Dr Len Hamey, Publications

Patents

L. G. C. Hamey, T. J. Watkins, and D. Wesche, "Pantograph damage and wear monitoring system," Australian provisional patent application 2007904219, QR Limited, 2007.

C. T. Westcott and L. G. C. Hamey, "Data recognition system," Patent application WO96/18975, Arnott's Biscuits Limited; also published as AU4111496 (1996), CA 2207326 (1996), GB 2311369 (1997), DE 19581867 (1997), CN 11704679 (1998), JP 10511786 (1998), NZ 296487 (2000), 1996. (Full Paper)

Abstract: The present invention is directed towards a method of determining the state of a substance (2) having a color property dependent on the state of the substance. The method includes the following steps: forming a pixel image of the substance (3) using a camera (3); projecting the pixel image into a three-dimensional color space using a neural network (7); and comparing the projection (37) with a projection of a second position of the substance (2) to determine the state of the substance (2) using a second neural network (9).

Journal Contributions

L. G. C. Hamey and S. N. Goldrei, "Implementing a domain-specific language using Stratego/XT: An experience paper," Electronic Notes in Theoretical Computer Science, vol. 203, no. 2, pp. 37-51, 2008. To appear in Electronic Notes in Theoretical Computer Science. (Full Paper)

Abstract: We describe the experience of implementing a Domain-Specific Language using transformation to a General Purpose Language. The domain of application is image processing and low-level computer vision. The transformation is accomplished using the Stratego/XT language transformation toolset. The implementation presented here is contrasted with the original implementation carried out many years ago using standard compiler implementation tools of the day. We highlight some of the unexpected advantages afforded to us, as language designers and implementers, by the source-to-source transformation technique. We also present some of the practical challenges faced in the implementation and show how these issues were addressed.

L. G. C. Hamey, "XOR has no local minima: A case study in neural network error surface analysis," Neural Networks, vol. 11, pp. 669-681, 1998. (Reprint available)

Abstract: This paper presents a case study of the analysis of local minima in feedforward neural networks. Firstly, a new methodology for analysis is presented, based upon considering trajectories through weight space by which a training algorithm might escape a hypothesized local minimum. This analysis method is then applied to the well known XOR (exclusive-or) problem, which has previously been considered to exhibit local minima. The analysis proves the absence of local minima, eliciting significant aspects of the structure of the error surface. The present work is important for the study of the existence of local minima in feedforward neural networks, and also for the development of training algorithms which avoid or escape entrapment in local minima.

T. RayChaudhuri and L. G. C. Hamey, "Active learning-approaches and issues," Journal of Intelligent Systems, vol. 7, pp. 205-243, 1997. (Reprint available)

Abstract: This paper surveys published work in active learning research with the purpose of providing a unified understanding of the area. A passive learning system relies entirely on pre-gathered information, whereas an active learning algorithm has the capability of interacting with its environment in order to collect information and/or to select learning policy. Active learning systems produce improved generalisation, reduce data costs and are most useful where data is expensive and computation is cheap. There are three major recognised approaches to the implementation of active learning-goal-driven learning, reinforcement learning and querying. While the first is largely a meta-level symbolic approach, the second is more a class of problems employing a policy-based approach to learning in non-deterministic dynamic environments; the third is based upon gathering the most useful examples by asking `intelligent' questions. Research in the area is mostly at a theoretical level yet.

T. RayChaudhuri, L. G. C. Hamey, and R. D. Bell, "From conventional control to autonomous intelligent methods," IEEE Control Systems, vol. 16, pp. 78-84, Oct. 1996.

Abstract: In this article we review the growth and development of control engineering, leading to modern adaptive methods and finally to autonomous intelligent control. Although the use of feedback control can be traced back to ancient and medieval times, it is really during the 20th century, with the evolution of the electronics age, that control engineering has become a recognized discipline. Well-established methods to model and control plants with linear characteristics and unchanging parameters are already in existence. Nonlinear plants with time-varying internal parameters are more challenging and the so-called ``adaptive'' methods have been developed to address this issue. The abundance of powerful computers has led us to think in terms of controllers that can ``learn'' by using AI techniques such as expert systems, genetic algorithms, neural networks, etc. These paradigms have evolved mostly from studying biological learning processes. ``Intelligent control'' and ``neurocontrol'' are terms that are recognized in the literature today as methods distinct from the more ``conventional'' control methods of the past few decades. Future advances in this science will be in the direction of the development of controllers that can learn to improve their performance and to plan while they learn in a particular task.

L. G. C. Hamey, "Comments on ``can backpropagation error surface not have local minima''," IEEE Transactions on Neural Networks, vol. 5, p. 844, Sept. 1994.

L. G. C. Hamey, J. A. Webb, and I.-C. Wu, "An architecture independent programming language for low-level vision," Computer Vision, Graphics and Image Processing, vol. 48, pp. 246-264, 1989.

Abstract: Low-level vision is particularly amenable to implementation on parallel architectures, which offers an enormous speedup at this level. To take advantage of this, the algorithm must be adapted to the particular parallel architecture. Having to adapt programs in this way poses a significant barrier to the vision programmer, who must learn and practice a different method of parallelization for each different parallel machine. There is also no possibility of portability for programs written for a particular parallel architecture. We have developed a specialized programming language, called Apply, which reduces the problem of writing the algorithm for this class of programs to the task of writing the function to be applied to a window surrounding a single pixel. Apply provides a method for programming these applications which is easy, consistent and efficient. Apply is programming model specific-it implements the input partitioning model-but is architecture independent. It is possible to implement versions of Apply which run efficiently on a wide variety of computers. We describe implementations of Apply on Warp, various Warp-like architectures, UNIX, and the Hughes HBA and sketch implementations on bit-serial processor arrays and distributed memory machines.

