### Ground Roll Suppression using Convolutional Neural Networks

Seismic data processing plays a major role in seismic exploration as it conditions much of the seismic interpretation performance. In this context, generating reliable post-stack seismic data depends also on disposing of an efficient pre-stack noise attenuation tool. Here we tackle ground roll noise, one of the most challenging and common noises observed in pre-stack seismic data. Since ground roll is characterized by relative low frequencies and high amplitudes, most commonly used approaches for its suppression are based on frequency-amplitude filters for ground roll characteristic bands. However, when signal and noise share the same frequency ranges, these methods usually deliver also signal suppression or residual noise. In this paper we take advantage of the highly non-linear features of convolutional neural networks, and propose to use different architectures to detect ground roll in shot gathers and ultimately to suppress them using conditional generative adversarial networks. Additionally, we propose metrics to evaluate ground roll suppression, and report strong results compared to expert filtering. Finally, we discuss generalization of trained models for similar and different geologies to better understand the feasibility of our proposal in real applications.

### Safety-Aware Cascade Controller Tuning Using Constrained Bayesian Optimization

This paper presents an automated, model-free, data-driven method for the safe tuning of PID cascade controller gains based on Bayesian optimization. The optimization objective is composed of data-driven performance metrics and modeled using Gaussian processes. We further introduce a data-driven constraint that captures the stability requirements from system data. Numerical evaluation shows that the proposed approach outperforms relay feedback autotuning and quickly converges to the global optimum, thanks to a tailored stopping criterion. We demonstrate the performance of the method in simulations and experiments on a linear axis drive of a grinding machine. For experimental implementation, in addition to the introduced safety constraint, we integrate a method for automatic detection of the critical gains and extend the optimization objective with a penalty depending on the proximity of the current candidate points to the critical gains. The resulting automated tuning method optimizes system performance while ensuring stability and standardization.

### Accurate Prostate Cancer Detection and Segmentation on Biparametric MRI using Non-local Mask R-CNN with Histopathological Ground Truth

Purpose: We aimed to develop deep machine learning (DL) models to improve the detection and segmentation of intraprostatic lesions (IL) on bp-MRI by using whole amount prostatectomy specimen-based delineations. We also aimed to investigate whether transfer learning and self-training would improve results with small amount labelled data. Methods: 158 patients had suspicious lesions delineated on MRI based on bp-MRI, 64 patients had ILs delineated on MRI based on whole mount prostatectomy specimen sections, 40 patients were unlabelled. A non-local Mask R-CNN was proposed to improve the segmentation accuracy. Transfer learning was investigated by fine-tuning a model trained using MRI-based delineations with prostatectomy-based delineations. Two label selection strategies were investigated in self-training. The performance of models was evaluated by 3D detection rate, dice similarity coefficient (DSC), 95 percentile Hausdrauff (95 HD, mm) and true positive ratio (TPR). Results: With prostatectomy-based delineations, the non-local Mask R-CNN with fine-tuning and self-training significantly improved all evaluation metrics. For the model with the highest detection rate and DSC, 80.5% (33/41) of lesions in all Gleason Grade Groups (GGG) were detected with DSC of 0.548[0.165], 95 HD of 5.72[3.17] and TPR of 0.613[0.193]. Among them, 94.7% (18/19) of lesions with GGG > 2 were detected with DSC of 0.604[0.135], 95 HD of 6.26[3.44] and TPR of 0.580[0.190]. Conclusion: DL models can achieve high prostate cancer detection and segmentation accuracy on bp-MRI based on annotations from histologic images. To further improve the performance, more data with annotations of both MRI and whole amount prostatectomy specimens are required.

### Cloud-Based Dynamic Programming for an Electric City Bus Energy Management Considering Real-Time Passenger Load Prediction

Electric city bus gains popularity in recent years for its low greenhouse gas emission, low noise level, etc. Different from a passenger car, the weight of a city bus varies significantly with different amounts of onboard passengers, which is not well studied in existing literature. This study proposes a passenger load prediction model using day-of-week, time-of-day, weather, temperatures, wind levels, and holiday information as inputs. The average model, Regression Tree, Gradient Boost Decision Tree, and Neural Networks models are compared in the passenger load prediction. The Gradient Boost Decision Tree model is selected due to its best accuracy and high stability. Given the predicted passenger load, dynamic programming algorithm determines the optimal power demand for supercapacitor and battery by optimizing the battery aging and energy usage in the cloud. Then rule extraction is conducted on dynamic programming results, and the rule is real-time loaded to onboard controllers of vehicles. The proposed cloud-based dynamic programming and rule extraction framework with the passenger load prediction shows 4% and 11% fewer bus operating costs in off-peak and peak hours, respectively. The operating cost by the proposed framework is less than 1% shy of the dynamic programming with the true passenger load information.

### GloFlow: Global Image Alignment for Creation of Whole Slide Images for Pathology from Video

The application of deep learning to pathology assumes the existence of digital whole slide images of pathology slides. However, slide digitization is bottlenecked by the high cost of precise motor stages in slide scanners that are needed for position information used for slide stitching. We propose GloFlow, a two-stage method for creating a whole slide image using optical flow-based image registration with global alignment using a computationally tractable graph-pruning approach. In the first stage, we train an optical flow predictor to predict pairwise translations between successive video frames to approximate a stitch. In the second stage, this approximate stitch is used to create a neighborhood graph to produce a corrected stitch. On a simulated dataset of video scans of WSIs, we find that our method outperforms known approaches to slide-stitching, and stitches WSIs resembling those produced by slide scanners.

### ACCDOA: Activity-Coupled Cartesian Direction of Arrival Representation for Sound Event Localization and Detection

Neural-network (NN)-based methods show high performance in sound event localization and detection (SELD). Conventional NN-based methods use two branches for a sound event detection (SED) target and a direction-of-arrival (DOA) target. The two-branch representation with a single network has to decide how to balance the two objectives during optimization. Using two networks dedicated to each task increases system complexity and network size. To address these problems, we propose an activity-coupled Cartesian DOA (ACCDOA) representation, which assigns a sound event activity to the length of a corresponding Cartesian DOA vector. The ACCDOA representation enables us to solve a SELD task with a single target and has two advantages: avoiding the necessity of balancing the objectives and model size increase. In experimental evaluations with the DCASE 2020 Task 3 dataset, the ACCDOA representation outperformed the two-branch representation in SELD metrics with a smaller network size. The ACCDOA-based SELD system also performed better than state-of-the-art SELD systems in terms of localization and location-dependent detection.

### DeviceTTS: A Small-Footprint, Fast, Stable Network for On-Device Text-to-Speech

With the number of smart devices increasing, the demand for on-device text-to-speech (TTS) increases rapidly. In recent years, many prominent End-to-End TTS methods have been proposed, and have greatly improved the quality of synthesized speech. However, to ensure the qualified speech, most TTS systems depend on large and complex neural network models, and it's hard to deploy these TTS systems on-device. In this paper, a small-footprint, fast, stable network for on-device TTS is proposed, named as DeviceTTS. DeviceTTS makes use of a duration predictor as a bridge between encoder and decoder so as to avoid the problem of words skipping and repeating in Tacotron. As we all know, model size is a key factor for on-device TTS. For DeviceTTS, Deep Feedforward Sequential Memory Network (DFSMN) is used as the basic component. Moreover, to speed up inference, mix-resolution decoder is proposed for balance the inference speed and speech quality. Experiences are done with WORLD and LPCNet vocoder. Finally, with only 1.4 million model parameters and 0.099 GFLOPS, DeviceTTS achieves comparable performance with Tacotron and FastSpeech. As far as we know, the DeviceTTS can meet the needs of most of the devices in practical application.

