In industrial applications, the early detection of malfunctioning factory machinery is crucial.In this paper, we consider acoustic malfunction via transfer learning.Contrary to majority current approaches which are based on deep autoencoders, propose extract features using neural networks that were pretrained task image classification.We then use these train a variety anomaly models and show improves results compared convolutional autoencoders in recordings four different machines noisy...

In this work, we present a general procedure for acoustic leak detection in water networks that satisfies multiple real-world constraints such as energy efficiency and ease of deployment. Based on recordings from seven contact microphones attached to the supply network municipal suburb, trained several shallow deep anomaly models. Inspired by how human experts detect leaks using electronic sounding-sticks, use these models repeatedly listen over predefined decision horizon. This way avoid...

In this work, we thoroughly evaluate the efficacy of pretrained neural networks as feature extractors for anomalous sound detection.In doing so, leverage knowledge that is contained in these to extract semantically rich features (representations) serve input a Gaussian Mixture Model which used density estimator model normality.We compare were trained on data from various domains, namely: images, environmental sounds and music.Our approach evaluated recordings factory machinery such valves,...

We apply the vision transformer, a deep machine learning model build around attention mechanism, on mel-spectrogram representations of raw audio recordings.When adding melbased data augmentation techniques and sample-weighting, we achieve comparable performance both (PRS CCS challenge) tasks ComParE21, outperforming most single baselines.We further introduce overlapping vertical patching evaluate influence parameter configurations.

In many fields of research, labeled datasets are hard to acquire. This is where data augmentation promises overcome the lack training in context neural network engineering and classification tasks. The idea here reduce model over-fitting feature distribution a small under-descriptive dataset. We try evaluate such techniques gather insights performance boost they provide for several convolutional networks on mel-spectrogram representations audio data. show impact binary task surgical mask...

Detecting sleepiness from spoken language is an ambitious task, which addressed by the Interspeech 2019 Computational Paralinguistics Challenge (ComParE).We propose end-to-end deep learning approach to detect and classify patterns reflecting in human voice.Our based solely on a moderately complex neural network architecture.It may be applied directly audio data without requiring any specific feature engineering, thus remaining transferable other classification tasks.Nevertheless, our...

A key element of biological structures is self-replication. Neural networks are the prime structure used for emergent construction complex behavior in computers. We analyze how various network types lend themselves to Backpropagation turns out be natural way navigate space weights and allows non-trivial self-replicators arise naturally. perform an in-depth analysis show self-replicators' robustness noise. then introduce artificial chemistry environments consisting several neural examine...

An anomalous sound detection (ASD) system detects substantial deviations from the norm and reports degree of abnormality through an anomaly score. important application scenario is malfunctions in factory machinery. Recent approaches train autoencoders on small segments sound's time-frequency representation use reconstruction error as a measure abnormality. However, it was recently shown that this approach leads to consistently higher errors for edge frames segments. To alleviate problem,...

Abstract. Fully autonomously driving vehicles are expected to be a widely available technology in the near future. Privately owned cars, which remain parked for majority of their lifetime, may therefore capable independently during usual long parking periods (e.g. owners working hours). Our analysis aims focus on potential privately shared car concept as transition period between present usages cars towards transportation paradigm autonomous vehicles. We propose two methods field...

Recent advances in Multi-Agent Reinforcement Learning have prompted the modeling of intricate interactions between agents simulated environments. In particular, predator-prey dynamics captured substantial interest and various simulations been tailored to unique requirements. To prevent further time-intensive developments, we introduce Aquarium, a comprehensive environment for interaction, enabling study emergent behavior. Aquarium is open source offers seamless integration PettingZoo...

Emergent effects can arise in multi-agent systems (MAS) where execution is decentralized and reliant on local information. These may range from minor deviations behavior to catastrophic system failures. To formally define these effects, we identify misalignments between the global inherent specification (the true specification) its approximation (such as configuration of different reward components or observations). Using established safety terminology, develop a framework understand...

The foundation of biological structures is self-replication. Neural networks are the prime structure used for emergent construction complex behavior in computers. We analyze how various netw...

In the near future, more and machines will perform tasks in vicinity of human spaces or support them directly their spatially bound activities. order to simplify verbal communication interaction between robotic units and/or humans, reliable robust systems w.r.t. noise processing results are needed. This work builds a foundation address this task. By using continuous representation spatial perception interiors learned from trajectory data, our approach clusters movement dependency its...

The ubiquitous availability of mobile devices capable location tracking led to a significant rise in the collection GPS data. Several compression methods have been developed order reduce amount storage needed while keeping important information. In this paper, we present an lstm-autoencoder based approach compress and reconstruct trajectories, which is evaluated on both gaming real-world dataset. We consider various ratios trajectory lengths. performance compared other algorithms, i.e.,...