Autoencoder

Fault diagnosis of electric motors is a fundamental task for production line testing, and it is usually performed by experienced human operators. In the recent years, several methods have been proposed in the literature for detecting faults automatically. Deep neural networks have been successfully employed for this task, but, up to the authors' knowledge, they have never been used in an unsupervised scenario. This paper proposes an unsupervised method for diagnosing faults of electric motors by using a novelty detection approach based on deep autoencoders. In the proposed method, vibration signals are acquired by using accelerometers and processed to extract Log-Mel coefficients as features. Autoencoders are trained by using normal data only, i.e., data that do not contain faults. Three different autoencoders architectures have been evaluated: the multi-layer perceptron (MLP) autoencoder, the convolutional neural network autoencoder, and the recurrent autoencoder composed of long short-term memory (LSTM) units. The experiments have been conducted by using a dataset created by the authors, and the proposed approaches have been compared to the one-class support vector machine (OC-SVM) algorithm. The performance has been evaluated in terms area under curve (AUC) of the receiver operating characteristic curve, and the results showed that all the autoencoder-based approaches outperform the OC-SVM algorithm. Moreover, the MLP autoencoder is the most performing architecture, achieving an AUC equal to 99.11%.

The Free Energy Principle and Active Inference Framework (FEP-AI) begins with the understanding that persisting systems must regulate environmental exchanges and prevent entropic accumulation. In FEP-AI, minds and brains are predictive controllers for autonomous systems, where action-driven perception is realized as probabilistic inference. Integrated Information Theory (IIT) begins with considering the preconditions for a system to intrinsically exist, as well as axioms regarding the nature of consciousness. IIT has produced controversy because of its surprising entailments: quasi-panpsychism; subjectivity without referents or dynamics; and the possibility of fully-intelligent-yet-unconscious brain simulations. Here, I describe how these controversies might be resolved by integrating IIT with FEP-AI, where integrated information only entails consciousness for systems with perspectival reference frames capable of generating models with spatial, temporal, and causal coherence for self and world. Without that connection with external reality, systems could have arbitrarily high amounts of integrated information, but nonetheless would not entail subjective experience. I further describe how an integration of these frameworks may contribute to their evolution as unified systems theories and models of emergent causation. Then, inspired by both Global Neuronal Workspace Theory (GNWT) and the Harmonic Brain Modes framework, I describe how streams of consciousness may emerge as an evolving generation of sensorimotor predictions, with the precise composition of experiences depending on the integration abilities of synchronous complexes as self-organizing harmonic modes (SOHMs). These integrating dynamics may be particularly likely to occur via richly connected subnetworks affording body-centric sources of phenomenal binding and executive control. Along these connectivity backbones, SOHMs are proposed to implement turbo coding via loopy message-passing over predictive (autoencoding) networks, so generating maximum a posteriori estimates as coherent vectors governing neural evolution, with alpha frequencies generating basic awareness, and cross-frequency phase-coupling within theta frequencies for access consciousness and volitional control. These dynamic cores of integrated information also function as global workspaces, centered on posterior cortices, but capable of being entrained with frontal cortices and interoceptive hierarchies, so affording agentic causation. Integrated World Modeling Theory (IWMT) represents a synthetic approach to understanding minds that reveals compatibility between leading theories of consciousness, so enabling inferential synergy.

Deep learning has been successfully applied to solve various complex problems ranging from big data analytics to computer vision and human-level control. Deep learning advances however have also been employed to create software that can cause threats to privacy, democracy and national security. One of those deep learning-powered applications recently emerged is "deepfake". Deepfake algorithms can create fake images and videos that humans cannot distinguish them from authentic ones. The proposal of technologies that can automatically detect and assess the integrity of digital visual media is therefore indispensable. This paper presents a survey of algorithms used to create deepfakes and, more importantly, methods proposed to detect deepfakes in the literature to date. We present extensive discussions on challenges, research trends and directions related to deepfake technologies. By reviewing the background of deepfakes and state-of-the-art deepfake detection methods, this study provides a comprehensive overview of deepfake techniques and facilitates the development of new and more robust methods to deal with the increasingly challenging deepfakes.