### A New "Model-Free" Method Combined with Neural Network for MIMO Systems

In this brief, a model-free adaptive predictive control (MFAPC) is proposed. It outperforms the current model-free adaptive control (MFAC) for not only solving the time delay problem in multiple-input multiple-output (MIMO) systems but also relaxing the current rigorous assumptions for sake of a wider applicable range. The most attractive merit of the proposed controller is that the controller design, performance analysis and applications are easy for engineers to realize. Furthermore, the problem of how to choose the matrix {\lambda} is finished by analyzing the function of the closed-loop poles rather than the previous contraction mapping method. Additionally, in view of the nonlinear modeling capability and adaptability of neural networks (NNs), we combine these two classes of algorithms together. The feasibility and several interesting results of the proposed method are shown in simulations.

### Advanced Biophysical Model to Capture Channel Variability for EQS Capacitive HBC

Human Body Communication (HBC) has come up as a promising alternative to traditional radio frequency (RF) Wireless Body Area Network (WBAN) technologies. This is essentially due to HBC providing a broadband communication channel with enhanced signal security in the physical layer due to lower radiation from the human body as compared to its RF counterparts. An in-depth understanding of the mechanism for the channel loss variability and associated biophysical model needs to be developed before EQS-HBC can be used more frequently in WBAN consumer and medical applications. Biophysical models characterizing the human body as a communication channel didn't exist in literature for a long time. Recent developments have shown models that capture the channel response for fixed transmitter and receiver positions on the human body. These biophysical models do not capture the variability in the HBC channel for varying positions of the devices with respect to the human body. In this study, we provide a detailed analysis of the change in path loss in a capacitive-HBC channel in the electroquasistatic (EQS) domain. Causes of channel loss variability namely: inter-device coupling and effects of fringe fields due to body's shadowing effects are investigated. FEM based simulation results are used to analyze the channel response of human body for different positions and sizes of the device which are further verified using measurement results to validate the developed biophysical model. Using the bio-physical model, we develop a closed form equation for the path loss in a capacitive HBC channel which is then analyzed as a function of the geometric properties of the device and the position with respect to the human body which will help pave the path towards future EQSHBC WBAN design.

### Distance Invariant Sparse Autoencoder for Wireless Signal Strength Mapping

Wireless signal strength based localization can enable robust localization for robots using inexpensive sensors. For this, a location-to-signal-strength map has to be learned for each access point in the environment. Due to the ubiquity of Wireless networks in most environments, this can result in tens or hundreds of maps. To reduce the dimensionality of this problem, we employ autoencoders, which are a popular unsupervised approach for feature extraction and data compression. In particular, we propose the use of sparse autoencoders that learn latent spaces that preserve the relative distance between inputs. Distance invariance between input and latent spaces allows our system to successfully learn compact representations that allow precise data reconstruction but also have a low impact on localization performance when using maps from the latent space rather than the input space. We demonstrate the feasibility of our approach by performing experiments in outdoor environments.

### An automated and multi-parametric algorithm for objective analysis of meibography images

Meibography is a non-contact imaging technique used by ophthalmologists to assist in the evaluation and diagnosis of meibomian gland dysfunction (MGD). While artificial qualitative analysis of meibography images could lead to low repeatability and efficiency and multi-parametric analysis is demanding to offer more comprehensive information in discovering subtle changes of meibomian glands during MGD progression, we developed an automated and multi-parametric algorithm for objective and quantitative analysis of meibography images. The full architecture of the algorithm can be divided into three steps: (1) segmentation of the tarsal conjunctiva area as the region of interest (ROI); (2) segmentation and identification of glands within the ROI; and (3) quantitative multi-parametric analysis including newly defined gland diameter deformation index (DI), gland tortuosity index (TI), and glands signal index (SI). To evaluate the performance of the automated algorithm, the similarity index (k) and the segmentation error including the false positive rate (r_P) and the false negative rate (r_N) are calculated between the manually defined ground truth and the automatic segmentations of both the ROI and meibomian glands of 15 typical meibography images. The feasibility of the algorithm is demonstrated in analyzing typical meibograhy images.

### Solving Sparse Linear Inverse Problems in Communication Systems: A Deep Learning Approach With Adaptive Depth

Sparse signal recovery problems from noisy linear measurements appear in many areas of wireless communications. In recent years, deep learning (DL) based approaches have attracted interests of researchers to solve the sparse linear inverse problem by unfolding iterative algorithms as neural networks. Typically, research concerning DL assume a fixed number of network layers. However, it ignores a key character in traditional iterative algorithms, where the number of iterations required for convergence changes with varying sparsity levels. By investigating on the projected gradient descent, we unveil the drawbacks of the existing DL methods with fixed depth. Then we propose an end-to-end trainable DL architecture, which involves an extra halting score at each layer. Therefore, the proposed method learns how many layers to execute to emit an output, and the network depth is dynamically adjusted for each task in the inference phase. We conduct experiments using both synthetic data and applications including random access in massive MTC and massive MIMO channel estimation, and the results demonstrate the improved efficiency for the proposed approach.

### ProCAN: Progressive Growing Channel Attentive Non-Local Network for Lung Nodule Classification

Lung cancer classification in screening computed tomography (CT) scans is one of the most crucial tasks for early detection of this disease. Many lives can be saved if we are able to accurately classify malignant/ cancerous lung nodules. Consequently, several deep learning based models have been proposed recently to classify lung nodules as malignant or benign. Nevertheless, the large variation in the size and heterogeneous appearance of the nodules makes this task an extremely challenging one. We propose a new Progressive Growing Channel Attentive Non-Local (ProCAN) network for lung nodule classification. The proposed method addresses this challenge from three different aspects. First, we enrich the Non-Local network by adding channel-wise attention capability to it. Second, we apply Curriculum Learning principles, whereby we first train our model on easy examples before hard/ difficult ones. Third, as the classification task gets harder during the Curriculum learning, our model is progressively grown to increase its capability of handling the task at hand. We examined our proposed method on two different public datasets and compared its performance with state-of-the-art methods in the literature. The results show that the ProCAN model outperforms state-of-the-art methods and achieves an AUC of 98.05% and accuracy of 95.28% on the LIDC-IDRI dataset. Moreover, we conducted extensive ablation studies to analyze the contribution and effects of each new component of our proposed method.

### Novel Digital Camera with the PCIe Interface

Digital cameras are commonly used for diagnostic purposes in large-scale physics experiments. A typical image diagnostic system consists of an optical setup, digital camera, frame grabber, image processing CPU, and data analysis tool. The standard architecture of the imaging system has a number of disadvantages. Data transmitted from a camera are buffered multiple times and must be converted between various protocols before they are finally transmitted to the host memory. Such an architecture makes the system quite complicated, limits its performance and, in consequence, increases its price. The limitations are even more critical for control or protection systems operating in real-time. Modern megapixel cameras generate large data throughput, easily exceeding 10 Gb/s, which often requires some additional processing on the host side. The optimal system architecture should assure low overhead and high performance of the data transmission and processing. It is particularly important during the processing of data streams from several imaging devices, which can be as high as several terabits per second. A novel architecture of image acquisition and processing system based on the PCI Express interface was proposed to meet the requirements of real-time imaging systems applied in large-scale physics experiments. The architecture allows to transfer an image stream directly from the camera to the data processing unit and therefore significantly decreases the overhead and improves performance.