Books and Chapters in Books

L. G. C. Hamey, Computer Perception of Repetitive Textures. PhD thesis, Carnegie Mellon University, Pittsburgh, PA, Feb. 1988. Also available as Technical Report CMU-CS-88-149. (Full thesis: searchable scanned PDF 7.7MB)

L. G. C. Hamey, J. A. Webb, and I.-C. Wu, "Low-level vision on Warp and the Apply programming model," in Parallel Computation and Computers for Artificial Intelligence (J. Kowalik, ed.), ch. 10, pp. 185-199, Kluwer Academic Publishers, 1987.

Refereed Conference Papers

L. G. C. Hamey, T. Watkins, and S. Wong Too Yen, "Pancam: in-service inspection of locomotive pantographs," in Proceedings of Digital Image Computing: Techniques and Applications (M. J. Bottema, A. Maeder, N. Redding, and A. van der Hengel, eds.), (Glenelg, South Australia, Australia), pp. 493-499, IEEE Computer Society, ISBN 0 7695 3067 2, 2007. (Full Paper)

Abstract: Pancam is an inspection system for in-service monitoring of wear and damage to pantographs on electric locomotives. Damage and wear of pantographs can cause damage to the locomotive and overhead wiring with subsequent interruption of service. Pancam uses CCD cameras to capture side and top views of the pantographs of locomotives in normal service. The side view is analysed for wear and damage to the carbon current collectors while the top view is analysed for damage to the horns on the pantograph. Analysis results are reported to railway staff through a database with a thin-client Web interface. This paper presents an overview of the Pancam system and discusses some of the analysis techniques. Performance results are also presented.

L. G. C. Hamey, "Efficient image processing with the Apply language," in Proceedings of Digital Image Computing: Techniques and Applications (M. J. Bottema, A. Maeder, N. Redding, and A. van der Hengel, eds.), (Glenelg, South Australia, Australia), pp. 533-540, IEEE Computer Society, ISBN 0 7695 3067 2, 2007. (Full Paper)

Abstract: Apply is a Domain-Specific Language for image processing and low-level computer vision. Apply allows programmers to write kernel operations that focus on the computation for a single pixel location. The compiler generates code to perform the kernel computation over entire images. The original Apply implementation was developed 20 years ago for efficient processing on parallel architectures. The current-generation Apply compiler targets efficient code generation for general-purpose computers, typically outperforming handwritten code, while maintaining the simplicity of the original language. The use of modern compiler writing tools, specifically Stratego/XT, has facilitated improvements in the language design and made it easy to target the compiler to different environments. A large number of computer vision and image processing operations can be expressed in Apply. However, some algorithms require additional features. To motivate future language development, we analyse the requirements of the algorithms provided in a commercial machine vision library.

L. G. C. Hamey, "Simultaneous estimation of camera response function, target reflectance and irradiance values," in Proceedings of Digital Image Computing: Techniques and Applications (B. C. Lovell, A. J. Maeder, T. Caelli, and S. Ourselin, eds.), (Cairns, Australia), pp. 51-58, IEEE Computer Society, ISBN 0 7695 2467 2, 2005. (Full Paper, On-line conference proceedings)

Abstract: Video cameras may be used for radiometric measurement in machine vision applications including colour measurement and shape from shading. These techniques assume that the camera provides a linear measurement of light intensity. Even when the video gamma factor is disabled, video camera response is still often nonlinear and varies from one camera to another. We describe a method of measuring a camera.s radiometric response function using unknown but stable reflectance targets and illumination. The technique does not require sophisticated equipment or precise measurement or control of physical properties. The technique simultaneously estimates the camera's radiometric response, the relative reflectance values of reflectance targets and the relative illumination levels of multiple light sources. The technique has been practically applied for colour measurement applications with an experimentally verified accuracy of 0.3%

L. G. C. Hamey and C. Priest, "Automatic number plate recognition for Australian conditions," in Proceedings of Digital Image Computing: Techniques and Applications (B. C. Lovell, A. J. Maeder, T. Caelli, and S. Ourselin, eds.), (Cairns, Australia), pp. 87-94, IEEE Computer Society, ISBN 0 7695 2467 2, 2005. (Full Paper, On-line conference proceedings)

Abstract: We consider the task of recognition of Australian vehicle number plates (also called license plates or registration plates in other countries). A system for Australian number plate recognition must cope with wide variations in the appearance of the plates. Each state uses its own range of designs with font variations between the designs. There are special designs issued for significant events such as the Sydney 2000 Olympic Games. Also, vehicle owners may place the plates inside glass covered frames or use plates made of non-standard materials. These issues compound the complexity of automatic number plate recognition, making existing approaches inadequate. We have developed a system that incorporates a novel combination of image processing and artificial neural network technologies to successfully locate and read Australian vehicle number plates in digital images. Commercial application of the system is envisaged.

