ALBERT

Description: Assistive technologies (ATs) often have a high-dimensionality of possible movements (e.g., assistive robot with several degrees of freedom or a computer), but the users have to control them with low-dimensionality sensors and interfaces (e.g., switches). This paper presents the development of an open-source interface based on a sequence-matching algorithm for the control of ATs. Sequence matching allows the user to input several different commands with low-dimensionality sensors by not only recognizing their output, but also their sequential pattern through time, similarly to Morse code. In this paper, the algorithm is applied to the recognition of hand gestures, inputted using an inertial measurement unit worn by the user. An SVM-based algorithm, that is aimed to be robust, with small training sets (e.g., five examples per class) is developed to recognize gestures in real-time. Finally, the interface is applied to control a computer’s mouse and keyboard. The interface was compared against (and combined with) the head movement-based AssystMouse software. The hand gesture interface showed encouraging results for this application but could also be used with other body parts (e.g., head and feet) and could control various ATs (e.g., assistive robotic arm and prosthesis).

Description: Light curve analysis usually involves extracting manually designed features associated with physical parameters and visual inspection. The large amount of data collected nowadays in astronomy by different surveys represents a major challenge of characterizing these signals. Therefore, finding good informative representation for them is a key non-trivial task. Some studies have tried unsupervised machine learning approaches to generate this representation without much effectiveness. In this article, we show that variational auto-encoders can learn these representations by taking the difference between successive timestamps as an additional input. We present two versions of such auto-encoders: Variational Recurrent Auto-Encoder plus time (VRAEt) and re-Scaling Variational Recurrent Auto Encoder plus time (S-VRAEt). The objective is to achieve the most likely low-dimensional representation of the time series that matched latent variables and, in order to reconstruct it, should compactly contain the pattern information. In addition, the S-VRAEt embeds the re-scaling preprocessing of the time series into the model in order to use the Flux standard deviation in the learning of the light curves structure. To assess our approach, we used the largest transit light curve dataset obtained during the 4 years of the Kepler mission and compared to similar techniques in signal processing and light curves. The results show that the proposed methods obtain improvements in terms of the quality of the deep representation of phase-folded transit light curves with respect to their deterministic counterparts. Specifically, they present a good balance between the reconstruction task and the smoothness of the curve, validated with the root mean squared error, mean absolute error, and auto-correlation metrics. Furthermore, there was a good disentanglement in the representation, as validated by the Pearson correlation and mutual information metrics. Finally, a useful representation to distinguish categories was validated with the F1 score in the task of classifying exoplanets. Moreover, the S-VRAEt model increases all the advantages of VRAEt, achieving a classification performance quite close to its maximum model capacity and generating light curves that are visually comparable to a Mandel–Agol fit. Thus, the proposed methods present a new way of analyzing and characterizing light curves.

Description: We propose an instance-based learning approach with data augmentation and similarity evaluation to estimate the remaining useful life (RUL) of a mechanical component for health management. The publicly available PRONOSTIA datasets, which provide accelerated degradation test data for bearings, are used in our study. The challenges with the datasets include a very limited number of run-to-failure examples, no failure mode information, and a wide range of bearing life spans. Without a large number of training samples, feature engineering is necessary. Principal component analysis is applied to the spectrogram of vibration signals to obtain prognostic feature sequences. A data augmentation strategy is developed to generate synthetic prognostic feature sequences using learning instances. Subsequently, similarities between the test and learning instances can be assessed using a root mean squared (RMS) difference measure. Finally, an ensemble method is developed to aggregate the RUL estimates based on multiple similar prognostic feature sequences. The proposed approach demonstrates comparable performance with published solutions in the literature. It serves as an alternative method for solving the RUL estimation problem.

Description: We propose a method for the blind separation of sounds of musical instruments in audio signals. We describe the individual tones via a parametric model, training a dictionary to capture the relative amplitudes of the harmonics. The model parameters are predicted via a U-Net, which is a type of deep neural network. The network is trained without ground truth information, based on the difference between the model prediction and the individual time frames of the short-time Fourier transform. Since some of the model parameters do not yield a useful backpropagation gradient, we model them stochastically and employ the policy gradient instead. To provide phase information and account for inaccuracies in the dictionary-based representation, we also let the network output a direct prediction, which we then use to resynthesize the audio signals for the individual instruments. Due to the flexibility of the neural network, inharmonicity can be incorporated seamlessly and no preprocessing of the input spectra is required. Our algorithm yields high-quality separation results with particularly low interference on a variety of different audio samples, both acoustic and synthetic, provided that the sample contains enough data for the training and that the spectral characteristics of the musical instruments are sufficiently stable to be approximated by the dictionary.

