Smart cities deploy various sensors such as microphones and RGB cameras to collect data improve the safety comfort of citizens. As annotation is expensive, self-supervised methods contrastive learning are used learn audio-visual representations for downstream tasks. Focusing on surveillance data, we investigate two common limitations learning: false negatives minimal sufficient information bottleneck. Irregular, yet frequently recurring events can lead a considerable number false-negative...

Unmanned Aerial Vehicles (UAVs) combined with Hyperspectral imaging (HSI) offer potential for environmental and agricultural applications by capturing detailed spectral information that enables the prediction of invisible features like biochemical leaf properties. However, data-intensive nature HSI poses challenges remote devices, which have limited computational resources storage. This paper introduces an Online Simple Linear Iterative Clustering algorithm (OHSLIC) framework real-time tree...

Soil compaction is a widespread problem, leading to soil degradation, yield losses, and adverse environmental impacts. Nowadays, various measurement methods exist assess map compaction, with vertical cone penetration resistance measurements being one of the most commonly used. This method easy, rapid, inexpensive, generally accepted. However, manual are time-consuming, labor-intensive, often less accurate due inconsistent speed. To address these limitations, an automated penetrometer was...

In this work we explore the generalization characteristics of unsupervised representation learning by leveraging disentangled VAE's to learn a useful latent space on set relational reasoning problems derived from Raven Progressive Matrices. We show that representations, learned training using right objective function, significantly outperform same architectures trained with purely supervised learning, especially when it comes generalization.

Automated anomaly detection in surveillance videos has attracted much interest as it provides a scalable alternative to manual monitoring. Most existing approaches achieve good performance on clean benchmark datasets recorded well-controlled environments. However, detecting anomalies is more challenging the real world. Adverse weather conditions like rain or changing brightness levels cause significant shift input data distribution, which turn can lead detector model incorrectly reporting...

Deep residual networks (ResNets) made a recent breakthrough in deep learning. The core idea of ResNets is to have shortcut connections between layers that allow the network be much deeper while still being easy optimize avoiding vanishing gradients. These interesting side-effects make behave differently from other typical architectures. In this work we use these properties design based on ResNet but with parameter sharing and adaptive computation time. resulting smaller than original can...

Inference of Deep Neural Networks for stream signal (Video/Audio) processing in edge devices is still challenging. Unlike the most state art inference engines which are efficient static signals, our brain optimized real-time dynamic processing. We believe one important feature (asynchronous state-full processing) key to its excellence this domain. In work, we show how asynchronous with neurons allows exploitation existing sparsity natural signals. This paper explains three different types...

Nowadays artificial neural networks are widely used to accurately classify and recognize patterns. An interesting application area is the Internet of Things (IoT), where physical things connected Internet, generate a huge amount sensor data that can be for myriad new, pervasive applications. Neural networks' ability comprehend unstructured make them useful building block such IoT As require lot processing power, especially during training phase, these most often deployed in cloud...

Internet applications rely on strong encryption techniques to protect the content of all communications between client and server. These algorithms ensure that third parties are unable obtain plain text data but also make it hard for network administrator enforce restrictions types traffic allowed. In this paper we show can train accurate machine learning models which predict type going through an IPsec or TOR tunnel based features extracted from encrypted streams. We use small, fast execute...

Artificial Neural Networks (ANNs) show great performance in several data analysis tasks including visual and auditory applications. However, direct implementation of these algorithms without considering the sparsity requires high processing power, consume vast amounts energy suffer from scalability issues. Inspired by biology, one methods which can reduce power consumption allow neural networks is asynchronous communication means action potentials, so-called spikes. In this work, we use...

Deep neural networks are the state of art technique for a wide variety classification problems. Although deeper able to make more accurate classifications, value brought by an additional hidden layer diminishes rapidly. Even shallow achieve relatively good results on various Only small subset samples do layers significant difference. We describe architecture in which only that can not be classified with sufficient confidence network have processed layers. Instead training one output at end...

Deep neural networks require large amounts of resources which makes them hard to use on resource constrained devices such as Internet-of-things devices. Offloading the computations cloud can circumvent these constraints but introduces a privacy risk since operator is not necessarily trustworthy. We propose technique that obfuscates data before sending it remote computation node. The obfuscated unintelligible for human eavesdropper still be classified with high accuracy by network trained...

Cyber-physical systems (CPS) in the factory of thefuture will consist cloud-hosted software governing an agileproduction process that is executed by mobile robots and thatis controlled analyzing data from a vast number ofsensors. CPSs thus operate on distributed production floorinfrastructure set-up continuously changes with eachnew manufacturing task. In this paper, we present our OSGibasedmiddleware abstracts deployment servicebasedCPS components platformcomprising robots, actuators,...

We present four training and prediction schedules from the same character-level recurrent neural network. The efficiency of these is tested in terms model effectiveness as a function time amount data seen. show that choice schedule potentially has considerable impact on for given budget.

As warehouses, storage facilities and factories become more expanded equipped with smart devices, there is a substantial need for rapid, intelligent autonomous detection of unusual potentially hazardous situations, also called anomalies. In particular Autonomous Guided Vehicles (AGVs) that drive around these premises independently, unforeseen obstructions along their path-e.g. cardboard box in the middle corridor or bumps floor-and sudden unexpected actions executed by personnel-e.g. someone...

Alternaria solani is the second most devastating foliar pathogen of potato crops worldwide, causing premature defoliation plants. This disease currently prevented through regular application detrimental crop protection products and guided by early warnings based on weather predictions visual observations farmers. To reduce use products, without additional production losses, it would be beneficial to able automatically detect in fields. In recent years, potential deep learning precision...