The identification of biological echoes in radar data has revolutionized research into airborne migratory species. Deep learning applied to polarimetric weather radar observations can reveal signature patterns of mass movement by bio-scatterers such as birds, bats, and insects. However, due to the difficulties in labelling bio-scatterers in these data, threshold approaches have been proposed in the literature. In this research, we used the depolarization ratio (DR) based on differential reflectivity (zDR) and the cross-correlation coefficient (pHV), along with citizen scientist-reported data, to label bio-scatterers for deep learning. This method of labeling biological echoes in radar signatures is prone to noise, which impacts the accuracy of any model that relies on it. We introduce a novel semi-supervised co-training approach that uses a bootstrap ensemble with a confidence threshold. Our ensemble consists of the newly proposed STNet and two modified FNet models, which incorporate co-learning through bootstrap sampling for label correction. This innovative method significantly improves classification accuracy across all three multivariate numerical datasets compared to baseline models that lack co-learning with bootstrap-based label correction. • We used depolarization ratio and citizen science data to label biological echoes in polarimetric weather radar observation. • We used a deep learning approach to correct noisy biological-scatterer labels in polarimetric weather radar observations. • We developed a novel approach that uses semi-supervised co-training method based on bootstrap ensemble with a confidence threshold. • We introduced ensembles comprising STNet and a modified FNet with bootstrap sparse categorical cross-entropy loss. • We tested our method on other similar multivariate numerical datasets.

The adoption of high-accuracy speech recognition algorithms without an effective evaluation of their impact on the target computational resource is impractical for mobile and embedded systems. In this paper, techniques are adopted to minimise the required computational resource for an effective mobile-based speech recognition system. A Dynamic Multi-Layer Perceptron speech recognition technique, capable of running in real time on a state-of-the-art mobile device, has been introduced. Even though a conventional hidden Markov model when applied to the same dataset slightly outperformed our approach, its processing time is much higher. The Dynamic Multi-layer Perceptron presented here has an accuracy level of 96.94% and runs significantly faster than similar techniques.

For a considerable time, it has been the goal of computational neuroscientists to understand biological nervous systems. However, the vast complexity of such systems has made it very difficult to fully understand even basic functions such as movement. Because of its small neuron count, the C. elegans nematode offers the opportunity to study a fully described connectome and attempt to link neural network activity to behaviour. In this paper a simulation of the neural network in C. elegans that responds to chemical stimulus is presented and a consequent realistic head movement demonstrated. An evolutionary algorithm (EA) has been utilised to search for estimates of the values of the synaptic conductances and also to determine whether each synapse is excitatory or inhibitory in nature. The chemotaxis neural network was designed and implemented, using the parameterisation obtained with the EA, on the Si elegans platform a state-of-the-art hardware emulation platform specially designed to emulate the C. elegans nematode.

The nematode C. elegans (Caenorhabditis elegans) has for many years been instrumental as a model organism for fundamental research into biological neural networks, mainly to understand the behaviour and physiology of nervous systems. The Si elegans EU FP7 project aims to develop a Hardware Neural Network (HNN) to accurately replicate the C. elegans nervous system, behaviour and response to environmental changes that will enable neuroscientists to better understand these basic functions. In this paper, we focus on consideration of some specific behaviours of the worm for modelling in the hardware based Si elegans system. Environmental response to oxygen, food, light, temperature and other chemicals are reviewed to establish specific neurons responsible for the worm behaviours. Neuron interconnects mechanisms and suitable models simulating the biological worm are also discussed.

Understanding how simple biological entities perform complex tasks (such as moving away from predators, reproduction and seeking for food) is one of the main goals of researchers across the world. The C. elegans nematode is one of the simplest and well characterised biological nervous systems (BNS) and the Si elegans EU FP7 project aims to develop a framework to mimic its functionality and better understand basic functions. This paper presents a small scale emulation environment composed of 8 neurons running on field programmable gate arrays (FPGAs), where each neuron FPGA is interconnected to all of the other neuron FPGAs via an innovative optical synaptic connectivity strategy. This paper evaluates the fidelity of synaptic communication for transmission of spikes using both wired and optical methods.

The nematode C. elegans is the only organism to have its connectome fully documented. All 302 neurons of the worm are already known, however, their description is somewhat lacking. Using current computational techniques for clustering, such as Growing Self Organizing Networks and K-Means algorithm, this paper aims to provide a deeper analysis and understanding of the C. elegans connectome and the particulars of its individual neurons. The algorithms show that there are several groups of neurons, although the rational for the grouping is not clear and the clusters are inconsistent.

The hardware layer of the Si elegans EU FP7 project is a massively parallel architecture designed to accurately emulate the C. elegans nematode in biological real-time. The C. elegans nematode is one of the simplest and well characterized Biological Nervous Systems (BNS) yet many questions related to basic functions such as movement and learning remain unanswered. The hardware layer includes a Hardware Neural Network (HNN) composed of 302 FPGAs (one per neuron), a Hardware Muscle Network (HMN) composed of 27 FPGAs (one per 5 muscles) and one Interface Manager FPGA, which is physically connected through 2 Local Area Networks (LANs) and through an innovative 3D optical connectome. Neuron structures (gap junctions and synapses) and muscles are modelled in the design environment of the software layer and their simulation data (spikes, variable values and parameters) generate data packets sent across the Local Area Networks (LAN). Furthermore, a software layer gives the user a set of design tools giving the required flexibility and high level hardware abstraction to design custom neuronal models. In this paper the authors present an overview of the hardware layer, connections infrastructure and communication protocol.