L. G. C. Hamey, "Teaching secure data communications using a game representation," in Proceedings of the Fifth Australasian Computing Education Conference (T. Greening and R. Lister, eds.), vol. 20 of Conferences in Research and Practice in Information Technology, (Adelaide, Australia), pp. 187-196, Australian Computer Society, ISBN 0-909925-98-4, 2003. (Full Paper, On-line conference proceedings)

Abstract: The Security Protocol Game is a highly visual and interactive game for teaching secure data communications protocols. Students use the game to simulate protocols and explore possible attacks against them. The power of the game lies in the representation of secret and public key cryptography. Specifically, the game provides representations for plain text and encrypted messages, message digests, digital signatures and cryptographic keys. Using these representations, students can construct public key certificates and perform multiple encryption, tunnelling and encrypted key transmission. They can simulate a wide range of protocols including authentication, key exchange and blind signature protocols. Application protocols such as Transport Layer Security and Pretty Good Privacy can be simulated in detail. The game clearly reveals the key issues of confidentiality, integrity, authentication and non-repudiation in secure data communications. Used as a small group learning activity, students gain a deep understanding of protocol design and operation issues. The game is suitable for use in tertiary and professional education courses for managers and information technology students at all levels.

L. G. C. Hamey, "Using the Security Protocol Game to teach computer network security," in Proceedings of Improving Learning Outcomes Through Flexible Science Teaching (K. Placing, ed.), (Sydney, Australia), pp. 96-101, Uniserve Science, 2003. (Full Paper, On-line symposium proceedings)

Abstract: The Security Protocol Game is a highly interactive game for teaching secure data communications protocols. Students use the game to simulate security protocols and explore possible attacks against them. The power of the game lies in the representation it provides for secret and public key cryptography - a unique combination of game rules and playing pieces has been devised that accurately represents the mathematical capabilities of cryptographic systems. Using pen and paper, envelopes and printed game pieces, students can simulate a wide range of computer network security protocols including well-known protocols such as SSL and Pretty Good Privacy. Such simulations enable students to gain a deep understanding of how the protocols operate and how protocol design affects security of the protocol. Student response to the game is positive and engaging. It has been successfully used with both information technology students and management students. This paper will include results from recently conducted student surveys.

L. G. C. Hamey, J. C.-H. Yeh, T. Westcott, and S. K. Y. Sung, "Pre-processing colour images with a self-organising map: Baking curve identification and bake image segmentation," in Proceedings of the 14th International Conference on Pattern Recognition, (Brisbane, Australia), pp. 1771-1775, Piscataway, NJ: IEEE, ISBN 0 8186 8512 3, 1998. (Full Paper)

Abstract: Kohonen's self-organising map is used to identify the colour development of baked goods from samples taken during baking. The resulting bake curves represent the colours characteristic of a particular baked product. Images of baked goods can be segmented and foreign bodies identified using these baking curves.

J. C.-H. Yeh, L. G. C. Hamey, and T. Westcott, "Developing FFNN applications using cross-validated validation training," in Proceedings of the 2nd IEEE International Conference on Intelligent Processing Systems, (Gold Coast, Australia), pp. 565-569, Piscataway, NJ: IEEE, ISBN 0 646 33229 5, 1998. (Full Paper)

Abstract: In this paper, we present a novel, effective and reliable training technique for feed-forward neural networks (FFNN). We call it cross-validated validation training (CVVT) since it combines statistical cross-validation with the validation training technique used in FFNNs. CVVT improves the generalisation estimation of validation training, enabling reliable comparison and selection of network architectures. Since it utilises validation training, CVVT also preserves the generalisation performance of FFNNs with excess weights. These benefits are demonstrated using statistical analysis of real-life results from a bake inspection system. Contrary to previous work, we found that significant excess weights may actually deteriorate the generalisation preserving ability of validation training.

J. Gibb and L. Hamey, "MINNI: micromouse incorporating neural network intelligence," in Proceedings of the Twentieth Australasian Computer Science Conference, (Sydney, Australia), pp. 194-201, Australian Computer Science Communications volume 19, number 1, ISSN 0157-3055, 1997. (Full Paper)

Abstract: MINNI is a system whereby a back propagation neural network is used to control the steering of a micromouse (small robot) in following a straight path. The neural network must be minimised to run on the hardware/software platform available on the Macquarie Micromouse. The development of the network follows intensive trials of algorithm and architecture variations in MATLAB. The implementation is programmed in ADA.

A. M. Pleasants and L. G. C. Hamey, "Photometric stereo using extended rectangular lights," in Proceedings of the First Joint Australia & New Zealand Biennial Conference on: Digital Image & Vision Computing - Techniques and Applications, (Auckland, New Zealand), pp. 533-538, Palmerston North, New Zealand: Dept. Production Technology, Massey University, ISBN 0 473 04947 3, 1997.

Abstract: This paper applies the photometric stereo technique for finding the shape and reflectance of an object using as illumination an extended rectangular light source. It derives the photometric stereo equations that are applicable and includes the geometry of the lights explicitly. The formulae are tested by finding the surface shape of a wedge with a known angle from four images. The results encourage confidence in the process.

L. G. C. Hamey, "Analysis of the error surface of the XOR network with two hidden nodes," in Proceedings of the Seventh Australian Conference on Neural Networks (P. Bartlett, A. Burkitt, and R. C. Williamson, eds.), (Canberra, Australia), pp. 179-183, The Australian National University, ISBN 0 7315 2429 2, Apr. 1996. (Full Paper)

Abstract: The exclusive-or learning task in a feed-forward neural network with two hidden nodes is investigated. Constraint equations are derived which fully describe the finite stationary points of the error surface. It is shown that the stationary points occur in a single connected union of eighteen manifolds. A Taylor series expansion is applied to the network error surface and it is shown that all points within the enumerated manifolds are arbitrarily close to points of lower error. It follows that the finite stationary points of the exclusive-or task are not relative minima. This result is surprising in view of the commonly held belief that the exclusive-or task exhibits local minima. The present result complements a recent result of the author's which proves the absence of regional local minima in the exclusive-or task.