Heterogeneous domain adaptation network based on autoencoder, J. Parallel Distrib. Comput. (2017), Abstract: Heterogeneous domain adaptation is a more challenging problem than homogeneous domain adaptation. The transfer effect is not ideal caused by shallow structure which cannot adequately describe the probability distribution and obtain more effective features. In this paper, we propose a heterogeneous domain adaptation network based on autoencoder, in which two sets of autoencoder networks are used to project the source-domain and target-domain data to a shared feature space to obtain more abstractive feature representations. In the last feature and classification layer, the marginal and conditional distributions can be matched by empirical maximum mean discrepancy metric to reduce distribution difference. To preserve the consistency of geometric structure and label information, a manifold alignment term based on labels is introduced. The classification performance can be improved further by making full use of label information of both domains. The experimental results of 16 cross-domain transfer tasks verify that HDANA outperforms several state-of-the-art methods.

This paper presents and evaluates a deep learning architecture for automated basal cell carcinoma cancer detection that integrates (1) image representation learning, (2) image classification and (3) result interpretability. A novel characteristic of this approach is that it extends the deep learning architecture to also include an interpretable layer that highlights the visual patterns that contribute to discriminate between cancerous and normal tissues patterns, working akin to a digital staining which spotlights image regions important for diagnostic decisions. Experimental evaluation was performed on set of 1,417 images from 308 regions of interest of skin histopathology slides, where the presence of absence of basal cell carcinoma needs to be determined. Different image representation strategies, including bag of features (BOF), canonical (discrete cosine transform (DCT) and Haar-based wavelet transform (Haar)) and proposed learned-from-data representations, were evaluated for comparison. Experimental results show that the representation learned from a large histology image data set has the best overall performance (89.4% in F-measure and 91.4% in balanced accuracy), which represents an improvement of around 7% over canonical representations and 3% over the best equivalent BOF representation.

Designing a robust anomaly detection system for a computer system and applications, is very essential for system security. Host-based Intrusion Detection System monitors system call sequences to prevent the execution of malicious codes on the host. This study applies and compares deep learning based LSTM (Long Short Term Memory)
Sequence-to-Sequence (Seq-Seq) and AutoEncoder in detecting an anomaly in Windows system call sequences. We use memory forensic techniques to extract system call sequences data of benign and malicious processes. Subsequently, we pre-process the extracted system call sequences to obtain valid system call sequences through filtering and ordering of redundant system calls. Our comparison analysis shows that LSTM Seq-Seq with a sequence length of ten gives high anomaly detection rate of 97.2% as compared to autoencoder.

Wireless sensor networks (WSN) are fundamental to the Internet of Things (IoT) by bridging the gap between the physical and the cyber worlds. Anomaly detection is a critical task in this context as it is responsible for identifying various events of interests such as equipment faults and undiscovered phenomena. However, this task is challenging because of the elusive nature of anomalies and the volatility of the ambient environments. In a resource-scarce setting like WSN, this challenge is further elevated and weakens the suitability of many existing solutions. In this paper, for the first time, we introduce autoencoder neural networks into WSN to solve the anomaly detection problem. We design a two-part algorithm that resides on sensors and the IoT cloud respectively, such that (i) anomalies can be detected at sensors in a fully distributed manner without the need for communicating with any other sensors or the cloud, and (ii) the relatively more computationintensive learning task can be handled by the cloud with a much lower (and configurable) frequency. In addition to the minimal communication overhead, the computational load on sensors is also very low (of polynomial complexity) and readily affordable by most COTS sensors. Using a real WSN indoor testbed and sensor data collected over 4 consecutive months, we demonstrate via experiments that our proposed autoencoderbased anomaly detection mechanism achieves high detection accuracy and low false alarm rate. It is also able to adapt to unforeseeable and new changes in a non-stationary environment, thanks to the unsupervised learning feature of our chosen autoencoder neural networks.