Description: Nowadays, time, scope and cost constraints along with knowledge requirements and personnel training constitute blocking restrictions for effective Offensive Cyberspace Operations (OCO). This paper presents RedHerd, an open-source, collaborative and serverless orchestration framework that overcomes these limitations. RedHerd leverages the ‘as a Service’ paradigm in order to seamlessly deploy a ready-to-use infrastructure that can be also adopted for effective simulation and training purposes, by reliably reproducing a real-world cyberspace battlefield in which red and blue teams can challenge each other. We discuss both the design and implementation of the proposed solution, by focusing on its main functionality, as well as by highlighting how it perfectly fits the Open Systems Architecture design pattern, thanks to the adoption of both open standards and wide-spread open-source software components. The paper also presents a complete OCO simulation based on the usage of RedHerd to perform a fictitious attack and fully compromise an imaginary enterprise following the Cyber Kill Chain (CKC) phases.

Description: With power designers always demanding for faster power switches, electromagnetic interference has become an issue of primary concern. As known, the commutation of power transistors is the main cause of the electromagnetic noise, which can be worsened by the presence of unwanted oscillations superimposed onto the switching waveforms. This work proposes a solution to mitigate the oscillations caused by the turn-on of a power transistor by exploiting its source inductance plus an external one. In this context, an optimization method is proposed to find the optimal value of the source inductance as a trade-off between oscillation damping and power dissipation. The experimental results performed on a prototyped power converter assess the proposed technique as the spectrum of the conducted emission is attenuated by 20 dB at the oscillation frequency. With respect to traditional solution based on snubbers, the proposed solution results in a similar oscillation damping, but with a 0.5% higher power efficiency.

Description: Most IoT networks implement one-way messages from the sensor nodes to the “application host server” via a gateway. Messages from any sensor node in the network are sent when its sensor is triggered or at regular intervals as dictated by the application, such as a Smart-City deployment of LoRaWAN traps/sensors for rat detection. However, these traps can, due to the nature of this application, be moved out of signal range from their original location, or obstructed by objects, resulting in under 69% of the messages reaching the gateway. Therefore, applications of this type would benefit from control messages from the “application host server” back to the sensor nodes for enhancing their performance/connectivity. This paper has implemented a cloud-based AI engine, as part of the “application host server”, that dynamically analyses all received messages from the sensor nodes and exchanges data/enhancement back and forth with them, when necessary. Hundreds of sensor nodes in various blocked/obstructed IoT network connectivity scenarios are used to test our DXN solution. We achieved 100% reporting success if access to any blocked sensor node was possible via a neighbouring node. DXN is based on DNN and Time Series models.

Description: Generative Adversarial Networks (GANs) have gained significant attention in recent years, with impressive applications highlighted in computer vision, in particular. Compared to such examples, however, there have been more limited applications of GANs to time series modeling, including forecasting. In this work, we present the Mixture Density Conditional Generative Adversarial Model (MD-CGAN), with a focus on time series forecasting. We show that our model is capable of estimating a probabilistic posterior distribution over forecasts and that, in comparison to a set of benchmark methods, the MD-CGAN model performs well, particularly in situations where noise is a significant component of the observed time series. Further, by using a Gaussian mixture model as the output distribution, MD-CGAN offers posterior predictions that are non-Gaussian.

Description: This paper proposes an automatic image correction method for portrait photographs, which promotes consistency of facial skin color by suppressing skin color changes due to background colors. In portrait photographs, skin color is often distorted due to the lighting environment (e.g., light reflected from a colored background wall and over-exposure by a camera strobe). This color distortion is emphasized when artificially synthesized with another background color, and the appearance becomes unnatural. In our framework, we, first, roughly extract the face region and rectify the skin color distribution in a color space. Then, we perform color and brightness correction around the face in the original image to achieve a proper color balance of the facial image, which is not affected by luminance and background colors. Our color correction process attains natural results by using a guide image, unlike conventional algorithms. In particular, our guided image filtering for the color correction does not require a perfectly-aligned guide image required in the original guide image filtering method proposed by He et al. Experimental results show that our method generates more natural results than conventional methods on not only headshot photographs but also natural scene photographs. We also show automatic yearbook style photo generation as another application.

Description: This paper describes an automatic drum transcription (ADT) method that directly estimates a tatum-level drum score from a music signal in contrast to most conventional ADT methods that estimate the frame-level onset probabilities of drums. To estimate a tatum-level score, we propose a deep transcription model that consists of a frame-level encoder for extracting the latent features from a music signal and a tatum-level decoder for estimating a drum score from the latent features pooled at the tatum level. To capture the global repetitive structure of drum scores, which is difficult to learn with a recurrent neural network (RNN), we introduce a self-attention mechanism with tatum-synchronous positional encoding into the decoder. To mitigate the difficulty of training the self-attention-based model from an insufficient amount of paired data and to improve the musical naturalness of the estimated scores, we propose a regularized training method that uses a global structure-aware masked language (score) model with a self-attention mechanism pretrained from an extensive collection of drum scores. The experimental results showed that the proposed regularized model outperformed the conventional RNN-based model in terms of the tatum-level error rate and the frame-level F-measure, even when only a limited amount of paired data was available so that the non-regularized model underperformed the RNN-based model.