### FlatNet: Towards Photorealistic Scene Reconstruction from Lensless Measurements

Lensless imaging has emerged as a potential solution towards realizing ultra-miniature cameras by eschewing the bulky lens in a traditional camera. Without a focusing lens, the lensless cameras rely on computational algorithms to recover the scenes from multiplexed measurements. However, the current iterative-optimization-based reconstruction algorithms produce noisier and perceptually poorer images. In this work, we propose a non-iterative deep learning based reconstruction approach that results in orders of magnitude improvement in image quality for lensless reconstructions. Our approach, called $\textit{FlatNet}$, lays down a framework for reconstructing high-quality photorealistic images from mask-based lensless cameras, where the camera's forward model formulation is known. FlatNet consists of two stages: (1) an inversion stage that maps the measurement into a space of intermediate reconstruction by learning parameters within the forward model formulation, and (2) a perceptual enhancement stage that improves the perceptual quality of this intermediate reconstruction. These stages are trained together in an end-to-end manner. We show high-quality reconstructions by performing extensive experiments on real and challenging scenes using two different types of lensless prototypes: one which uses a separable forward model and another, which uses a more general non-separable cropped-convolution model. Our end-to-end approach is fast, produces photorealistic reconstructions, and is easy to adopt for other mask-based lensless cameras.

### Progressive Voice Trigger Detection: Accuracy vs Latency

We present an architecture for voice trigger detection for virtual assistants. The main idea in this work is to exploit information in words that immediately follow the trigger phrase. We first demonstrate that by including more audio context after a detected trigger phrase, we can indeed get a more accurate decision. However, waiting to listen to more audio each time incurs a latency increase. Progressive Voice Trigger Detection allows us to trade-off latency and accuracy by accepting clear trigger candidates quickly, but waiting for more context to decide whether to accept more marginal examples. Using a two-stage architecture, we show that by delaying the decision for just 3% of detected true triggers in the test set, we are able to obtain a relative improvement of 66% in false rejection rate, while incurring only a negligible increase in latency.

### A Novel Fast 3D Single Image Super-Resolution Algorithm

This paper introduces a novel computationally efficient method of solving the 3D single image super-resolution (SR) problem, i.e., reconstruction of a high-resolution volume from its low-resolution counterpart. The main contribution lies in the original way of handling simultaneously the associated decimation and blurring operators, based on their underlying properties in the frequency domain. In particular, the proposed decomposition technique of the 3D decimation operator allows a straightforward implementation for Tikhonov regularization, and can be further used to take into consideration other regularization functions such as the total variation, enabling the computational cost of state-of-the-art algorithms to be considerably decreased. Numerical experiments carried out showed that the proposed approach outperforms existing 3D SR methods.

### Experimental validation of MDL emulation and estimation techniques for SDM transmission systems

We experimentally validate a mode-dependent loss (MDL) estimation technique employing acorrection factor to remove the MDL estimation dependence on the SNR when using a minimum meansquare error (MMSE) equalizer. A reduction of the MDL estimation error is observed for both transmitter-side and in-span MDL emulation.

### 1 Tbit/s/$λ$ Transmission Over a 130 km Link Consisting of Graded-Index 50 $μ$m Core Multi-Mode Fiber and 6LP Few-Mode Fiber

We demonstrate 1 Tbit/s/$\lambda$ single-span transmission over a heterogeneous link consisting of graded-index 50 $\mu$m core multi-mode fiber and 6LP few-mode fiber using a Kramers-Kronig receiver structure. Furthermore, the link budget increase by transmitting only three modes while employing more than three receivers is investigated.

### FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement

This paper proposes a full-band and sub-band fusion model, named as FullSubNet, for single-channel real-time speech enhancement. Full-band and sub-band refer to the models that input full-band and sub-band noisy spectral feature, output full-band and sub-band speech target, respectively. The sub-band model processes each frequency independently. Its input consists of one frequency and several context frequencies. The output is the prediction of the clean speech target for the corresponding frequency. These two types of models have distinct characteristics. The full-band model can capture the global spectral context and the long-distance cross-band dependencies. However, it lacks the ability to modeling signal stationarity and attending the local spectral pattern. The sub-band model is just the opposite. In our proposed FullSubNet, we connect a pure full-band model and a pure sub-band model sequentially and use practical joint training to integrate these two types of models' advantages. We conducted experiments on the DNS challenge (INTERSPEECH 2020) dataset to evaluate the proposed method. Experimental results show that full-band and sub-band information are complementary, and the FullSubNet can effectively integrate them. Besides, the performance of the FullSubNet also exceeds that of the top-ranked methods in the DNS Challenge (INTERSPEECH 2020).

### UNetGAN: A Robust Speech Enhancement Approach in Time Domain for Extremely Low Signal-to-noise Ratio Condition

Speech enhancement at extremely low signal-to-noise ratio (SNR) condition is a very challenging problem and rarely investigated in previous works. This paper proposes a robust speech enhancement approach (UNetGAN) based on U-Net and generative adversarial learning to deal with this problem. This approach consists of a generator network and a discriminator network, which operate directly in the time domain. The generator network adopts a U-Net like structure and employs dilated convolution in the bottleneck of it. We evaluate the performance of the UNetGAN at low SNR conditions (up to -20dB) on the public benchmark. The result demonstrates that it significantly improves the speech quality and substantially outperforms the representative deep learning models, including SEGAN, cGAN fo SE, Bidirectional LSTM using phase-sensitive spectrum approximation cost function (PSA-BLSTM) and Wave-U-Net regarding Short-Time Objective Intelligibility (STOI) and Perceptual evaluation of speech quality (PESQ).

### A comparison of automatic multi-tissue segmentation methods of the human fetal brain using the FeTA Dataset

It is critical to quantitatively analyse the developing human fetal brain in order to fully understand neurodevelopment in both normal fetuses and those with congenital disorders. To facilitate this analysis, automatic multi-tissue fetal brain segmentation algorithms are needed, which in turn requires open databases of segmented fetal brains. Here we introduce a publicly available database of 50 manually segmented pathological and non-pathological fetal magnetic resonance brain volume reconstructions across a range of gestational ages (20 to 33 weeks) into 7 different tissue categories (external cerebrospinal fluid, grey matter, white matter, ventricles, cerebellum, deep grey matter, brainstem/spinal cord). In addition, we quantitatively evaluate the accuracy of several automatic multi-tissue segmentation algorithms of the developing human fetal brain. Four research groups participated, submitting a total of 10 algorithms, demonstrating the benefits the database for the development of automatic algorithms.

### Probabilistic interval predictor based on dissimilarity functions

This work presents a new method to obtain probabilistic interval predictions of a dynamical system. The method uses stored past system measurements to estimate the future evolution of the system. The proposed method relies on the use of dissimilarity functions to estimate the conditional probability density function of the outputs. A family of empirical probability density functions, parameterized by means of two parameters, is introduced. It is shown that the the proposed family encompasses the multivariable normal probability density function as a particular case. We show that the proposed method constitutes a generalization of classical estimation methods. A cross-validation scheme is used to tune the two parameters on which the methodology relies. In order to prove the effectiveness of the methodology presented, some numerical examples and comparisons are provided.

### Coordinated Formation Control for Intelligent and Connected Vehicles in Multiple Traffic Scenarios

In this paper, a unified multi-vehicle formation control framework for Intelligent and Connected Vehicles (ICVs) that can apply to multiple traffic scenarios is proposed. In the one-dimensional scenario, different formation geometries are analyzed and the interlaced structure is mathematically modelized to improve driving safety while making full use of the lane capacity. The assignment problem for vehicles and target positions is solved using Hungarian Algorithm to improve the flexibility of the method in multiple scenarios. In the two-dimensional scenario, an improved virtual platoon method is proposed to transfer the complex two-dimensional passing problem to the one-dimensional formation control problem based on the idea of rotation projection. Besides, the vehicle regrouping method is proposed to connect the two scenarios. Simulation results prove that the proposed multi-vehicle formation control framework can apply to multiple typical scenarios and have better performance than existing methods.

