Papers by Lambert Spaanenburg
Cloud Connectivity and Embedded Sensory Systems, 2010
Proceedings …, 2004
The paper presents the architecture and building blocks for a digital CNN, extending the previous... more The paper presents the architecture and building blocks for a digital CNN, extending the previous ILVA design with packet switched communication and an OSI-compliant functional structure. Locality of operation results in a generic VHDL description that can be easily ...
This paper presents the design of a neurocontroller for solving complex control problems. The non... more This paper presents the design of a neurocontroller for solving complex control problems. The non-linear system dynamics of the vehicle and its tangled parameterization make the controller design for autonomous lateral vehicle guidance by conventional system- theoretical methods difficult. Furthermore, experts have to adjust this type of controller for each new kind of vehicle. Neural networks provide an easy solution, that can be learned from measured human driving data without knowledge of physical details of the vehicle. Simulations have shown that already small-sized neural networks are able to approximate the human driving behaviour. Practical tests with the Daimler Benz autonomous vehicle OSCAR (Optical Steered CAR) have been successfully performed on public highways up to a speed of 130 km/h. (A) For the covering abstract see IRRD 866746.
| Basic arguments are described to create a uniied CoDesign environment. A unique feature is the ... more | Basic arguments are described to create a uniied CoDesign environment. A unique feature is the use of a single language to describe systems. Diierent subsets describe hardware as well a software, both structure and behavior. It vividly shows software and hardware are much more alike than usually perceived. Borders between diier-ent descriptions now become artiicial, meaning we can walk much more freely in the design space. Also our tools beneet from the uniication, we need fewer of them and they are inherently more general. Recent developments in functional languages, integrating functional and imperative languages, made this approach practical for the rst time. The tool we are developing is intended to stimulate the integration of software and hardware design tools into genuine CoDesign tools. The use of a modern typed functional language such as Haskell also makes the language into a broad spectrum CoDe-sign language. Keywords| codesign, design re-use, functional imperative programming.
International Neural Network Conference, 1990
A neural network learns from a long history of applied patterns. To create fault-tolerance, an ex... more A neural network learns from a long history of applied patterns. To create fault-tolerance, an extended learning period is required. For actual hardware such an initialization period is too long. It is investigated how under the bounded delay assumption training of the neural network can be accelerated. It is concluded that the optimal training set takes not only the smallest number of learning cycles but is also delay-insensitive and therefore allows a large speed-up using the non-normal mode of operation.
Lecture Notes in Computer Science, 2009
Modeling of human and animal behavior is of interest for a number of diagnostic purposes. Convolu... more Modeling of human and animal behavior is of interest for a number of diagnostic purposes. Convolutional neural networks offer a constructive approach allowing learning on a limited number of examples. Chaotic tendencies make that learning is not always successful. The paper looks into a number of applications to find the reason for this anomaly and identifies the need for behavioral references to provide determinism in the diagnostic model.
Applications, Concepts and Technologies
In the early days of photography, camera movement was a nuisance that could blur a picture. Once ... more In the early days of photography, camera movement was a nuisance that could blur a picture. Once movement becomes measurable by micro-mechanical means, the effects can be compensated by optical, mechanical or digital technology to enhance picture quality. Alternatively movement can be quantified by processing image streams. This opens up for new functionality upon convergence of the camera and the mobile phone, for instance by ’actively extending the hand’ for remote control and interactive signage.
Image processing is one of the popular applications of Cellular Neural Networks. Macro enriched f... more Image processing is one of the popular applications of Cellular Neural Networks. Macro enriched field-programmable gate-arrays can be used to realize such systems on silicon. The paper discusses a pipelined implementation that supports the handling of gray-level images at 180 to 240 Mpixels per second by exploiting the Virtex-II macros to spatially unroll the local feedback.
Cloud Connectivity and Embedded Sensory Systems, 2010
Cloud Connectivity and Embedded Sensory Systems, 2010
The typical embedded system design starts with listing the requirements. But there will be little... more The typical embedded system design starts with listing the requirements. But there will be little to go on, when the system is to be a loosely coupled network on which a number of functionalities are to be created. First, the design space has to be explored to find the practical operating limits despite the fact that not everything is clearly defined at the beginning. Applying wisdom is one approach to solve the dilemma, but a more structured development scheme is advisable.
Journal of the Institution of Electronic and Radio Engineers
A methodology is presented for the hierarchical, constructionally correct specification of digita... more A methodology is presented for the hierarchical, constructionally correct specification of digital controllers. It allows easy adaptation to architectural changes and straightforward preparation of documentation. Means are discussed for the rapid realization of the state diagram specification with efficient use of silicon, thus paving the way to general-purpose silicon compilation.