A potential threat to embedded systems is the execution of unknown or malicious software capable of triggering harmful system behavior, aimed at theft of sensitive data or causing damage to the system. Commercial off-the-shelf embedded devices, such as embedded medical equipment, are more vulnerable as these type of products cannot be amended conventionally or have limited resources to implement protection mechanisms. In this paper, we present a self-organizing map (SOM)-based approach to enhance embedded system security by detecting abnormal program behavior. The proposed method extracts features derived from processor's program counter and cycles per instruction, and then utilises the features to identify abnormal behavior using the SOM. Results achieved in our experiment show that the proposed method can identify unknown program behaviors not included in the training set with over 98.4% accuracy.

This paper presents a system for automatically classifying the resting location of a moving object in an indoor environment. The system uses an unsupervised neural network (Self Organising Feature Map) fully implemented on a low-cost, low-power automated home-based surveillance system, capable of monitoring activity level of elders living alone independently. The proposed system runs on an embedded platform with a specialised ceiling-mounted video sensor for intelligent activity monitoring. The system has the ability to learn resting locations, to measure overall activity levels and to detect specific events such as potential falls. First order motion information, including first order moving average smoothing, is generated from the 2D image coordinates (trajectories). A novel edge-based object detection algorithm capable of running at a reasonable speed on the embedded platform has been developed. The classification is dynamic and achieved in real-time. The dynamic classifier is achieved using a SOFM and a probabilistic model. Experimental results show less than 20% classification error, showing the robustness of our approach over others in literature with minimal power consumption. The head location of the subject is also estimated by a novel approach capable of running on any resource limited platform with power constraints.

This paper introduces a tri-state logic self-organizing map (bSOM) designed and implemented on a field programmable gate array (FPGA) chip. The bSOM takes binary inputs and maintains tri-state weights. A novel training rule is presented. The bSOM is well suited to FPGA implementation, trains quicker than the original self-organizing map (SOM), and can be used in clustering and classification problems with binary input data. Two practical applications, character recognition and appearance-based object identification, are used to illustrate the performance of the implementation. The appearance-based object identification forms part of an end-to-end surveillance system implemented wholly on FPGA. In both applications, binary signatures extracted from the objects are processed by the bSOM. The system performance is compared with a traditional SOM with real-valued weights and a strictly binary weighted SOM.

Methods of segmenting objects of interest from video data typically use a background model to represent an empty, static scene. However, dynamic processes in the background, such as moving foliage and water, can act to undermine the robustness of such methods and result in false positive object detections. Techniques for reducing errors have been proposed, including Markov Random Field (MRF) based pixel classification schemes, and also the use of region-based models. The work we present here combines these two approaches, using a region-based background model to provide robust likelihoods for multi-class MRF pixel labelling. Our initial results show the effectiveness of our method, by comparing performance with an analogous per-pixel likelihood model.

A binary Self Organizing Map (SOM) has been designed and implemented on a Field Programmable Gate Array (FPGA) chip. A novel learning algorithm which takes binary inputs and maintains tri-state weights is presented. The binary SOM has the capability of recognizing binary input sequences after training. A novel tri-state rule is used in updating the network weights during the training phase. The rule implementation is highly suited to the FPGA architecture, and allows extremely rapid training. This architecture may be used in real-time for fast pattern clustering and classification of binary features.

The sparse distributed memory (SDM) proposed by Kanerva provides a simple model for human long-term memory, with a strong underlying mathematical theory. However, there are problematic features in the original SDM model that affect its efficiency and performance in real world applications and for hardware implementation. In this paper, we propose modifications to the SDM model that improve its efficiency and performance in pattern recall. First, the address matrix is built using training samples rather than random binary sequences. This improves the recall performance significantly. Second, the content matrix is modified using a simple tri-state logic rule. This reduces the storage requirements of the SDM and simplifies the implementation logic, making it suitable for hardware implementation. The modified model has been tested using pattern recall experiments. It is found that the modified model can recall clean patterns very well from noisy inputs.

The lobula giant movement detector (LGMD) is a wide-field visual neuron that is located in the lobula layer of the locust nervous system. The LGMD increases its firing rate in response to both the velocity of the approaching object and its proximity. It has been found that it can respond to looming stimuli very quickly and can trigger avoidance reactions whenever a rapidly approaching object is detected. It has been successfully applied in visual collision avoidance systems for vehicles and robots. This paper proposes a modified LGMD model that provides additional movement depth direction information. The proposed model retains the simplicity of the previous neural network model, adding only a few new cells. It has been tested on both simulated and recorded video data sets. The experimental results shows that the modified model can very efficiently provide stable information on the depth direction of movement.

This paper introduces a real-time connected component labelling algorithm designed for field programmable gate array (FPGA) implementation. The algorithm run-length encodes the image, and performs connected component analysis on this representation. The run-length encoding, together with other parts of the algorithm, is performed in parallel; sequential operations are minimized as the number of runs are typically less than the number of pixels. The architecture is designed mainly on Block RAM (i.e. internal RAM) of the FPGA. A comparison with the multi-pass algorithm in hardware and software is presented to show the advantages of the algorithm. The algorithm runs comfortably in real-time with reasonably low resource utilization, making integration with other real-time algorithms feasible.