### Genetic U-Net: Automatically Designing Lightweight U-shaped CNN Architectures Using the Genetic Algorithm for Retinal Vessel Segmentation

Many previous works based on deep learning for retinal vessel segmentation have achieved promising performance by manually designing U-shaped convolutional neural networks (CNNs). However, the manual design of these CNNs is time-consuming and requires extensive empirical knowledge. To address this problem, we propose a novel method using genetic algorithms (GAs) to automatically design a lightweight U-shaped CNN for retinal vessel segmentation, called Genetic U-Net. Here we first design a special search space containing the structure of U-Net and its corresponding operations, and then use genetic algorithm to search for superior architectures in this search space. Experimental results show that the proposed method outperforms the existing methods on three public datasets, DRIVE, CHASE_DB1 and STARE. In addition, the architectures obtained by the proposed method are more lightweight but robust than the state-of-the-art models.

### Sampling and Reconstruction of Sparse Signals in Shift-Invariant Spaces: Generalized Shannon's Theorem Meets Compressive Sensing

This paper introduces a novel framework and corresponding methods for sampling and reconstruction of sparse signals in shift-invariant (SI) spaces. We reinterpret the random demodulator, a system that acquires sparse bandlimited signals, as a system for acquisition of linear combinations of the samples in the SI setting with the box function as the sampling kernel. The sparsity assumption is exploited by compressive sensing (CS) framework for recovery of the SI samples from a reduced set of measurements. The samples are subsequently filtered by a discrete-time correction filter in order to reconstruct expansion coefficients of an observed signal. Furthermore, we offer a generalization of the proposed framework to other sampling kernels that lie in arbitrary SI spaces. The generalized method embeds the correction filter in a CS optimization problem which directly reconstructs expansion coefficients of the signal. Both approaches recast an inherently infinite-dimensional inverse problem as a finite-dimensional CS problem in an exact way. Finally, we conduct numerical experiments on signals in B-spline spaces whose expansion coefficients are assumed to be sparse in a certain transform domain. The coefficients can be regarded as parametric models of an underlying continuous signal, obtained from a reduced set of measurements. Such continuous signal representations are particularly suitable for signal processing without converting them into samples.

### One-Bit Direct Position Determination of Narrowband Gaussian Signals

One of the main drawbacks of the well-known Direct Position Determination (DPD) method is the requirement that raw signal data be transferred to a common processor. It would therefore be of high practical value if DPD$-$or a modified version thereof$-$could be successfully applied to a coarsely quantized version of the raw data, thus alleviating the requirements on the communication links between the different base stations. Motivated by the above, and inspired by recent work in the rejuvenated one-bit array processing field, we present One-Bit DPD: a method for direct localization based on one-bit quantized measurements. We show that despite the coarse quantization, the proposed method nonetheless yields an estimate for the unknown emitter position with appealing asymptotic properties. We further establish the underlying identifiability conditions of this model, which rely only on second-order statistics. Empirical simulation results corroborate our analytical derivations, demonstrating that much of the information regarding the unknown emitter position is preserved under this crude form of quantization.

### Brain Tumor Segmentation Network Using Attention-based Fusion and Spatial Relationship Constraint

Delineating the brain tumor from magnetic resonance (MR) images is critical for the treatment of gliomas. However, automatic delineation is challenging due to the complex appearance and ambiguous outlines of tumors. Considering that multi-modal MR images can reflect different tumor biological properties, we develop a novel multi-modal tumor segmentation network (MMTSN) to robustly segment brain tumors based on multi-modal MR images. The MMTSN is composed of three sub-branches and a main branch. Specifically, the sub-branches are used to capture different tumor features from multi-modal images, while in the main branch, we design a spatial-channel fusion block (SCFB) to effectively aggregate multi-modal features. Additionally, inspired by the fact that the spatial relationship between sub-regions of tumor is relatively fixed, e.g., the enhancing tumor is always in the tumor core, we propose a spatial loss to constrain the relationship between different sub-regions of tumor. We evaluate our method on the test set of multi-modal brain tumor segmentation challenge 2020 (BraTs2020). The method achieves 0.8764, 0.8243 and 0.773 dice score for whole tumor, tumor core and enhancing tumor, respectively.

### Identification of complex mixtures for Raman spectroscopy using a novel scheme based on a new multi-label deep neural network

With noisy environment caused by fluoresence and additive white noise as well as complicated spectrum fingerprints, the identification of complex mixture materials remains a major challenge in Raman spectroscopy application. In this paper, we propose a new scheme based on a constant wavelet transform (CWT) and a deep network for classifying complex mixture. The scheme first transforms the noisy Raman spectrum to a two-dimensional scale map using CWT. A multi-label deep neural network model (MDNN) is then applied for classifying material. The proposed model accelerates the feature extraction and expands the feature graph using the global averaging pooling layer. The Sigmoid function is implemented in the last layer of the model. The MDNN model was trained, validated and tested with data collected from the samples prepared from substances in palm oil. During training and validating process, data augmentation is applied to overcome the imbalance of data and enrich the diversity of Raman spectra. From the test results, it is found that the MDNN model outperforms previously proposed deep neural network models in terms of Hamming loss, one error, coverage, ranking loss, average precision, F1 macro averaging and F1 micro averaging, respectively. The average detection time obtained from our model is 5.31 s, which is much faster than the detection time of the previously proposed models.

### FD Cell-Free mMIMO: Analysis and Optimization

We consider a full-duplex cell-free massive multiple-input-multiple-output system with limited capacity fronthaul links. We derive its downlink/uplink closed-form spectral efficiency (SE) lower bounds with maximum-ratio transmission/maximum-ratio combining and optimal uniform quantization. To reduce carbon footprint, this paper maximizes the non-convex weighted sum energy efficiency (WSEE) via downlink and uplink power control, and successive convex approximation framework. We show that with low fronthaul capacity, the system requires a higher number of fronthaul quantization bits to achieve high SE and WSEE. For high fronthaul capacity, higher number of bits, however, achieves high SE but a reduced WSEE.

### Maximum a posteriori signal recovery for optical coherence tomography angiography image generation and denoising

Optical coherence tomography angiography (OCTA) is a novel and clinically promising imaging modality to image retinal and sub-retinal vasculature. Based on repeated optical coherence tomography (OCT) scans, intensity changes are observed over time and used to compute OCTA image data. OCTA data are prone to noise and artifacts caused by variations in flow speed and patient movement. We propose a novel iterative maximum a posteriori signal recovery algorithm in order to generate OCTA volumes with reduced noise and increased image quality. This algorithm is based on previous work on probabilistic OCTA signal models and maximum likelihood estimates. Reconstruction results using total variation minimization and wavelet shrinkage for regularization were compared against an OCTA ground truth volume, merged from six co-registered single OCTA volumes. The results show a significant improvement in peak signal-to-noise ratio and structural similarity. The presented algorithm brings together OCTA image generation and Bayesian statistics and can be developed into new OCTA image generation and denoising algorithms.