Gravitational wave astronomy is a rapidly growing field of modern astrophysics, with observations being made frequently by the LIGO detectors. Gravitational wave signals are often extremely weak and the data from the detectors, such as LIGO, is contaminated with non-Gaussian and non-stationary noise, often containing transient disturbances which can obscure real signals. Traditional denoising methods, such as principal component analysis and dictionary learning, are not optimal for dealing with this non-Gaussian noise, especially for low signal-to-noise ratio gravitational wave signals. Furthermore, these methods are computationally expensive on large datasets. To overcome these issues, we apply state-of-the-art signal processing techniques, based on recent groundbreaking advancements in deep learning, to denoise gravitational wave signals embedded either in Gaussian noise or in real LIGO noise. We introduce SMTDAE, a Staired Multi-Timestep Denoising Autoencoder, based on sequence-to-sequence bi-directional Long-Short-Term-Memory recurrent neural networks. We demonstrate the advantages of using our unsupervised deep learning approach and show that, after training only using simulated Gaussian noise, SMTDAE achieves superior recovery performance for gravitational wave signals embedded in real non-Gaussian LIGO noise.

Rapid growth in numbers of connected devices including sensors, mobile, wearable, and other Internet of Things (IoT) devices, is creating an explosion of data that are moving across the network. To carry out machine learning (ML), IoT data are typically transferred to the cloud or another centralized system for storage and processing; however, this causes latencies and increases network traffic. Edge computing has the potential to remedy those issues by moving computation closer to the network edge and data sources. On the other hand, edge computing is limited in terms of computational power and thus is not well suited for ML tasks. Consequently, this paper aims to combine edge and cloud computing for IoT data analytics by taking advantage of edge nodes to reduce data transfer. In order to process data close to the source, sensors are grouped according to locations, and feature learning is performed on the close by edge node. For comparison reasons, similarity-based processing is also considered. Feature learning is carried out with deep learning: the encoder part of the trained autoencoder is placed on the edge and the decoder part is placed on the cloud. The evaluation was performed on the task of human activity recognition from sensor data. The results show that when sliding windows are used in the preparation step, data can be reduced on the edge up to 80% without significant loss in accuracy.

Deep neural networks (DNN) have proven high efficiency in many solutions in the industry and the academic research. However, they face many limitations, and challenges such as the insufficiency in data or the noise effects that leads to the common overfitting problem. Among many solutions that were proposed to face these challenges, we propose the use of deep Autoencoder (DAE) to improve the accuracy of DNN with and without the effect of noise. The experiment has been done on three medical datasets for diagnosing diabetes, heart attack seizures and autism in toddlers. We have built DAE model, and a model of DNN using Keras, and evaluated the performance of DNN with and without the use of DAE as an auxiliary task. Then we added noise and evaluated the results. The DAE has proven to enhance the efficiency of one model with and without noise, while it has proven to enhance performance in the two other models if the noise was applied.

The rapid increase in network traffic has recently led to the importance of flow-based intrusion detection systems processing a small amount of traffic data. Furthermore, anomaly-based methods, which can identify unknown attacks are also integrated into these systems. In this study, the focus is concentrated on the detection of anomalous network traffic (or intrusions) from flow-based data using unsupervised deep learning methods with semi-supervised learning approach. More specifically, Autoencoder and Variational Autoencoder methods were employed to identify unknown attacks using flow features. In the experiments carried out, the flow-based features extracted out of network traffic data, including typical and different types of attacks, were used. The Receiver Operating Characteristics (ROC) and the area under ROC curve, resulting from these methods were calculated and compared with One-Class Support Vector Machine. The ROC curves were examined in detail to analyze the performance of the methods in various threshold values. The experimental results show that Variational Autoencoder performs, for the most part, better than Autoencoder and One-Class Support Vector Machine.

