From the article: Abstract: An overview of neural network architectures is presented. Some of these architectures have been created in recent years, whereas others originate from many decades ago. Apart from providing a practical tool for comparing deep learning models, the Neural Network Zoo also uncovers a taxonomy of network architectures, their chronology, and traces back lineages and inspirations for these neural information processing systems.
DOCUMENT
We present a novel architecture for an AI system that allows a priori knowledge to combine with deep learning. In traditional neural networks, all available data is pooled at the input layer. Our alternative neural network is constructed so that partial representations (invariants) are learned in the intermediate layers, which can then be combined with a priori knowledge or with other predictive analyses of the same data. This leads to smaller training datasets due to more efficient learning. In addition, because this architecture allows inclusion of a priori knowledge and interpretable predictive models, the interpretability of the entire system increases while the data can still be used in a black box neural network. Our system makes use of networks of neurons rather than single neurons to enable the representation of approximations (invariants) of the output.
LINK
The recent success of Machine Learning encouraged research using artificial neural networks (NNs) in computer graphics. A good example is the bidirectional texture function (BTF), a data-driven representation of surface materials that can encapsulate complex behaviors that would otherwise be too expensive to calculate for real-time applications, such as self-shadowing and interreflections. We propose two changes to the state-of-the-art using neural networks for BTFs, specifically NeuMIP. These changes, suggested by recent work in neural scene representation and rendering, aim to improve baseline quality, memory footprint, and performance. We conduct an ablation study to evaluate the impact of each change. We test both synthetic and real data, and provide a working implementation within the Mitsuba 2 rendering framework. Our results show that our method outperforms the baseline in all these metrics and that neural BTF is part of the broader field of neural scene representation. Project website: https://traverse-research.github.io/NeuBTF/.
DOCUMENT
Passenger flow management is an important issue at many airports around the world. There are high concentrations of passengers arriving and leaving the airport in waves of large volumes in short periods, particularly in big hubs. This might cause congestion in some locations depending on the layout of the terminal building. With a combination of real airport data, as well as synthetic data obtained through an airport simulator, a Long Short-Term Memory Recurrent Neural Network has been implemented to predict the possible trajectories that passengers may travel within the airport depending on user-defined passenger profiles. The aim of this research is to improve passenger flow predictability and situational awareness to make a more efficient use of the airport, that could also positively impact communication with public and private land transport operators.
DOCUMENT
There is a growing literature investigating the relationship between oscillatory neural dynamics measured using electroencephalography (EEG) and/or magnetoencephalography (MEG), and sentence-level language comprehension. Recent proposals have suggested a strong link between predictive coding accounts of the hierarchical flow of information in the brain, and oscillatory neural dynamics in the beta and gamma frequency ranges. We propose that findings relating beta and gamma oscillations to sentence-level language comprehension might be unified under such a predictive coding account. Our suggestion is that oscillatory activity in the beta frequency range may reflect both the active maintenance of the current network configuration responsible for representing the sentence-level meaning under construction, and the top-down propagation of predictions to hierarchically lower processing levels based on that representation. In addition, we suggest that oscillatory activity in the low and middle gamma range reflect the matching of top-down predictions with bottom-up linguistic input, while evoked high gamma might reflect the propagation of bottom-up prediction errors to higher levels of the processing hierarchy. We also discuss some of the implications of this predictive coding framework, and we outline ideas for how these might be tested experimentally.
LINK
Objective. In this study it was investigated whether an artificial neural network can be used to determine the horizontal, fore-aft component of the ground reaction force from insole pressure patterns. Design. An artificial neural network was applied to map insole pressures and ground reaction forces. Method. To train an artificial neural network insole pressure patterns and ground reaction force data were simultaneously determined for a wide range of different speeds (0.9-2.3 m s−1) for five subjects. Both intrasubject and intersubject generalizability were evaluated. Results. At the intrasubject level generalizability was good when the speed for which the force was to be predicted was within the range of speeds from which data were used to train the network. Besides in some cases, generalizability to a condition outside the range of training conditions could be demonstrated. At the intersubject level the quality of generalization differed widely over subjects, from poor to good. Conclusions. It was found that an artificial neural network is able to map the relationship between insole pressure patterns and the fore-aft component of the ground reaction force. Relevance Good intrasubject generalization of 'knowledge' obtained by an artificial neural network will allow the assessment of the fore-aft component of ground reaction force in condition that cannot be evaluated with force plates, e.g. activities of daily living or real sport situations. Additionally, intersubject generalization will allow shear-force recordings in subjects that are not able to complete a great number of runs to acquire enough force-plate hits.
DOCUMENT
Masonry structures represent the highest proportion of building stock worldwide. Currently, the structural condition of such structures is predominantly manually inspected which is a laborious, costly and subjective process. With developments in computer vision, there is an opportunity to use digital images to automate the visual inspection process. The aim of this study is to examine deep learning techniques for crack detection on images from masonry walls. A dataset with photos from masonry structures is produced containing complex backgrounds and various crack types and sizes. Different deep learning networks are considered and by leveraging the effect of transfer learning crack detection on masonry surfaces is performed on patch level with 95.3% accuracy and on pixel level with 79.6% F1 score. This is the first implementation of deep learning for pixel-level crack segmentation on masonry surfaces. Codes, data and networks relevant to the herein study are available in: github.com/dimitrisdais/crack_detection_CNN_masonry.
DOCUMENT
A considerable amount of literature has been published on Corporate Reputation, Branding and Brand Image. These studies are extensive and focus particularly on questionnaires and statistical analysis. Although extensive research has been carried out, no single study was found which attempted to predict corporate reputation performance based on data collected from media sources. To perform this task, a biLSTM Neural Network extended with attention mechanism was utilized. The advantages of this architecture are that it obtains excellent performance for NLP tasks. The state-of-the-art designed model achieves highly competitive results, F1 scores around 72%, accuracy of 92% and loss around 20%.
DOCUMENT
Het doel van dit onderzoek is te onderzoeken onder welke omstandigheden en onder welke condities relatief moderne modelleringstechnieken zoals support vector machines, neural networks en random forests voordelen zouden kunnen hebben in medisch-wetenschappelijk onderzoek en in de medische praktijk in vergelijking met meer traditionele modelleringstechnieken, zoals lineaire regressie, logistische regressie en Cox regressie.
MULTIFILE
Anomaly detection is a key factor in the processing of large amounts of sensor data from Wireless Sensor Networks (WSN). Efficient anomaly detection algorithms can be devised performing online node-local computations and reducing communication overhead, thus improving the use of the limited hardware resources. This work introduces a fixed-point embedded implementation of Online Sequential Extreme Learning Machine (OS-ELM), an online learning algorithm for Single Layer Feed forward Neural Networks (SLFN). To overcome the stability issues introduced by the fixed precision, we apply correction mechanisms previously proposed for Recursive Least Squares (RLS). The proposed implementation is tested extensively with generated and real-world datasets, and compared with RLS, Linear Least Squares Estimation, and a rule-based method as benchmarks. The methods are evaluated on the prediction accuracy and on the detection of anomalies. The experimental results demonstrate that fixed-point OS-ELM can be successfully implemented on resource-limited embedded systems, with guarantees of numerical stability. Furthermore, the detection accuracy of fixed-point OS-ELM shows better generalization properties in comparison with, for instance, fixed-point RLS. © 2013 IEEE.
DOCUMENT