### Resilient Energy Efficient Healthcare Monitoring Infrastructure with Server and Network Protection

In this paper, a 1+1 server protection scheme is considered where two servers, a primary and a secondary processing server are used to serve ECG monitoring applications concurrently. The infrastructure is designed to be resilient against server failure under two scenarios related to the geographic location of primary and secondary servers and resilient against both server and network failures. A Mixed Integer Linear Programming (MILP) model is used to optimise the number and locations of both primary and secondary processing servers so that the energy consumption of the networking equipment and processing are minimised. The results show that considering a scenario for server protection without geographical constraints compared to the non-resilient scenario has resulted in both network and processing energy penalty as the traffic is doubled. The results also reveal that increasing the level of resilience to consider geographical constraints compared to case without geographical constraints resulted in higher network energy penalty when the demand is low as more nodes are utilised to place the processing servers under the geographic constraints. Also, increasing the level of resilience to consider network protection with link and node disjoint selection has resulted in a low network energy penalty at high demands due to the activation of a large part of the network in any case due to the demands. However, the results show that the network energy penalty is reduced with the increasing number of processing servers at each candidate node. Meanwhile, the same energy for processing is consumed regardless of the increasing level of resilience as the same number of processing servers are utilised. A heuristic is developed for each resilience scenario for real-time implementation where the results show that the performance of the heuristic is approaching the results of the MILP model.

### Deep Autofocus for Synthetic Aperture Sonar

Synthetic aperture sonar (SAS) requires precise positional and environmental information to produce well-focused output during the image reconstruction step. However, errors in these measurements are commonly present resulting in defocused imagery. To overcome these issues, an \emph{autofocus} algorithm is employed as a post-processing step after image reconstruction for the purpose of improving image quality using the image content itself. These algorithms are usually iterative and metric-based in that they seek to optimize an image sharpness metric. In this letter, we demonstrate the potential of machine learning, specifically deep learning, to address the autofocus problem. We formulate the problem as a self-supervised, phase error estimation task using a deep network we call Deep Autofocus. Our formulation has the advantages of being non-iterative (and thus fast) and not requiring ground truth focused-defocused images pairs as often required by other deblurring deep learning methods. We compare our technique against a set of common sharpness metrics optimized using gradient descent over a real-world dataset. Our results demonstrate Deep Autofocus can produce imagery that is perceptually as good as benchmark iterative techniques but at a substantially lower computational cost. We conclude that our proposed Deep Autofocus can provide a more favorable cost-quality trade-off than state-of-the-art alternatives with significant potential of future research.

### Quickest detection of false data injection attack in remote state estimation

In this paper, quickest detection of false data injection attack on remote state estimation is considered. A set of $N$ sensors make noisy linear observations of a discrete-time linear process with Gaussian noise, and report the observations to a remote estimator. The challenge is the presence of a few potentially malicious sensors which can start strategically manipulating their observations at a random time in order to skew the estimates. The quickest attack detection problem for a known linear attack scheme is posed as a constrained Markov decision process in order to minimise the expected detection delay subject to a false alarm constraint, with the state involving the probability belief at the estimator that the system is under attack. State transition probabilities are derived in terms of system parameters, and the structure of the optimal policy is derived analytically. It turns out that the optimal policy amounts to checking whether the probability belief exceeds a threshold. Numerical results demonstrate significant performance gain under the proposed algorithm against competing algorithms.

### Raw Audio for Depression Detection Can Be More Robust Against Gender Imbalance than Mel-Spectrogram Features

Depression is a large-scale mental health problem and a challenging area for machine learning researchers in terms of the detection of depression. Datasets such as the Distress Analysis Interview Corpus - Wizard of Oz have been created to aid research in this area. However, on top of the challenges inherent in accurately detecting depression, biases in datasets may result in skewed classification performance. In this paper we examine gender bias in the DAIC-WOZ dataset using audio-based deep neural networks. We show that gender biases in DAIC-WOZ can lead to an overreporting of performance, which has been overlooked in the past due to the same gender biases being present in the test set. By using raw audio and different concepts from Fair Machine Learning, such as data re-distribution, we can mitigate against the harmful effects of bias.

### On the Optimality and Convergence Properties of the Learning Model Predictive Controller

In this technical note we analyse the performance improvement and optimality properties of the Learning Model Predictive Control (LMPC) strategy for linear deterministic systems. The LMPC framework is a policy iteration scheme where closed-loop trajectories are used to update the control policy for the next execution of the control task. We show that, when a Linear Independence Constraint Qualification (LICQ) condition holds, the LMPC scheme guarantees strict iterative performance improvement and optimality, meaning that the closed-loop cost evaluated over the entire task converges asymptotically to the optimal cost of the infinite-horizon control problem. Compared to previous works this sufficient LICQ condition can be easily checked, it holds for a larger class of systems and it can be used to adaptively select the prediction horizon of the controller, as demonstrated by a numerical example.

### Improving Perceptual Quality by Phone-Fortified Perceptual Loss for Speech Enhancement

Speech enhancement (SE) aims to improve speech quality and intelligibility, which are both related to a smooth transition in speech segments that may carry linguistic information, e.g. phones and syllables. In this study, we took phonetic characteristics into account in the SE training process. Hence, we designed a phone-fortified perceptual (PFP) loss, and the training of our SE model was guided by PFP loss. In PFP loss, phonetic characteristics are extracted by wav2vec, an unsupervised learning model based on the contrastive predictive coding (CPC) criterion. Different from previous deep-feature-based approaches, the proposed approach explicitly uses the phonetic information in the deep feature extraction process to guide the SE model training. To test the proposed approach, we first confirmed that the wav2vec representations carried clear phonetic information using a t-distributed stochastic neighbor embedding (t-SNE) analysis. Next, we observed that the proposed PFP loss was more strongly correlated with the perceptual evaluation metrics than point-wise and signal-level losses, thus achieving higher scores for standardized quality and intelligibility evaluation metrics in the Voice Bank--DEMAND dataset.

### Inference of ventricular activation properties from non-invasive electrocardiography

The realisation of precision cardiology requires novel techniques for the non-invasive characterisation of individual patients' cardiac function to inform therapeutic and diagnostic decision-making. The electrocardiogram (ECG) is the most widely used clinical tool for cardiac diagnosis. Its interpretation is, however, confounded by functional and anatomical variability in heart and torso. In this study, we develop new computational techniques to estimate key ventricular activation properties for individual subjects by exploiting the synergy between non-invasive electrocardiography and image-based torso-biventricular modelling and simulation. More precisely, we present an efficient sequential Monte Carlo approximate Bayesian computation-based inference method, integrated with Eikonal simulations and torso-biventricular models constructed based on clinical cardiac magnetic resonance (CMR) imaging. The method also includes a novel strategy to treat combined continuous (conduction speeds) and discrete (earliest activation sites) parameter spaces, and an efficient dynamic time warping-based ECG comparison algorithm. We demonstrate results from our inference method on a cohort of twenty virtual subjects with cardiac volumes ranging from 74 cm3 to 171 cm3 and considering low versus high resolution for the endocardial discretisation (which determines possible locations of the earliest activation sites). Results show that our method can successfully infer the ventricular activation properties from non-invasive data, with higher accuracy for earliest activation sites, endocardial speed, and sheet (transmural) speed in sinus rhythm, rather than the fibre or sheet-normal speeds.

### Semantic video segmentation for autonomous driving

We aim to solve semantic video segmentation in autonomous driving, namely road detection in real time video, using techniques discussed in (Shelhamer et al., 2016a). While fully convolutional network gives good result, we show that the speed can be halved while preserving the accuracy. The test dataset being used is KITTI, which consists of real footage from Germany's streets.

### DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality metric to evaluate Noise Suppressors

Human subjective evaluation is the gold standard to evaluate speech quality optimized for human perception. Perceptual objective metrics serve as a proxy for subjective scores. The conventional and widely used metrics require a reference clean speech signal, which is unavailable in real recordings. The no-reference approaches correlate poorly with human ratings and are not widely adopted in the research community. One of the biggest use cases of these perceptual objective metrics is to evaluate noise suppression algorithms. This paper introduces a multi-stage self-teaching based perceptual objective metric that is designed to evaluate noise suppressors. The proposed method generalizes well in challenging test conditions with a high correlation to human ratings.