Based on neural network and machine learning, we apply the energy disaggregation for both classification (prediction on usage time) and estimation (prediction on usage amount) on 150 AMI (Advanced Metering Infrastructure) smart meters and a small amount of HEMS (Home Energy Management System) smart plugs in a community in New Taipei City, Taiwan. The aim of this paper is to clarify how we lower the cost, obtain the model of appliance usage from only a small portion of households, improve it with simple questionnaire, and generalize it for prediction on collective households. Our investigation demonstrates the benefits and various possibilities for power suppliers and the government, and won the Elite Award in the Presidential Hackathon 2020, Taiwan.

Variational Autoencoders play important role in text generation tasks, when semantically consistent latent space is needed. However , training VAE for text is not a trivial task due to mode collapse issue. In this paper, autoencoder with binary latent space trained using straight-through estimator is shown to have advantages over VAE on text modeling task. In our model, Bernoulli distribution is used instead of Gaussian (usual for VAE). The model can be trained with only reconstruction objective, without using any additional terms such as KL divergence. Experiments reported in this paper show binary autoencoder to have the main features of VAE: semantic consistency and good latent space coverage; while not suffering from the mode collapse and being a lot easier to train than VAE.

Restricted Boltzmann Machines (RBMs) and autoencoders have been used-in several variants-for similar tasks, such as reducing dimensionality or extracting features from signals. Even though their structures are quite similar, they rely on different training theories. Lately, they have been largely used as building blocks in deep learning architectures that are called deep belief networks (instead of stacked RBMs) and stacked autoencoders.
In light of this, the student has worked on this thesis with the aim to understand the extent of the similarities and the overall pros and cons of using either RBMs, autoencoders or denoising autoencoders in deep networks. Important characteristics are tested, such as the robustness to noise, the influence on training of the availability of data and the tendency to overtrain. The author has then dedicated part of the thesis to study how the three deep networks in exam form their deep internal representations and how similar these can be to each other. In result of this, a novel approach for the evaluation of internal representations is presented with the name of F-Mapping. Results are reported and discussed.

The  problem  of  predicting  links  has  gained  much  attention  in  recent  years  due  toits vast application in various domains such as sociology, network analysis, information science,etc.  Many  methods  have  been  proposed  for  link  prediction  such  as  RA,  AA,  CCLP,  etc.  Thesemethods required hand-crafted structural features to calculate the similarity scores between a pairof  nodes  in  a  network.  Some  methods  use  local  structural  information  while  others  use  globalinformation of a graph. These methods do not tell which properties are better than others. Withan  in-depth  analysis  of  these  methods,  we  understand  that  one  way  to  overcome  this  problemis  to  consider  network  structure  and  node  attribute  information  to  capture  the  discriminativefeatures for link prediction tasks. We proposed a deep learning Autoencoder based Link Prediction(ALP)  architecture  for  the  latent  representation  of  a  graph,  unified  with  non-negative  matrixfactorization to automatically determine the underlying roles in a network, after that assigning amixed-membership of these roles to each node in the network. The idea is to transfer these rolesas a feature vector for the link prediction task in the network. Further, cosine similarity is appliedafter getting the required features to compute the pairwise similarity score between the nodes. Wepresent the performance of the algorithm on the real-world datasets, where it gives the competitiveresult compared to other algorithms.

The Bag-of-Visual Words has been recognised as an effective mean of representing images for image classification. However, its reliance on hand crafted image feature extraction algorithms often results in significant computational overhead, and poor classification accuracies. Therefore, this paper presents a Bag-of-Visual Word Modelling in which Image Feature Extraction is achieved using Deep Feature Learning via Stacked-Autoencoder. The proposed method is tested using three image collections constituted from the Caltech 101 image collection and the results confirm the ability of deep feature learning to yield optimum image categorisation performance.

Rapid growth in numbers of connected devices including sensors, mobile, wearable, and other Internet of Things (IoT) devices, is creating an explosion of data that are moving across the network. To carry out machine learning (ML), IoT data are typically transferred to the cloud or another centralized system for storage and processing; however, this causes latencies and increases network traffic. Edge computing has the potential to remedy those issues by moving computation closer to the network edge and data sources. On the other hand, edge computing is limited in terms of computational power and thus is not well suited for ML tasks. Consequently, this paper aims to combine edge and cloud computing for IoT data analytics by taking advantage of edge nodes to reduce data transfer. In order to process data close to the source, sensors are grouped according to locations, and feature learning is performed on the close by edge node. For comparison reasons, similarity-based processing is al...