L. G. C. Hamey, "Results on weight configurations that are not local minima in feed-forward neural networks," in Proceedings of the Seventh Australian Conference on Neural Networks (P. Bartlett, A. Burkitt, and R. C. Williamson, eds.), (Canberra, Australia), pp. 173-178, The Australian National University, ISBN 0 7315 2429 2, Apr. 1996. (Full Paper)

Abstract: Local minima in the error surfaces of feed-forward neural networks are significant because they may entrap gradient based training algorithms. Recent results have identified conditions under which local minima do not occur. The present paper considers three distinct definitions of local minimum, concluding that a new definition, called regional minimum, corresponds most closely to intuition. Using this definition, we analyse weight configurations in which a hidden node is ignored or redundant and show that these are not local minima. The practical implications of this result for gradient based learning are discussed.

T. RayChaudhuri and L. G. C. Hamey, "Active learning for nonlinear system identification and control," in Proceedings of IFAC World Congress 1996 (J. J. Gertler, J. B. Cruz, Jr, and M. Peshkin, eds.), vol. F, (San Fransisco), pp. 193-197, Pergamon, ISBN 0 08 042605 0, 1996. (Full Paper)

Abstract: The identification and control of nonlinear systems continues to remain a challenging issue. Neural network models and controllers have often been effective in addressing the situation. However current neural network learning methods have been found to be limited in their generalisation abilities. Recent research has shown active learning methods to be effective in increasing the modelling reliability of a neural network system. An active learning agent has the ability to query its environment in order to make a selection of its training data. One approach to the implementation of active leaning is to use `querying-by-committee'. This results in considerably reduced data collection and at the same time does not compromise the accuracy of identification. A nonlinear plant with both clean and noisy data is successfully modelled by such a technique and a feedforward neural network controller based upon such a model is demonstrated to perform effectively.

T. RayChaudhuri and L. G. C. Hamey, "Accurate modelling with minimised data collection-an active learning algorithm," in Proceedings of the Seventh Australian Conference on Neural Networks (P. Bartlett, A. Burkitt, and R. C. Williamson, eds.), (Canberra, Australia), pp. 11-15, The Australian National University, ISBN 0 7315 2429 2, Apr. 1996. (Full Paper)

Abstract: A data gathering method based on active querying is described. In this method data is reduced to a minimum, yet modelling accuracy is uncompromised. Our active querying criterion is determined by whether or not several neural network models agree when they are fitted to random subsamples of a small amount of collected data. Experiments have established the feasibility of our algorithm. It is also shown that our approach results in a more samples being collected in the neighbourhood of the more significant inputs.

L. G. C. Hamey, "The structure of neural network error surfaces," in Proceedings of the Sixth Australian Conference on Neural Networks (M. Charles and C. Latimer, eds.), (Sydney, Australia), pp. 197-200, Dept. of Electrical Engineering, University of Sydney, ISBN 0 909391 03 3, Feb. 1995. (Full Paper)

Abstract: Analysis of the error surfaces of feed-forward neural networks is complicated by the high dimensionality of the weight space. Visualisation over one- and two-dimensional slices, and Monte Carlo analysis of stationary points can produce misleading results. We show that, in some situations, important features of the error surface can only be visualised by considering the error over non-planar manifolds of weight space. We also show that Monte Carlo simulations can depend critically upon the random step size chosen. The relationship can reveal key properties of the local structure of the error surface.

T. RayChaudhuri, J. C.-H. Yeh, L. G. C. Hamey, and C. T. Westcott, "Baked product classification with the use of a self-organising map," in Proceedings of the Sixth Australian Conference on Neural Networks (M. Charles and C. Latimer, eds.), (Sydney, Australia), pp. 152-155, Dept. of Electrical Engineering, University of Sydney, ISBN 0 909391 03 3, Feb. 1995. (Full Paper)

Abstract: Study of the baking of biscuits involves among other aspects detailed analysis of colour changes in the product during the process. Previous study has shown the existence of a colour development curve (known as the baking curve) by examining colour development in the RGB and HSI colour spaces. In the current work a different approach to extracting the baking curve is presented. Using a Kohonen self-organising map with an optimum number of output nodes a well-defined baking curve is automatically extracted from preprocessed data of images gathered during the actual baking process. We propose that these curves can be used as a basis for characterising the colour bake level of a biscuit.

T. RayChaudhuri and L. G. C. Hamey, "Minimisation of data collection by active learning," in Proceedings of the IEEE International Conference on Neural Networks, vol. 3, (Perth, Australia), pp. 1338-1341, Piscataway, NJ: IEEE, ISBN 0 7803 2768 3, Nov. 1995. (Full Paper)

Abstract: We use the `query-by-committee' approach for building an active scheme for data collection. In this method data gathering is reduced to a minimum, yet modelling accuracy is uncompromised. Our active querying criterion is determined by whether or not several models agree when they are fitted to random subsamples of a small amount of collected data. Experiments with neural network models to establish the feasibility of our algorithm have produced encouraging results.