### Object sieving and morphological closing to reduce false detections in wide-area aerial imagery

For object detection in wide-area aerial imagery, post-processing is usually needed to reduce false detections. We propose a two-stage post-processing scheme which comprises an area-thresholding sieving process and a morphological closing operation. We use two wide-area aerial videos to compare the performance of five object detection algorithms in the absence and in the presence of our post-processing scheme. The automatic detection results are compared with the ground-truth objects. Several metrics are used for performance comparison.

### Representation learning for improved interpretability and classification accuracy of clinical factors from EEG

Despite extensive standardization, diagnostic interviews for mental health disorders encompass substantial subjective judgment. Previous studies have demonstrated that EEG-based neural measures can function as reliable objective correlates of depression, or even predictors of depression and its course. However, their clinical utility has not been fully realized because of 1) the lack of automated ways to deal with the inherent noise associated with EEG data at scale, and 2) the lack of knowledge of which aspects of the EEG signal may be markers of a clinical disorder. Here we adapt an unsupervised pipeline from the recent deep representation learning literature to address these problems by 1) learning a disentangled representation using $\beta$-VAE to denoise the signal, and 2) extracting interpretable features associated with a sparse set of clinical labels using a Symbol-Concept Association Network (SCAN). We demonstrate that our method is able to outperform the canonical hand-engineered baseline classification method on a number of factors, including participant age and depression diagnosis. Furthermore, our method recovers a representation that can be used to automatically extract denoised Event Related Potentials (ERPs) from novel, single EEG trajectories, and supports fast supervised re-mapping to various clinical labels, allowing clinicians to re-use a single EEG representation regardless of updates to the standardized diagnostic system. Finally, single factors of the learned disentangled representations often correspond to meaningful markers of clinical factors, as automatically detected by SCAN, allowing for human interpretability and post-hoc expert analysis of the recommendations made by the model.

### Point Cloud Attribute Compression via Successive Subspace Graph Transform

Inspired by the recently proposed successive subspace learning (SSL) principles, we develop a successive subspace graph transform (SSGT) to address point cloud attribute compression in this work. The octree geometry structure is utilized to partition the point cloud, where every node of the octree represents a point cloud subspace with a certain spatial size. We design a weighted graph with self-loop to describe the subspace and define a graph Fourier transform based on the normalized graph Laplacian. The transforms are applied to large point clouds from the leaf nodes to the root node of the octree recursively, while the represented subspace is expanded from the smallest one to the whole point cloud successively. It is shown by experimental results that the proposed SSGT method offers better R-D performances than the previous Region Adaptive Haar Transform (RAHT) method.

### Exploring Generative Adversarial Networks for Image-to-Image Translation in STEM Simulation

The use of accurate scanning transmission electron microscopy (STEM) image simulation methods require large computation times that can make their use infeasible for the simulation of many images. Other simulation methods based on linear imaging models, such as the convolution method, are much faster but are too inaccurate to be used in application. In this paper, we explore deep learning models that attempt to translate a STEM image produced by the convolution method to a prediction of the high accuracy multislice image. We then compare our results to those of regression methods. We find that using the deep learning model Generative Adversarial Network (GAN) provides us with the best results and performs at a similar accuracy level to previous regression models on the same dataset. Codes and data for this project can be found in this GitHub repository, https://github.com/uw-cmg/GAN-STEM-Conv2MultiSlice.

### The IQIYI System for Voice Conversion Challenge 2020

This paper presents the IQIYI voice conversion system (T24) for Voice Conversion 2020. In the competition, each target speaker has 70 sentences. We have built an end-to-end voice conversion system based on PPG. First, the ASR acoustic model calculates the BN feature, which represents the content-related information in the speech. Then the Mel feature is calculated through an improved prosody tacotron model. Finally, the Mel spectrum is converted to wav through an improved LPCNet. The evaluation results show that this system can achieve better voice conversion effects. In the case of using 16k rather than 24k sampling rate audio, the conversion result is relatively good in naturalness and similarity. Among them, our best results are in the similarity evaluation of the Task 2, the 2nd in the ASV-based objective evaluation and the 5th in the subjective evaluation.

### Improvement of EAST Data Acquisition Configuration Management

The data acquisition console is an important component of the EAST data acquisition system which provides unified data acquisition and long-term data storage for diagnostics. The data acquisition console is used to manage the data acquisition configuration information and control the data acquisition workflow. The data acquisition console has been developed many years, and with increasing of data acquisition nodes and emergence of new control nodes, the function of configuration management has become inadequate. It is going to update the configuration management function of data acquisition console. The upgraded data acquisition console based on LabVIEW should be oriented to the data acquisition administrator, with the functions of managing data acquisition nodes, managing control nodes, setting and publishing configuration parameters, batch management, database backup, monitoring the status of data acquisition nodes, controlling the data acquisition workflow, and shot simulation data acquisition test. The upgraded data acquisition console has been designed and under testing recently.

### Identifying safe intersection design through unsupervised feature extraction from satellite imagery

The World Health Organization has listed the design of safer intersections as a key intervention to reduce global road trauma. This article presents the first study to systematically analyze the design of all intersections in a large country, based on aerial imagery and deep learning. Approximately 900,000 satellite images were downloaded for all intersections in Australia and customized computer vision techniques emphasized the road infrastructure. A deep autoencoder extracted high-level features, including the intersection's type, size, shape, lane markings, and complexity, which were used to cluster similar designs. An Australian telematics data set linked infrastructure design to driving behaviors captured during 66 million kilometers of driving. This showed more frequent hard acceleration events (per vehicle) at four- than three-way intersections, relatively low hard deceleration frequencies at T-intersections, and consistently low average speeds on roundabouts. Overall, domain-specific feature extraction enabled the identification of infrastructure improvements that could result in safer driving behaviors, potentially reducing road trauma.

### Sea-Net: Squeeze-And-Excitation Attention Net For Diabetic Retinopathy Grading

Diabetes is one of the most common disease in individuals. \textit{Diabetic retinopathy} (DR) is a complication of diabetes, which could lead to blindness. Automatic DR grading based on retinal images provides a great diagnostic and prognostic value for treatment planning. However, the subtle differences among severity levels make it difficult to capture important features using conventional methods. To alleviate the problems, a new deep learning architecture for robust DR grading is proposed, referred to as SEA-Net, in which, spatial attention and channel attention are alternatively carried out and boosted with each other, improving the classification performance. In addition, a hybrid loss function is proposed to further maximize the inter-class distance and reduce the intra-class variability. Experimental results have shown the effectiveness of the proposed architecture.

### Self-supervised Pre-training Reduces Label Permutation Instability of Speech Separation

Speech separation has been well-developed while there are still problems waiting to be solved. The main problem we focus on in this paper is the frequent label permutation switching of permutation invariant training (PIT). For N-speaker separation, there would be N! possible label permutations. How to stably select correct label permutations is a long-standing problem. In this paper, we utilize self-supervised pre-training to stabilize the label permutations. Among several types of self-supervised tasks, speech enhancement based pre-training tasks show significant effectiveness in our experiments. When using off-the-shelf pre-trained models, training duration could be shortened to one-third to two-thirds. Furthermore, even taking pre-training time into account, the entire training process could still be shorter without a performance drop when using a larger batch size.

### Learning Audio Embeddings with User Listening Data for Content-based Music Recommendation

Personalized recommendation on new track releases has always been a challenging problem in the music industry. To combat this problem, we first explore user listening history and demographics to construct a user embedding representing the user's music preference. With the user embedding and audio data from user's liked and disliked tracks, an audio embedding can be obtained for each track using metric learning with Siamese networks. For a new track, we can decide the best group of users to recommend by computing the similarity between the track's audio embedding and different user embeddings, respectively. The proposed system yields state-of-the-art performance on content-based music recommendation tested with millions of users and tracks. Also, we extract audio embeddings as features for music genre classification tasks. The results show the generalization ability of our audio embeddings.