Advances on bidirectional intelligence are overviewed along three threads, with extensions and new perspectives. The first thread is about bidirectional learning architecture, exploring five dualities that enable Lmser six cognitive functions and provide new perspectives on which a lot of extensions and particularlly flexible Lmser are proposed. Interestingly, either or two of these dualities actually takes an important role in recent models such as U-net, ResNet, and DenseNet. The second thread is about bidirectional learning principles unified by best yIng-yAng (IA) harmony in BYY system. After getting insights on deep bidirectional learning from a bird-viewing on existing typical learning principles from one or both of the inward and outward directions, maximum likelihood, variational principle, and several other learning principles are summarised as exemplars of the BYY learning, with new perspectives on advanced topics. The third thread further proceeds to deep bidirectional intelligence, driven by long term dynamics (LTD) for parameter learning and short term dynamics (STD) for image thinking and rational thinking in harmony. Image thinking deals with information flow of continuously valued arrays and especially image sequence, as if thinking was displayed in the real world, exemplified by the flow from inward encoding/cognition to outward reconstruction/transformation performed in Lmser learning and BYY learning. In contrast, rational thinking handles symbolic strings or discretely valued vectors, performing uncertainty reasoning and problem solving. In particular, a general thesis is proposed for bidirectional intelligence, featured by BYY intelligence potential theory (BYY-IPT) and nine essential dualities in architecture, fundamentals, and implementation, respectively. Then, problems of combinatorial solving and uncertainty reasoning are investigated from this BYY IPT perspective. First, variants and extensions are suggested for AlphaGoZero like searching tasks, such as traveling salesman problem (TSP) and attributed graph matching (AGM) that are turned into Go like problems with help of a feature enrichment technique. Second, reasoning activities are summarized under guidance of BYY IPT from the aspects of constraint satisfaction, uncertainty propagation, and path or tree searching. Particularly, causal potential theory is proposed for discovering causal direction, with two roads developed for its implementation.

According to the World Health Organization, the number of people suffering from dementia worldwide will grow to 150 million by mid-century, and Alzheimer’s disease is the most common form of dementia contributing to 60%–70% of cases. The problem is compounded by the fact that current pharmacologic treatments are only symptomatic, and therapies are ineffective in slow down or cure the degenerative process. An automatic and standardize classifier for Alzheimer’s disease is thereby extremely important to rapidly respond and deliver as preventive as possible interventions. Speech alterations might be one of the earliest signs of cognitive defect and, recently, the researchers showed that they can be observable well in advance other cognitive deficits become manifest. In this paper, we propose a full automated method able to classify the spontaneous spoken production of the subjects. In particular, we trained an artificial neural network using the spectrogram of the audio signal, which is the visual representation of the speech of the subject. Moreover, to overcome the problem of the large amount of annotated data usually required for training deep learning models, we used a specific data augmentation approach that avoids distorting the original samples. We evaluated the proposed method using the English Pitt Corpus from DementiaBank. The used dataset consists of 180 subjects: 43 healthy controls and 137 Alzheimer’s disease patients. The proposed method outperformed the other approaches in the literature based on manual and semi-automatic transcription and annotation of speech, improving the classification capability by 5.93%, and obtained good classification results compared to the state-of-the-art neuropsychological screening tests (i.e., the Mini-Mental State Examination and the Activities of Daily Living portion of the Blessed Dementia Rating Scale) exhibiting an accuracy of 93.30% and an F1 score of 88.50%.

