceptance and popularity of wireless networking technologies has proliferated, the security of the IEEE 802.11 wireless local area network (WLAN) has advanced in leaps and bounds. From tenuous beginnings, where the only safe way to deploy a WLAN was to assume it was hostile and employ higherlayer information security controls, to the current state of the art, all manner of improvements have been conceived and many implemented. This work investigates some of the remaining issues surrounding IEEE 802.11 WLAN operation. While the inherent issues in WLAN deployments and the problems of the original Wired Equivalent Privacy (WEP) provisions are well known and widely documented, there still exist a number of unresolved security issues. These include the security of management and control frames and the data link layer protocols themselves. This research introduces a novel proposal to enhance security at the link layer of IEEE 802.11 WLANs and then conducts detailed theoretical and empirical investigation and analysis of the eects of such proposals. This thesis �rst de�nes the state of the art in WLAN technology and deployment, including an overview of the current and emerging standards, the various threats, numerous vulnerabilities and current exploits. The IEEE 802.11i MAC security enhancements are discussed in detail, along with the likely outcomes of the IEEE 802.11 Task Group W1, looking into protected management frames. The problems of the remaining unprotected management frames, the unprotected control frames and the unprotected link layer headers are reviewed and a solution is hypothesised, to encrypt the entire MAC Protocol Data Unit (MPDU), including the MAC headers, not just the MAC Service Data Unit (MSDU) commonly performed by existing protocols. The proposal is not just to encrypt a copy of the headers while still using cleartext addresses to deliver the frame, as used by some existing protocols to support the integrity and authenticity of the headers, but to pass the entire MPDU only as ciphertext to also support the con�dentiality of the frame header information. This necessitates the decryption of every received frame using every available key before a station can determine if it is the intended recipient. As such, this raises serious concerns as to the viability of any such proposal due to the likely impact on throughput and scalability. The bulk of the research investigates the impacts of such proposals on the current WLAN protocols. Some possible variations to the proposal are also provided to enhance both utility and speed. The viability this proposal with respect to the eect on network throughput is then tested using a well known and respected network simulation tool, along with a number of analysis tools developed speci�cally for the data generated here. The simulator's operation is �rst validated against recognised test outputs, before a comprehensive set of control data is established, and then the proposal is tested and and compared against the controls. This detailed analysis of the various simulations should be of bene�t to other researchers who need to validate simulation results. The analysis of these tests indicate areas of immediate improvement and so the protocols are adjusted and a further series of experiments conducted. These �nal results are again analysed in detail and �nal appraisals provided.

Multiple camera management using wide baseline matching

Camera calibration information is required in order for multiple camera networks to deliver more than the sum of many single camera systems. Methods exist for manually calibrating cameras with high accuracy. Manually calibrating networks with many cameras is, however, time consuming, expensive and impractical for networks that undergo frequent change. For this reason, automatic calibration techniques have been vigorously researched in recent years. Fully automatic calibration methods depend on the ability to automatically find point correspondences between overlapping views. In typical camera networks, cameras are placed far apart to maximise coverage. This is referred to as a wide base-line scenario. Finding sufficient correspondences for camera calibration in wide base-line scenarios presents a significant challenge. This thesis focuses on developing more effective and efficient techniques for finding correspondences in uncalibrated, wide baseline, multiple-camera scenarios. The project consists of two major areas of work. The first is the development of more effective and efficient view covariant local feature extractors. The second area involves finding methods to extract scene information using the information contained in a limited set of matched affine features. Several novel affine adaptation techniques for salient features have been developed. A method is presented for efficiently computing the discrete scale space primal sketch of local image features. A scale selection method was implemented that makes use of the primal sketch. The primal sketch-based scale selection method has several advantages over the existing methods. It allows greater freedom in how the scale space is sampled, enables more accurate scale selection, is more effective at combining different functions for spatial position and scale selection, and leads to greater computational efficiency. Existing affine adaptation methods make use of the second moment matrix to estimate the local affine shape of local image features. In this thesis, it is shown that the Hessian matrix can be used in a similar way to estimate local feature shape. The Hessian matrix is effective for estimating the shape of blob-like structures, but is less effective for corner structures. It is simpler to compute than the second moment matrix, leading to a significant reduction in computational cost. A wide baseline dense correspondence extraction system, called WiDense, is presented in this thesis. It allows the extraction of large numbers of additional accurate correspondences, given only a few initial putative correspondences. It consists of the following algorithms: An affine region alignment algorithm that ensures accurate alignment between matched features; A method for extracting more matches in the vicinity of a matched pair of affine features, using the alignment information contained in the match; An algorithm for extracting large numbers of highly accurate point correspondences from an aligned pair of feature regions. Experiments show that the correspondences generated by the WiDense system improves the success rate of computing the epipolar geometry of very widely separated views. This new method is successful in many cases where the features produced by the best wide baseline matching algorithms are insufficient for computing the scene geometry.