### Channel Estimation and Equalization for CP-OFDM-based OTFS in Fractional Doppler Channels

Orthogonal time frequency and space (OTFS) modulation is a promising technology that satisfies high Doppler requirements for future mobile systems. OTFS modulation encodes information symbols and pilot symbols into the two-dimensional (2D) delay-Doppler (DD) domain. The received symbols suffer from inter-Doppler interference (IDI) in the fading channels with fractional Doppler shifts that are sampled at noninteger indices in the DD domain. IDI has been treated as an unavoidable effect because the fractional Doppler shifts cannot be obtained directly from the received pilot symbols. In this paper, we provide a solution to channel estimation for fractional Doppler channels. The proposed estimation provides new insight into the OTFS input-output relation in the DD domain as a 2D circular convolution with a small approximation. According to the input-output relation, we also provide a low-complexity channel equalization method using the estimated channel information. We demonstrate the error performance of the proposed channel estimation and equalization in several channels by simulations. The simulation results show that in high-mobility environments, the total system utilizing the proposed methods outperforms orthogonal frequency division multiplexing (OFDM) with ideal channel estimation and a conventional channel estimation method using a pseudo sequence.

### Modeling and Control of COVID-19 Epidemic through Testing Policies

Testing for the infected cases is one of the most important mechanisms to control an epidemic. It enables to isolate the detected infected individuals, thereby limiting the disease transmission to the susceptible population. However, despite the significance of testing policies, the recent literature on the subject lacks a control-theoretic perspective. In this work, an epidemic model that incorporates the testing rate as a control input is presented. The proposed model differentiates the undetected infected from the detected infected cases, who are assumed to be removed from the disease spreading process in the population. First, the model is estimated and validated for COVID-19 data in France. Then, two testing policies are proposed, the so-called best-effort strategy for testing (BEST) and constant optimal strategy for testing (COST). The BEST policy is a suppression strategy that provides a lower bound on the testing rate such that the epidemic switches from a spreading to a non-spreading state. The COST policy is a mitigation strategy that provides an optimal value of testing rate that minimizes the peak value of the infected population when the total stockpile of tests is limited. Both testing policies are evaluated by predicting the number of active intensive care unit (ICU) cases and the cumulative number of deaths due to COVID-19.

### Self-awareness in intelligent vehicles: Feature based dynamic Bayesian models for abnormality detection

The evolution of Intelligent Transportation Systems in recent times necessitates the development of self-awareness in agents. Before the intensive use of Machine Learning, the detection of abnormalities was manually programmed by checking every variable and creating huge nested conditions that are very difficult to track. This paper aims to introduce a novel method to develop self-awareness in autonomous vehicles that mainly focuses on detecting abnormal situations around the considered agents. Multi-sensory time-series data from the vehicles are used to develop the data-driven Dynamic Bayesian Network (DBN) models used for future state prediction and the detection of dynamic abnormalities. Moreover, an initial level collective awareness model that can perform joint anomaly detection in co-operative tasks is proposed. The GNG algorithm learns the DBN models' discrete node variables; probabilistic transition links connect the node variables. A Markov Jump Particle Filter (MJPF) is applied to predict future states and detect when the vehicle is potentially misbehaving using learned DBNs as filter parameters. In this paper, datasets from real experiments of autonomous vehicles performing various tasks used to learn and test a set of switching DBN models.

### Beyond cross-entropy: learning highly separable feature distributions for robust and accurate classification

Deep learning has shown outstanding performance in several applications including image classification. However, deep classifiers are known to be highly vulnerable to adversarial attacks, in that a minor perturbation of the input can easily lead to an error. Providing robustness to adversarial attacks is a very challenging task especially in problems involving a large number of classes, as it typically comes at the expense of an accuracy decrease. In this work, we propose the Gaussian class-conditional simplex (GCCS) loss: a novel approach for training deep robust multiclass classifiers that provides adversarial robustness while at the same time achieving or even surpassing the classification accuracy of state-of-the-art methods. Differently from other frameworks, the proposed method learns a mapping of the input classes onto target distributions in a latent space such that the classes are linearly separable. Instead of maximizing the likelihood of target labels for individual samples, our objective function pushes the network to produce feature distributions yielding high inter-class separation. The mean values of the distributions are centered on the vertices of a simplex such that each class is at the same distance from every other class. We show that the regularization of the latent space based on our approach yields excellent classification accuracy and inherently provides robustness to multiple adversarial attacks, both targeted and untargeted, outperforming state-of-the-art approaches over challenging datasets.

### Modulation Pattern Detection Using Complex Convolutions in Deep Learning

Transceivers used for telecommunications transmit and receive specific modulation patterns that are represented as sequences of complex numbers. Classifying modulation patterns is challenging because noise and channel impairments affect the signals in complicated ways such that the received signal bears little resemblance to the transmitted signal. Although deep learning approaches have shown great promise over statistical methods in this problem space, deep learning frameworks continue to lag in support for complex-valued data. To address this gap, we study the implementation and use of complex convolutions in a series of convolutional neural network architectures. Replacement of data structure and convolution operations by their complex generalization in an architecture improves performance, with statistical significance, at recognizing modulation patterns in complex-valued signals with high SNR after being trained on low SNR signals. This suggests complex-valued convolutions enables networks to learn more meaningful representations. We investigate this hypothesis by comparing the features learned in each experiment by visualizing the inputs that results in one-hot modulation pattern classification for each network.

### Modeling biomedical breathing signals with convolutional deep probabilistic autoencoders

One of the main problems with biomedical signals is the limited amount of patient-specific data and the significant amount of time needed to record a sufficient number of samples for diagnostic and treatment purposes. We explore the use of Variational Autoencoder (VAE) and Adversarial Autoencoder (AAE) algorithms based on one-dimensional convolutional neural networks in order to build generative models able to capture and represent the variability of a set of unlabeled quasi-periodic signals using as few as 10 parameters. Furthermore, we introduce a modified AAE architecture that allows simultaneous semi-supervised classification and generation of different types of signals. Our study is based on physical breathing signals, i.e. time series describing the position of chest markers, generally used to describe respiratory motion. The time series are discretized into a vector of periods, with each period containing 6 time and position values. These vectors can be transformed back into time series through an additional reconstruction neural network and allow to generate extended signals while simplifying the modeling task. The obtained models can be used to generate realistic breathing realizations from patient or population data and to classify new recordings. We show that by incorporating the labels from around 10-15\% of the dataset during training, the model can be guided to group data according to the patient it belongs to, or based on the presence of different types of breathing irregularities such as baseline shifts. Our specific motivation is to model breathing motion during radiotherapy lung cancer treatments, for which the developed model serves as an efficient tool to robustify plans against breathing uncertainties. However, the same methodology can in principle be applied to any other kind of quasi-periodic biomedical signal, representing a generically applicable tool.