T. RayChaudhuri, L. G. C. Hamey, and R. D. Bell, "Neural network control using active learning," in Control 95, vol. 2, pp. 369-373, Barton, ACT, Australia: Inst. of Engineers, ISBN 0 85825 631 2, Oct. 1995. (Full Paper)

Abstract: Neural networks have been shown to give considerably better results when controlling complex non-linear systems than conventional control methods. Most neural network controllers today are built around `passive' learning methods whereby the network once trained is expected to perform repeatedly with equal accuracy on fresh sets of input-output data. This is not always suitable in real world situations where external environmental parameter variations cause changes in the plant and controller performance. In the current paper we propose the use of an autonomous `active' learning technique which will cause training to re-occur precisely when these parameter variations happen, yielding enhanced controller performance.

T. RayChaudhuri, J. C.-H. Yeh, L. G. C. Hamey, S. K. Y. Sung, and T. Westcott, "A connectionist approach to quality assessment of food products," in Proceedings of the Eighth Australian Joint Conference on Artificial Intelligence (X. Yao, ed.), (Canberra, Australia), pp. 435-441, Singapore: World Scientific, ISBN 981 02 2484 2, Nov. 1995. (Full Paper)

Abstract: Colour development of a product is often vital in the food industry. The study of the baking of biscuits reveals interesting colour development characteristic curves. Neural network methods are used to both represent and classify products according to these characteristics. Using self-organising maps well-defined characteristic curves are extracted. Colour data histogrammed along these curves are then accurately classified by feedforward neural networks trained by backpropagation. Image segmentation is implicit within this colour histogramming technique. The overall system has been shown to considerably outperform a human expert.

J. C.-H. Yeh, L. G. C. Hamey, C. T. Westcott, and S. K. Y. Sung, "Colour bake inspection system using hybrid artificial neural networks," in Proceedings of the IEEE International Conference on Neural Networks, vol. 1, (Perth, Australia), pp. 37-42, Piscataway, NJ: IEEE, ISBN 0 7803 2768 3, Nov. 1995. (Full Paper)

Abstract: The bake level of biscuits is of significant value to biscuit manufacturers as it determines the taste, texture and appearance of the products. Previous research explored and revealed the feasibility of biscuit bake inspection using feed forward neural networks (FFNN) with a back propagation learning algorithm and monochrome images. A second study revealed the existence of a curve in colour space, called a baking curve, along which the bake colour changes during the baking process. Combining these results, we proposed an automated bake inspection system with artificial neural networks that utilises colour instead of monochrome images. In this paper, we present the implementation of the inspection system with a hybrid neural network of self-organising maps and FFNNs. The system was tested and its grading performance on biscuit bake levels was evaluated and compared to that of a trained human inspector. We found that the proposed colour system with a hybrid neural network performed significantly better than the human.

J. C.-H. Yeh and L. G. C. Hamey, "Biscuit bake assessment by an artificial neural network," in Proceedings of the Fifth Australian Conference on Neural Networks (A. C. Tsoi and T. Downs, eds.), (Brisbane, Australia), pp. 266-269, Dept. of Electrical and Computer Engineering, University of Queensland, Feb. 1994. (Full Paper)

Abstract: A prototype artificial neural network system for assessing the bake level of biscuits has been implemented. We present performance results and compare the neural network approach with a statistical method and the performance of the trained inspector. The neural network system performs comparably with the other methods.

L. G. C. Hamey, A. J. Watson, and C. T. Westcott, "Machine inspection of biscuit bake," in Proceedings of Digital Image Computing: Techniques and Applications (K. K. Fung and A. Ginige, eds.), (Sydney, Australia), pp. 124-129, Australian Pattern Recognition Society, ISBN 0 646 16522 4, Dec. 1993. (Full Paper)

Abstract: A prototype system for automated assessment of biscuit bake has been developed. The system employs monochrome imaging and histogramming techniques to classify product samples as underbaked, correctly baked or overbaked. Two products have been investigated. Product ``A'' exhibits uneven browing due to the formation of blisters. An intensity histogram suitably characterises the overall browning and the prototype system classifies samples of product ``A'' with an error rate comparable to that of a trained inspector. Product ``B'' browns most heavily on the perimeter on the biscuit. A boundary-distance histogram is used to classify samples of product ``B''. In this case also, the system performance is comparable with a trained inspector.

L. G. C. Hamey, "Benchmarking feed-forward neural networks: Models and measures," in Advances in Neural Information Processing Systems 4 (J. E. Moody, S. J. Hanson, and R. P. Lippmann, eds.), pp. 1167-1174, San Mateo, CA: Morgan Kaufmann, 1992. (Full Paper)

Abstract: Existing metrics for the learning performance of feed-forward neural networks do not provide a satisfactory basis for comparison because the choice of the training epoch limit can determine the results of the comparison. I propose new metrics which have the desirable property of being independent of the training epoch limit. The efficiency measures the yield of correct networks in proportion to the training effort expended. The optimal epoch limit provides the greatest efficiency. The learning performance is modelled statistically, and asymptotic performance is estimated. Implementation details may be found in (Hamey, 1992).

M. Annaratone, F. Bitz, J. Deutch, L. Hamey, H. Kung, P. Maulik, P. Tsend, and J. Webb, "Applications experience on Warp," in Proceedings of the National Computer Conference, (AFIPS, Chicago, Il), pp. 149-158, June 1987.