The manufacturing of nanomaterials by the electrospinning process requires accurate and meticulous inspection of related scanning electron microscope (SEM) images of the electrospun nanofiber, to ensure that no structural defects are produced. The presence of anomalies prevents practical application of the electrospun nanofibrous material in nanotechnology. Hence, the automatic monitoring and quality control of nanomaterials is a relevant challenge in the context of Industry 4.0. In this paper, a novel automatic classification system for homogenous (anomaly-free) and non-homogenous (with defects) nanofibers is proposed. The inspection procedure aims at avoiding direct processing of the redundant full SEM image. Specifically, the image to be analyzed is first partitioned into sub-images (nanopatches) that are then used as input to a hybrid unsupervised and supervised machine learning system. In the first step, an autoencoder (AE) is trained with unsupervised learning to generate a code representing the input image with a vector of relevant features. Next, a multilayer perceptron (MLP), trained with supervised learning, uses the extracted features to classify non-homogenous nanofiber (NH-NF) and homogenous nanofiber (H-NF) patches. The resulting novel AE-MLP system is shown to outperform other standard machine learning models and other recent state-of-the-art techniques, reporting accuracy rate up to 92.5%. In addition, the proposed approach leads to model complexity reduction with respect to other deep learning strategies such as convolutional neural networks (CNN). The encouraging performance achieved in this benchmark study can stimulate the application of the proposed scheme in other challenging industrial manufacturing tasks.

Biometrics identification using multiple modalities has attracted the attention of many researchers as it produces more robust and trustworthy results than single modality biometrics. In this paper, we present a novel multimodal recognition system that trains a Deep Learning Network to automatically learn features after extracting multiple biometric modalities from a single data source, i.e., facial video clips. Utilizing different modalities, i.e., left ear, left profile face, frontal face, right profile face, and right ear, present in the facial video clips, we train supervised denosing autoencoders to automatically extract robust and non-redundant features. The automatically learned features are then used to train modality specific sparse classifiers to perform the multimodal recognition. Experiments conducted on the constrained facial video dataset (WVU) and the unconstrained facial video dataset (HONDA/UCSD), resulted in a 99.17% and 97.14% rank-1 recognition rates, respectively. The multimodal recognition accuracy demonstrates the superiority and robustness of the proposed approach irrespective of the illumination, non-planar movement, and pose variations present in the video clips.

This paper presents and evaluates a deep learning architecture for automated basal cell carcinoma cancer detection that integrates (1) image representation learning, (2) image classification and (3) result interpretability. A novel characteristic of this approach is that it extends the deep learning architecture to also include an interpretable layer that highlights the visual patterns that contribute to discriminate between cancerous and normal tissues patterns, working akin to a digital staining which spotlights image regions important for diagnostic decisions. Experimental evaluation was performed on set of 1,417 images from 308 regions of interest of skin histopathology slides, where the presence of absence of basal cell carcinoma needs to be determined. Different image representation strategies, including bag of features (BOF), canonical (discrete cosine transform (DCT) and Haar-based wavelet transform (Haar)) and proposed learnedfrom-data representations, were evaluated for comparison. Experimental results show that the representation learned from a large histology image data set has the best overall performance (89.4% in F-measure and 91.4% in balanced accuracy), which represents an improvement of around 7% over canonical representations and 3% over the best equivalent BOF representation.

Biometrics identification using multiple modalities has attracted the attention of many researchers as it produces more robust and trustworthy results than single modality biometrics. In this paper, we present a novel multimodal recognition system that trains a Deep Learning Network to automatically learn features after extracting multiple biometric modalities from a single data source, i.e., facial video clips. Utilizing different modalities, i.e., left ear, left profile face, frontal face, right profile face, and right ear, present in the facial video clips, we train supervised denosing autoencoders to automatically extract robust and non-redundant features. The automatically learned features are then used to train modality specific sparse classifiers to perform the multimodal recognition. Experiments conducted on the constrained facial video dataset (WVU) and the unconstrained facial video dataset (HONDA/UCSD), resulted in a 99.17% and 97.14% rank-1 recognition rates, respectively. The multimodal recognition accuracy demonstrates the superiority and robustness of the proposed approach irrespective of the illumination, non-planar movement, and pose variations present in the video clips.