Modularity and hierarchy are fundamental notions in structured system design. By subdividing a la... more Modularity and hierarchy are fundamental notions in structured system design. By subdividing a large and unstructured problem into smaller and tractable chunks, design automation becomes possible. In this paper we discuss the use of modularity and hierarchy for functional specialization during the development of neural networks. We study the behavioral differences and requirements for back- propagation training of feed-forward networks. Further we illustrate that a deliberate mix of hierarchically imposed evaluation functions will improve network accuracy and learning speed.
2006 10th International Workshop on Cellular Neural Networks and Their Applications, 2006
Ieee Transactions on Circuits and Systems Ii Analog and Digital Signal Processing, 2002
Despite their success-story, artificial neural networks have one major disadvantage compared to o... more Despite their success-story, artificial neural networks have one major disadvantage compared to other techniques: the inability to explain comprehensively how a trained neural network reaches its output; neural networks are not only (incorrectly) seen as a "magic tool" but possibly even more as a mysterious "black box." Although much research has already been done to "open the box," there is a notable hiatus in known publications on analysis of neural networks. So far, mainly sensitivity analysis and rule extraction methods have been used to analyze neural networks. However, these can only be applied in a limited subset of the problem domains where neural network solutions are encountered. In this paper we propose a wider applicable method which, for a given problem domain, involves identifying basic functions with which users in that domain are already familiar, and describing trained neural networks, or parts thereof, in terms of those basic functions. This will provide a comprehensible description of the neural network's function and, depending on the chosen base functions, it may also provide an insight into the neural network's inner "reasoning." It could further be used to optimize neural network systems. An analysis in terms of base functions may even make clear how to (re)construct a superior system using those base functions, thus using the neural network as a construction advisor.
Phys Rev Lett, 2002
This paper illustrates a novel method to analyze artificial neural networks so as to gain insight... more This paper illustrates a novel method to analyze artificial neural networks so as to gain insight into their internal functionality. To this purpose, the elements of a feedforward-backpropagation neural network, that has been trained to detect edges in images, are described in terms of differential operators of various orders and with various angles of operation.
Energy For Sustainable Development, Nov 26, 2003
... It takes a winner to take his share Suleyman Malki1, Lambert Spaanenburg1 and Berend-Jan van ... more ... It takes a winner to take his share Suleyman Malki1, Lambert Spaanenburg1 and Berend-Jan van der Zwaag2 Lund University / LTH Twente University / EWI ... 698 705. [15] RS Venema and L. Spaanenburg, Learning feed-forward multi-nets, ICANNGA'01 (Prague, 2001) pp. ...
Notes Sais Ssls Joint Workshop, 2003
The proper handling of alarms is crucial to automated process control. In practice, many alarms a... more The proper handling of alarms is crucial to automated process control. In practice, many alarms are only distractive and do not represent a fault situation. Here a toolbox, that uses an expert system called LARA, is proposed that aims to reduce the occurrence of so-called nuisance alarms. It is based on a signal classification scheme to propose suitable alarm handling algorithms according to the signals behavior. In a first prototype, a significant reduction of the alarms in a biofueled District Heating Plant has been achieved.
Journal of Intelligent Fuzzy Systems Applications in Engineering and Technology, 2004
Multilevel flow modeling (MFM) is a modeling method for complex technical systems in which the go... more Multilevel flow modeling (MFM) is a modeling method for complex technical systems in which the goals and functions of the system are explicitly described. MFM can be used as a basis for root cause analysis, where primary root causes are separated from consequential faults, in complex fault situations. Model representations for use in diagnostic reasoning usually describe causality, between parameters, faults, or process states. However, the causality of a system may vary depending on details in the construction, as well as over time with the process state. One contribution of this paper is a general method of describing varying causality in a simple and efficient way. The method has been tested using multilevel flow models. Causality is visible in measurements and can be used to increase process understanding. The standard cross - correlation technique is insufficient for causality detection in industrial processes. Another contribution of this paper is a new method that can detect causality in industrial signals, and thus be used to validate the design of multilevel flow models.
Journal of Intelligent Fuzzy Systems Applications in Engineering and Technology, 2004
Visual quality assurance techniques focus on the detection and qualification of abnormal structur... more Visual quality assurance techniques focus on the detection and qualification of abnormal structures in the image of an object. The features of abnormality are extracted through image mining, whereupon classification is performed on characteristic combinations. Many techniques for feature extraction have been proposed, but the feed-forward neural network is seldom utilized despite its popularity in other application areas. Based on the wide experience base, this paper shows how a multi-tier feed-forward network can be constructed to model detectable peaks using only the physical properties of the image domain. This generic architecture can easily be adapted for different applications, as in metal plate inspection and protein detection, with mean error rate below 5%.
Uploads
Papers by Lambert Spaanenburg