Unrefereed Conference Papers

L. G. C. Hamey, "A simulation game for teaching secure data communications protocols," in And Gladly Teche: Celebrating Teaching at Macquarie (A. Reid, M. Gosper, and S. Fraser, eds.), (Macquarie University, Australia), The Centre for Professional Development and the Centre for FLexible Learning, Macquarie University, Australia, ISBN 1-86408-793-5, 2002. (Full Paper, On-line conference proceedings)

Abstract: With the widespread commercial use of the Internet, secure data communications over the Internet has become an important aspect of business operations. Thus, it is an important study for information technology and management students. The Security Protocol Game is an interactive group activity for exploring secure data communication protocols. Using pen and paper, envelopes and game tokens, students simulate security protocols and possible attacks against them. The game provides simple and intuitive representations for cryptographic methods, including both public key and secret key techniques. Using these representations, students can simulate Internet application protocols such as Pretty Good Privacy (used to secure email) and Transport Layer Security (used for secure web transactions). They can explore well-known protocols for authentication, key exchange and blind signatures. Students can also develop and test their own protocols using public key certificates, encrypted key transmission, tunnelling and other well-known techniques. Through this learning activity, students gain a deep understanding of how security protocols operate and are designed. The game has been used in tertiary units of study for managers and information technology students.

L. G. C. Hamey, J. C.-H. Yeh, and C. Ng, "Objective bake assessment using image analysis and artificial intelligence," in Cereals '97: Proceedings of the 47th Australian Cereal Chemistry Conference, (Perth, Australia), pp. 180-184, North Melbourne, Australia: Royal Australian Chemical Institute, ISBN 0909589941, 1997. (Full Paper)

Abstract: Bake assessment in food manufacture is currently performed by trained human inspectors. The subjective nature of human assessment introduces short-term variation and long-term drift of bake standards due to inconsistencies in human performance. A durable, repeatable and transferable assessment method is desirable to ensure long-term consistency of product bake and to facilitate product migration between manufacturing sites. These goals are successfully met by employing computer analysis of colour digital images to assess product bake. Our studies have shown the existence of a baking curve, unique to each product, which describes the colour development of that product during baking. Based upon this discovery, a system has been developed employing Artificial Intelligence techniques, specifically Artificial Neural Networks, to match computer bake assessment to one or more expert bake assessors. The process of training the system for a particular product is driven by these assessors who select and classify product samples. By this means, the system can be deployed across the product range with minimal involvement of image analysis and computing technologists. The system has been shown to provide significant performance improvements in the bake quality assessment task.

L. Hamey, J. Yeh, and C. Ng, "Machine inspection system for bake colour quality," in Process Control and Optimisation Seminar, (Sydney, Australia), Cooperative Research Centre for International Food Manufacture and Packaging Science, Nov. 1997. (unpublished).

L. G. C. Hamey and J. C.-H. Yeh, "Segmentation of bake images by a self-organising map," in Image Segmentation Workshop, (Sydney, Australia), pp. 65-68, Australian Pattern Recognition Society, Dec. 1996. (Full Paper)

Abstract: A technique for segmentation of images of baked good is presented. The technique employs a Self-Organising Map to identify the characteristic colour development curve (bake curve) for each product. Segmentation is based upon the colour information contained in the bake curve. The technique is trained with only positive exemplars of the product.

L. G. C. Hamey, "XOR has no local minima," in NIPS*96 Workshop: Modeling Error Surfaces, (Denver, Co.), 1996. (invited presentation, unpublished). (Slide Show)

T. RayChaudhuri and L. G. C. Hamey, "An algorithm for active data collection applied to neural network training," in AAAI Fall Symposium on Active Learning, (MIT, Boston, USA), Nov. 1995. (unpublished).

L. G. C. Hamey and T. Kanade, "Computer analysis of regular repetitive textures," in Proceedings of the Image Understanding Workshop, pp. 1076-1088, 1989.

Theses Supervised

T. RayChaudhuri, Seeking the Valuable Domain - Query Learning in a Cost-Optimal Perspective. PhD thesis, Macquarie University, Computing Department, 1997. (Full Paper)

Abstract: This thesis is intended as a contribution to the theories and principles of active learning and dual control. It is an account of an investigation which essentially examines the problem of how a machine learning system can be designed so as to acquire useful data actively, i.e., with environment interaction and simultaneously perform cost-effectively. The primary aim of the research has been to evolve the general basis for developing an engineering solution for a self-designing system of the future. In other words, while developed theory and experimental investigations to substantiate the theory are presented in this dissertation, the learning system used herein is based on certain specific assumptions with a view to achieving a balance between extremely general theoretical abstraction and a practical method that addresses clearly-defined problem parameters. The outcome is a strategy of `learning while performing' - consisting of a set of algorithms directed to addressing a real world scenario such as an industrial manufacturing plant from which continuous data is available.

J. C.-H. Yeh, "Colour bake inspection using artificial neural networks," MSc(hons) thesis, Macquarie University, Computing Department, 1997.

J. Gibb, "Generalisation and performance of a back propagation network applied to a small robot," MSc(hons) thesis, Macquarie University, Computing Department, 1996.

C. G. Perrott, "Perceptual grouping for recognition of deformable objects," MSc(hons) thesis, Macquarie University, Computing Department, 1995.

Honours Theses Supervised

A. Ruckholdt, "Plant system emulation with on-line learning," BSc(hons) thesis, Macquarie University, Computing Department, 1995.

S. K. Y. Sung, "A study of baking curve," BSc(hons) thesis, Macquarie University, Computing Department, 1993.

J. C.-H. Yeh, "Baking inspection using artificial neural networks," BSc(hons) thesis, Macquarie University, Computing Department, 1993.

P. A. Bayliss, "Artificial neural networks for edge detection in computer vision," BSc(hons) thesis, Macquarie University, Computing Department, 1992.