### Shared Space Transfer Learning for analyzing multi-site fMRI data

Multi-voxel pattern analysis (MVPA) learns predictive models from task-based functional magnetic resonance imaging (fMRI) data, for distinguishing when subjects are performing different cognitive tasks -- e.g., watching movies or making decisions. MVPA works best with a well-designed feature set and an adequate sample size. However, most fMRI datasets are noisy, high-dimensional, expensive to collect, and with small sample sizes. Further, training a robust, generalized predictive model that can analyze homogeneous cognitive tasks provided by multi-site fMRI datasets has additional challenges. This paper proposes the Shared Space Transfer Learning (SSTL) as a novel transfer learning (TL) approach that can functionally align homogeneous multi-site fMRI datasets, and so improve the prediction performance in every site. SSTL first extracts a set of common features for all subjects in each site. It then uses TL to map these site-specific features to a site-independent shared space in order to improve the performance of the MVPA. SSTL uses a scalable optimization procedure that works effectively for high-dimensional fMRI datasets. The optimization procedure extracts the common features for each site by using a single-iteration algorithm and maps these site-specific common features to the site-independent shared space. We evaluate the effectiveness of the proposed method for transferring between various cognitive tasks. Our comprehensive experiments validate that SSTL achieves superior performance to other state-of-the-art analysis techniques.

### Expert Selection in High-Dimensional Markov Decision Processes

In this work we present a multi-armed bandit framework for online expert selection in Markov decision processes and demonstrate its use in high-dimensional settings. Our method takes a set of candidate expert policies and switches between them to rapidly identify the best performing expert using a variant of the classical upper confidence bound algorithm, thus ensuring low regret in the overall performance of the system. This is useful in applications where several expert policies may be available, and one needs to be selected at run-time for the underlying environment.

### Manifold learning-based feature extraction for structural defect reconstruction

Data-driven quantitative defect reconstructions using ultrasonic guided waves has recently demonstrated great potential in the area of non-destructive testing. In this paper, we develop an efficient deep learning-based defect reconstruction framework, called NetInv, which recasts the inverse guided wave scattering problem as a data-driven supervised learning progress that realizes a mapping between reflection coefficients in wavenumber domain and defect profiles in the spatial domain. The superiorities of the proposed NetInv over conventional reconstruction methods for defect reconstruction have been demonstrated by several examples. Results show that NetInv has the ability to achieve the higher quality of defect profiles with remarkable efficiency and provides valuable insight into the development of effective data driven structural health monitoring and defect reconstruction using machine learning.

### Semi-Supervised Speech Recognition via Graph-based Temporal Classification

Semi-supervised learning has demonstrated promising results in automatic speech recognition (ASR) by self-training using a seed ASR model with pseudo-labels generated for unlabeled data. The effectiveness of this approach largely relies on the pseudo-label accuracy, for which typically only the 1-best ASR hypothesis is used. However, alternative ASR hypotheses of an N-best list can provide more accurate labels for an unlabeled speech utterance and also reflect uncertainties of the seed ASR model. In this paper, we propose a generalized form of the connectionist temporal classification (CTC) objective that accepts a graph representation of the training targets. The newly proposed graph-based temporal classification (GTC) objective is applied for self-training with WFST-based supervision, which is generated from an N-best list of pseudo-labels. In this setup, GTC is used to learn not only a temporal alignment, similarly to CTC, but also a label alignment to obtain the optimal pseudo-label sequence from the weighted graph. Results show that this approach can effectively exploit an N-best list of pseudo-labels with associated scores, outperforming standard pseudo-labeling by a large margin, with ASR results close to an oracle experiment in which the best hypotheses of the N-best lists are selected manually.

### Playing a Part: Speaker Verification at the Movies

The goal of this work is to investigate the performance of popular speaker recognition models on speech segments from movies, where often actors intentionally disguise their voice to play a character. We make the following three contributions: (i) We collect a novel, challenging speaker recognition dataset called VoxMovies, with speech for 856 identities from almost 4000 movie clips. VoxMovies contains utterances with varying emotion, accents and background noise, and therefore comprises an entirely different domain to the interview-style, emotionally calm utterances in current speaker recognition datasets such as VoxCeleb; (ii) We provide a number of domain adaptation evaluation sets, and benchmark the performance of state-of-the-art speaker recognition models on these evaluation pairs. We demonstrate that both speaker verification and identification performance drops steeply on this new data, showing the challenge in transferring models across domains; and finally (iii) We show that simple domain adaptation paradigms improve performance, but there is still large room for improvement.

### What can we learn from gradients?

Recent work (\cite{zhu2019deep}) has shown that it is possible to reconstruct the input (image) from the gradient of a neural network. In this paper, our aim is to better understand the limits to reconstruction and to speed up image reconstruction by imposing prior image information and improved initialization. Firstly, we show that for the \textbf{non-linear} neural network, gradient-based reconstruction approximates to solving a high-dimension \textbf{linear} equations for both fully-connected neural network and convolutional neural network. Exploring the theoretical limits of input reconstruction, we show that a fully-connected neural network with a \textbf{one} hidden node is enough to reconstruct a \textbf{single} input image, regardless of the number of nodes in the output layer. Then we generalize this result to a gradient averaged over mini-batches of size B. In this case, the full mini-batch can be reconstructed in a fully-connected network if the number of hidden units exceeds B. For a convolutional neural network, the required number of filters in the first convolutional layer again is decided by the batch size B, however, in this case, input width d and the width after filter $d^{'}$ also play the role $h=(\frac{d}{d^{'}})^2BC$, where C is channel number of input. Finally, we validate and underpin our theoretical analysis on bio-medical data (fMRI, ECG signals, and cell images) and on benchmark data (MNIST, CIFAR100, and face images).

### Recurrent Neural Networks for video object detection

There is lots of scientific work about object detection in images. For many applications like for example autonomous driving the actual data on which classification has to be done are videos. This work compares different methods, especially those which use Recurrent Neural Networks to detect objects in videos. We differ between feature-based methods, which feed feature maps of different frames into the recurrent units, box-level methods, which feed bounding boxes with class probabilities into the recurrent units and methods which use flow networks. This study indicates common outcomes of the compared methods like the benefit of including the temporal context into object detection and states conclusions and guidelines for video object detection networks.

### A Helmholtz equation solver using unsupervised learning: Application to transcranial ultrasound

Transcranial ultrasound therapy is increasingly used for the non-invasive treatment of brain disorders. However, conventional numerical wave solvers are currently too computationally expensive to be used online during treatments to predict the acoustic field passing through the skull (e.g., to account for subject-specific dose and targeting variations). As a step towards real-time predictions, in the current work, a fast iterative solver for the heterogeneous Helmholtz equation in 2D is developed using a fully-learned optimizer. The lightweight network architecture is based on a modified UNet that includes a learned hidden state. The network is trained using a physics-based loss function and a set of idealized sound speed distributions with fully unsupervised training (no knowledge of the true solution is required). The learned optimizer shows excellent performance on the test set, and is capable of generalization well outside the training examples, including to much larger computational domains, and more complex source and sound speed distributions, for example, those derived from x-ray computed tomography images of the skull.

### GANs & Reels: Creating Irish Music using a Generative Adversarial Network

In this paper we present a method for algorithmic melody generation using a generative adversarial network without recurrent components. Music generation has been successfully done using recurrent neural networks, where the model learns sequence information that can help create authentic sounding melodies. Here, we use DC-GAN architecture with dilated convolutions and towers to capture sequential information as spatial image information, and learn long-range dependencies in fixed-length melody forms such as Irish traditional reel.

### The ins and outs of speaker recognition: lessons from VoxSRC 2020

The VoxCeleb Speaker Recognition Challenge (VoxSRC) at Interspeech 2020 offers a challenging evaluation for speaker recognition systems, which includes celebrities playing different parts in movies. The goal of this work is robust speaker recognition of utterances recorded in these challenging environments. We utilise variants of the popular ResNet architecture for speaker recognition and perform extensive experiments using a range of loss functions and training parameters. To this end, we optimise an efficient training framework that allows powerful models to be trained with limited time and resources. Our trained models demonstrate improvements over most existing works with lighter models and a simple pipeline. The paper shares the lessons learned from our participation in the challenge.