Restricted Boltzmann Machines (RBMs) and autoencoders have been used - in several variants - for similar tasks, such as reducing dimensionality or extracting features from signals. Even though their structures are quite similar, they rely on different training theories. Lately, they have been largely used as building blocks in deep learning architectures that are called deep belief networks (instead of stacked RBMs) and stacked autoencoders.In light of this, the student has worked on this thesis with the aim to understand the extent of the similarities and the overall pros and cons of using either RBMs, autoencoders or denoising autoencoders in deep networks. Important characteristics are tested, such as the robustness to noise, the influence on training of the availability of data and the tendency to overtrain. The author has then dedicated part of the thesis to study how the three deep networks in exam form their deep internal representations and how similar these can be to each o...

Image denoising is a crucial topic in image processing. Noisy images are generated due to technical and environmental errors. Therefore, it is reasonable to consider image denoising an important topic to study, as it also helps to resolve other image processing issues. However, the challenge is that the classical techniques used are time-consuming and not flexible enough. This article compares two major neural network architectures which look promising to resolve these issues. The AutoEncoder and UNET is now the most researched subject in deep learning for image denoising. Multiple model architectures are designed, implement, and evaluated. The dataset is preprocessed and then it is used to train and test the model. It is clearly shown in this paper which model performs the best in this task by comparing both models using the most used parameters to evaluate image quality PSNR and SSIM.

In recent years, advances in neural variational inference have achieved many successes in text processing. Examples include neural topic models which are typically built upon varia-tional autoencoder (VAE) with an objective of minimising the error of reconstructing original documents based on the learned latent topic vectors. However, minimising reconstruction errors does not necessarily lead to high quality topics. In this paper, we borrow the idea of reinforcement learning and incorporate topic coherence measures as reward signals to guide the learning of a VAE-based topic model. Furthermore , our proposed model is able to automatically separating background words dynamically from topic words, thus eliminating the pre-processing step of filtering infrequent and/or top frequent words, typically required for learning traditional topic models. Experimental results on the 20 Newsgroups and the NIPS datasets show superior performance both on perplexity and topic coherence measure compared to state-of-the-art neural topic models .

Peer-Led Team Learning (PLTL) is a learning methodology where a peer-leader coordinate a small-group of students to collaboratively solve technical problems. PLTL have been adopted for various science, engineering, technology and maths courses in several US universities. This paper proposed and evaluated a speech system for behavioral analysis of PLTL groups. It could help in identifying the best practices for PLTL. The CRSS-PLTL corpus was used for evaluation of developed algorithms. In this paper, we developed a robust speech activity detection (SAD) by fusing the outputs of a DNN-based pitch extractor and an unsupervised SAD based on voicing measures. Robust speaker diarization system consisted of bottleneck features (from stacked autoencoder) and informed HMM-based joint segmentation and clustering system. Behavioral characteristics such as participation, dominance, emphasis, curiosity and engagement were extracted by acoustic analyses of speech segments belonging to all students. We proposed a novel method for detecting question inflection and performed equal error rate analysis on PLTL corpus. In addition, a robust approach for detecting emphasized speech regions was also proposed. Further, we performed exploratory data analysis for understanding the distortion present in CRSS-PLTL corpus as it was collected in naturalistic scenario. The ground-truth Likert scale ratings were used for capturing the team dynamics in terms of student's responses to a variety of evaluation questions. Results suggested the applicability of proposed system for behavioral analysis of small-group conversations such as PLTL, workplace meetings etc..

Accurate segmentation of brain tumor is a critical component for diagnosis of cancer, treatment and evaluation of outcome. It consist of identification of different types of tumor tissues from normal brain MRI images. Recently, pathway CNNs have been used for semantic segmentation, however are computationally expensive. Build upon success of SegNet, in this paper, we presented different architectures of SegNeT encoder and decoder based on pixel-wise classification. Nonlinear up sampling are performed by the model. The end to end training and small number of parameters used for the training makes the computational process more higher than other deep learning architectures. We performed the semantic segmentation on the MRI brain tumor Figshare dataset and achieved the state of the arts results (99.93% global accuracy) in comparison to traditional CNN models.