W. Edwards, "An object-oriented approach to neural network simulators," BSc(hons) thesis, Macquarie University, Computing Department, 1992.

M. L. Allsop, "Applications of functional methods to image processing," BSc(hons) thesis, Macquarie University, Computing Department, 1991.

J. Gibb, "Benchtesting of backpropagation learning for neural networks," BSc(hons) thesis, Macquarie University, Computing Department, 1991.

Technical Reports

L. G. C. Hamey and S. N. Goldrei, "Implementing the Apply compiler using Stratego/XT," Tech. Rep. C/TR07-01, Department of Computing, Macquarie University, NSW 2109 Australia, Feb. 2007. (Full Paper)

Abstract: We describe the experience of implementing a Domain-Specific Language using transformation to a General Purpose Language. The domain of application is image processing and low-level computer vision. The transformation is accomplished using the Stratego/XT language transformation toolset. The implementation presented here is contrasted with the original implementation carried out many years ago using standard compiler implementation tools of the day. We highlight some of the unexpected advantages afforded to us, as language designers and implementers, by the source-to-source transformation technique. We also present some of the practical challenges faced in the implementation and show how these issues were addressed.

D. B. Hayward and L. Hamey, Araucaria Colour Matching. Forest and Wood Products Research and Development Corporation, Report on project PN01.1901, 2005. (Full Paper)

T. RayChaudhuri and L. G. C. Hamey, "Cost effective querying leading to dual control," Tech. Rep. C/TR96-07, Department of Computing, Macquarie University, NSW 2109 Australia, May 1996. (Full Paper)

Abstract: We propose a practical querying method that incorporates the notion of output dollar value. The approach leads to a method of dual control that uses querying methods directed towards cost optimisation. Employing the basic tenets of the query-by-committee philosophy of Seung and Freund, a novel jack-knifed approach for estimating the distribution of a label, is advocated. We define an expected value function of a label as an exploitation objective - in a cost- benefit analysis perspective. A confidence interval size on this expected value determines the degree of exploration. Experimental results upon simulated data are presented.

L. G. C. Hamey, "Analysis of the error surface of the XOR network with two hidden nodes," Tech. Rep. 95-167C, Department of Computing, Macquarie University, NSW 2109 Australia, Feb. 1995. (Full Paper)

Abstract: The exclusive-or learning task in a feed-forward neural network with two hidden nodes is investigated. Constraint equations are derived which fully describe the finite stationary points of the error surface. It is shown that the stationary points occur in a single connected union of eighteen manifolds. A Taylor series expansion is applied to the network error surface and it is shown that all points within the enumerated manifolds are arbitrarily close to points of lower error. It follows that the finite stationary points of the exclusive-or task are not relative minima. This result is surprising in view of the commonly held belief that the exclusive-or task exhibits local minima. The present result complements a recent result of the author's which proves the absence of regional local minima in the exclusive-or task.

J. Gibb and L. Hamey, "A comparison of back propagation implementations," Tech. Rep. C/TR95-06, Department of Computing, Macquarie University, NSW 2109 Australia, Sept. 1995. (Full Paper)

Abstract: Back propagation training algorithms have been implemented by many researchers for their own purposes and provided publicly on the internet for others to use in verification of published results and for reuse in unrelated research projects. Often, the source code of a package is used as the basis for a new package for demonstrating new algorithm variations, or some functionality is added specifically for analysis of results. However, there are rarely any guarantees that the original implementation is faithful to the algorithm it represents, or that its code is bug free or accurate. This report attempts to look at a few implementations and provide a test suite which shows deficienciesin some software available which the average researcher may not be aware of, and may not have the time to discover on their own. This test suite may then be used to test the correctness of new packages.

T. RayChaudhuri and L. G. C. Hamey, "From conventional control to intelligent neurocontrol methods-a survey of the literature," Tech. Rep. 95-170C, Department of Computing, Macquarie University, NSW 2109 Australia, Mar. 1995. (Full Paper)

Abstract: We present here a survey of published work covering the growth and development of control engineering, leading to modern adaptive methods and finally to intelligent neurocontrol. Ever since the advent of civilization, man has constantly attempted to replace human effort with machines and control systems. From the early days of steam, hydraulic and pneumatically driven machinery to the modern era of electrical and electronic devices, the level of control achieved over industrial systems has constantly increased. With the availability of fast, high-capacity digital computer equipment, automatic numerically-controlled industrial systems are common today. Well-established methods to model and control plants with linear characteristics and unchanging parameters are already in existence. Nonlinear plants with time-varying internal parameters are more challenging and the so-called `adaptive' methods have been developed to address this issue. The abundance of powerful computers has led us to think in terms of controllers that can `learn' by using neural networks - computing paradigms that have evolved from studying the functioning of the human brain. `Intelligent Control' and `Neurocontrol' are terms that are recognised in the literature today as methods distinct from the more `conventional' control methods of the past few decades. Future advances in this science will be in the direction of the development of controllers that can learn to improve their performance and to plan while they learn a particular task.

T. RayChaudhuri and L. G. C. Hamey, "An algorithm for active data collection for learning-feasibility study with neural networks," Tech. Rep. 95-173C, Department of Computing, Macquarie University, NSW 2109 Australia, May 1995. (Full Paper)

Abstract: Statisticians have considered query-based or `active' sampling of data as a means of reducing the expense of data measurement and collection in modelling tasks. The quest for more reliable neural network learning techniques has led researchers to examine statistical active querying as a means of obtaining training data that will produce greatly improved generalisation. Some algorithms evolved for active learning in neural networks perform active data subset selection. We propose using the `query-by-committee' approach. This leads to an active scheme for data collection where data gathering is reduced to a minimum and yet the accuracy of modelling remains high. Our method is built around the philosophy that `data gathering is expensive and computation is cheap'. Our active querying criterion is determined by whether or not several models agree when they are fitted to random subsamples of a small amount of collected data. Recent experimental investigations have established the effectiveness of this algorithm for both clean and noisy data so far as neural network learning is concerned.

L. G. C. Hamey, "On human perception of regular repetitive textures," Tech. Rep. 92-94C, Department of Computing, Macquarie University, NSW 2109 Australia, 1992. (Full Paper)

Abstract: Regular repetitive textures occur frequently in both natural and man-made scenes. The human visual system is highly effective at extracting the texture element and structure of such textures. Surprisingly, even when presented with random repetitive textures, human observers are able to immediately identify the repetitive nature of the texture and to describe a perceived structural unit. This paper presents a qualitative exploration of some of the capabilities of the human visual system in relation to the perception of regular repetitive textures. In particular, some limitations of the perception of regularity are demonstrated. I hypothesize that the perception of regularity is mediated by perceptual organisation of dominant primitive features or dominant meta-features (perceptual groupings of features). This hypothesis suggests that the perception of regular structure can be prevented by either suppressing the extraction of the primitive features, or by suppressing the dominance of a single feature or meta-feature. The suppression of feature extraction is difficult, and results obtained to date are only partially successful. The suppression of dominance, on the other hand, has already been achieved in commercial textile design. The surprising result of this paper is that the suppression of dominance can be effectively achieved by arranging only four pattern units in the texture element in such a way as to create competing perceptual groupings between the pattern units. The resulting four-dot texture produces the visual impression of a field of dots rather than a regular repetitive pattern. Modifications to the pattern that create a dominant primitive feature or dominant meta-features are shown to mediate the perception of regularity. It is expected that these patterns will aid in developing an understanding of the processes underlying human perception.

L. G. C. Hamey, "Efficient image processing on RISC workstations," Tech. Rep. 92-113C, Department of Computing, Macquarie University, NSW 2109 Australia, 1992. (Full Paper)

Abstract: The use of Reduced Instruction Set Computers for scientific applications is common today. Advances in processor design and compiler technology make it possible to perform large-scale computations on RISC workstations. The RISC design provides simple instructions that operate at high speed and compiler optimisations employ the machine's capabilities to perform naively programmed operations with reasonable efficiency. However, there remain opportunities for improvement in execution performance by optimising the design of the high-level language program, and, conversely, there are also traps for the unwary programmer. Partial unrolling of loops and employing scalar variables to assist the compiler in the use of registers are two simple techniques that can yield performance improvements of up to 50%depending upon the compiler and the architecture. Unexpected performance penalties as high as 500%may be paid for cache conflicts, where two large data arrays are contending for the same cache locations. This report discusses how to improve the performance of critical code sections in scientific programs, and how to detect and avoid cache conflict performance penalties.

C. G. Perrott and L. G. C. Hamey, "Object recognition-a survey of the literature," Tech. Rep. 91-65C, Department of Computing, Macquarie University, NSW 2109 Australia, Jan. 1991. (Full Paper)

Abstract: This paper surveys the techniques which have been applied to the problem of recognising three-dimensional objects in two-dimensional images. Human vision was discussed in the works of the ancient Greek philosophers, and has also been of interest to modern philosophers. The Gestalt school of psychology in the early part of the twentieth century provided a number of useful insights into human perception.

Computer vision research effectively started with the pioneering work of Roberts, who built a program capable of recognising simple objects in a blocks world. The blocks world paradigm provides a simplified model in which new approaches can be tested, and has been adopted from time to time by a number of researchers.

The dominant paradigm in modern computer vision research is that pioneered by Marr, and known as inverse optics or the Marr paradigm. In this approach, edges, surfaces and depth cues are identified before object recognition is attempted. Central problems in much of this work are edge detection and region segmentation, which have proved to be more difficult than was anticipated by early researchers. The results achieved up till now suggest that it may not be possible to perform a perfect segmentation of the image before proceeding to higher level processing.

Recently some researchers have investigated the use of cues from perceptual organisation in order to perform object recognition without using complete depth information. The perceptual organisation approach promises to reduce the amount of computation that has to be performed. This would be highly desirable since it is widely believed that a practical computer vision system for processing natural scenes would require many Gflops of processing power.

L. G. C. Hamey, J. A. Webb, and I.-C. Wu, "Low-level vision on Warp and the Apply programming model," Tech. Rep. CMU-RI-TR-87-17, Robotics Institute, Carnegie Mellon University, 1987. (Full Paper)

Abstract: In the course of implementing low-level (image to image) vision algorithms on Warp, we have understood the mapping of this class of algorithms well enough so that the programming of these algorithms is now a straightforward and stereotypical task. The partitioning method used is input partitioning, which provides an efficient, natural implementation of this class of algorithms. We have developed a special programming language called Apply, which reduces the problem of writing the algorithm for this class of programs to the task of writing the function to be applied to a window around a single pixel. Apply provides a method for programming Warp in these applications which is easy, consistent, and efficient. Apply is application specific, but machine independent - it is possible to implement versions of Apply which run efficiently on a wide variety of computers. We describe implementations of Apply on Warp, UNIX and the Hughes HBA, and sketch implementation on bit-serial processor arrays and distributed memory machines.

Copyright 2009 Dr Len Hamey