Lesions that appear hyperintense in both Fluid Attenuated Inversion Recovery (FLAIR) and T2-weighted magnetic resonance images (MRIs) of the human brain are common in the brains of the elderly population and may be caused by ischemia or demyelination. Lesions are biomarkers for various neurodegenerative diseases, making accurate quantification of them important for both disease diagnosis and progression. Automatic lesion detection using supervised learning requires manually annotated images, which can often be impractical to acquire. Unsupervised lesion detection, on the other hand, does not require any manual delineation; however, these methods can be challenging to construct due to the variability in lesion load, placement of lesions, and voxel intensities. Here we present a novel approach to address this problem using a convolutional autoencoder, which learns to segment brain lesions as well as the white matter, gray matter, and cerebrospinal fluid by reconstructing FLAIR images as conical combinations of softmax layer outputs generated from the corresponding T1, T2, and FLAIR images. Some of the advantages of this model are that it accurately learns to segment lesions regardless of lesion load, and it can be used to quickly and robustly segment new images that were not in the training set. Comparisons with state-of-the-art segmentation methods evaluated on ground truth manual labels indicate that the proposed method works well for generating accurate lesion segmentations without the need for manual annotations.

The interconnected and heterogeneous nature of the next-generation Electrical Grid (EG), widely known as Smart Grid (SG), bring severe cybersecurity and privacy risks that can also raise domino effects against other Critical Infrastructures (CIs). In this paper, we present an Intrusion Detection System (IDS) specially designed for the SG environments that use Modbus/Transmission Control Protocol (TCP) and Distributed Network Protocol 3 (DNP3) protocols. The proposed IDS called MENSA (anoMaly dEtection aNd claSsificAtion) adopts a novel Autoencoder-Generative Adversarial Network (GAN) architecture for (a) detecting operational anomalies and (b) classifying Modbus/TCP and DNP3 cyberattacks. In particular, MENSA combines the aforementioned Deep Neural Networks (DNNs) in a common architecture, taking into account the adversarial loss and the reconstruction difference. The proposed IDS is validated in four real SG evaluation environments, namely (a) SG lab, (b) substation, (c) hydropower plant and (d) power plant, solving successfully an outlier detection (i.e., anomaly detection) problem as well as a challenging multiclass classification problem consisting of 14 classes (13 Modbus/TCP cyberattacks and normal instances). Furthermore, MENSA can discriminate five cyberattacks against DNP3. The evaluation results demonstrate the efficiency of MENSA compared to other Machine Learning (ML) and Deep Learning (DL) methods in terms of Accuracy, False Positive Rate (FPR), True Positive Rate (TPR) and the F1 score.

Sparse events, such as malign attacks in real-time network traffic, have caused big organisations an immense hike in revenue loss. This is due to the excessive growth of the network and its exposure to a plethora of people. The standard methods used to detect intrusions are not promising and have significant failure to identify new malware. Moreover, the challenges in handling high volume data with sparsity, high false positives, fewer detection rates in minor class, training time and feature engineering of the dimensionality of data has promoted deep learning to take over the task with less time and great results. The existing system needs improvement in solving real-time network traffic issues along with feature engineering. Our proposed work overcomes these challenges by giving promising results using deep-stacked autoencoders in two stages. The two-stage deep learning combines with shallow learning using the random forest for classification in the second stage. This made the mod...

Uncertainties in machine learning are a significant roadblock for its application in safety-critical cyber-physical systems (CPS). One source of uncertainty arises from distribution shifts in the input data between training and test scenarios. Detecting such distribution shifts in real-time is an emerging approach to address the challenge. The high dimensional input space in CPS applications involving imaging adds extra difficulty to the task. Generative learning models are widely adopted for the task, namely out-of-distribution (OoD) detection. To improve the state-of-the-art, we studied existing proposals from both machine learning and CPS fields. In the latter, safety monitoring in real-time for autonomous driving agents has been a focus. Exploiting the spatiotemporal correlation of motion in videos, we can robustly detect hazardous motion around autonomous driving agents. Inspired by the latest advances in the Variational Autoencoder (VAE) theory and practice, we tapped into the...