Search

Due to the increasing size of integrated circuits (ICs), their design and optimization phases (i.e., computer-aided design, CAD) grow increasingly complex. At design time, a large design space needs to be explored to find an implementation that fulfills all specifications and then optimizes metrics like energy, area, delay, reliability, etc. At run time, a large configuration space needs to be searched to find the best set of parameters (e.g., voltage/frequency) to further optimize the system. Both spaces are infeasible for exhaustive search typically leading to heuristic optimization algorithms that find some tradeoff between design quality and computational overhead. Machine learning (ML) can build powerful models that have successfully been employed in related domains. In this survey, we categorize how ML may be used and is used for design-time and run-time optimization and exploration strategies of ICs. A metastudy of published techniques unveils areas in CAD that are well explored and underexplored with ML, as well as trends in the employed ML algorithms. We present a comprehensive categorization and summary of the state of the art on ML for CAD. Finally, we summarize the remaining challenges and promising open research directions. [ABSTRACT FROM AUTHOR]

This paper serves as a survey of recent advances in large margin training and its theoretical foundations, mostly for (nonlinear) deep neural networks (DNNs) that are probably the most prominent machine learning models for large-scale data in the community over the past decade. We generalize the formulation of classification margins from classical research to latest DNNs, summarize theoretical connections between the margin, network generalization, and robustness, and introduce recent efforts in enlarging the margins for DNNs comprehensively. Since the viewpoint of different methods is discrepant, we categorize them into groups for ease of comparison and discussion in the paper. Hopefully, our discussions and overview inspire new research work in the community that aim to improve the performance of DNNs, and we also point to directions where the large margin principle can be verified to provide theoretical evidence why certain regularizations for DNNs function well in practice. We managed to shorten the paper such that the crucial spirit of large margin learning and related methods are better emphasized. [ABSTRACT FROM AUTHOR]

There is an opportunity in modern power systems to explore the demand flexibility by incentivizing consumers with dynamic prices. In this paper, we quantify demand flexibility using an efficient tool called time-varying elasticity, whose value may change depending on the prices and decision dynamics. This tool is particularly useful for evaluating the demand response potential and system reliability. Recent empirical evidences have highlighted some abnormal features when studying demand flexibility, such as delayed responses and vanishing elasticities after price spikes. Existing methods fail to capture these complicated features because they heavily rely on some predefined (often over-simplified) regression expressions. Instead, this paper proposes a model-free methodology to automatically and accurately derive the optimal estimation pattern. We further develop a two-stage estimation process with Siamese long short-term memory (LSTM) networks. Here, a LSTM network encodes the price response, while the other network estimates the time-varying elasticities. In the case study, the proposed framework and models are validated to achieve higher overall estimation accuracy and better description for various abnormal features when compared with the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Correspondence analysis (CA) is a multivariate statistical tool used to visualize and interpret data dependencies by finding maximally correlated embeddings of pairs of random variables. CA has found applications in fields ranging from epidemiology to social sciences. However, current methods for CA do not scale to large, high-dimensional datasets. In this paper, we provide a novel interpretation of CA in terms of an information-theoretic quantity called the principal inertia components. We show that estimating the principal inertia components, which consists in solving a functional optimization problem over the space of finite variance functions of two random variable, is equivalent to performing CA. We then leverage this insight to design algorithms to perform CA at scale. Specifically, we demonstrate how the principal inertia components can be reliably approximated from data using deep neural networks. Finally, we show how the maximally correlated embeddings of pairs of random variables in CA further play a central role in several learning problems including multi-view and multi-modal learning methods and visualization of classification boundaries. [ABSTRACT FROM AUTHOR]

Correspondence analysis (CA) is a multivariate statistical tool used to visualize and interpret data dependencies by finding maximally correlated embeddings of pairs of random variables. CA has found applications in fields ranging from epidemiology to social sciences. However, current methods for CA do not scale to large, high-dimensional datasets. In this paper, we provide a novel interpretation of CA in terms of an information-theoretic quantity called the principal inertia components. We show that estimating the principal inertia components, which consists in solving a functional optimization problem over the space of finite variance functions of two random variable, is equivalent to performing CA. We then leverage this insight to design algorithms to perform CA at scale. Specifically, we demonstrate how the principal inertia components can be reliably approximated from data using deep neural networks. Finally, we show how the maximally correlated embeddings of pairs of random variables in CA further play a central role in several learning problems including multi-view and multi-modal learning methods and visualization of classification boundaries. [ABSTRACT FROM AUTHOR]

Parallelism and data reuse are the most critical issues for the design of hardware acceleration in a deep learning processor. Besides, abundant on-chip memories and precise data management are intrinsic design requirements because most of deep learning algorithms are data-driven and memory-bound. In this paper, we propose a compiler-architecture co-design scheme targeting a reconfigurable and algorithm-oriented array processor, named ReAAP. Given specific deep neural networks, the proposed co-design scheme is effective to perform parallelism and data reuse optimization on compute-intensive layers for guiding reconfigurable computing in hardware. Especially, the systemic optimization is performed in our proposed domain-specific compiler to deal with the intrinsic tensions between parallelism and data locality, for the purpose of automatically mapping diverse layer-level workloads onto our proposed reconfigurable array architecture. In this architecture, abundant on-chip memories are software-controlled and its massive data access is precisely handled by compiler-generated instructions. In our experiments, the ReAAP is implemented on an embedded FPGA platform. Experimental results demonstrate that our proposed co-design scheme is effective to integrate software flexibility with hardware parallelism for accelerating diverse deep learning workloads. As a whole system, ReAAP achieves a consistently high utilization of hardware resource for accelerating all the diverse compute-intensive layers in ResNet, MobileNet, and BERT. [ABSTRACT FROM AUTHOR]

Edge systems are required to autonomously make real-time decisions based on large quantities of input data under strict power, performance, area, and other constraints. Meeting these constraints is only possible by specializing systems through hardware accelerators purposefully built for machine learning and data analysis algorithms. However, data science evolves at a quick pace, and manual design of custom accelerators has high non-recurrent engineering costs: general solutions are needed to automatically and rapidly transition from the formulation of a new algorithm to the deployment of a dedicated hardware implementation. Our solution is the SOftware Defined Architectures (SODA) Synthesizer, an end-to-end, multi-level, modular, extensible compiler toolchain providing a direct path from machine learning tools to hardware. The SODA Synthesizer frontend is based on the multilevel intermediate representation (MLIR) framework; it ingests pre-trained machine learning models, identifies kernels suited for acceleration, performs high-level optimizations, and prepares them for hardware synthesis. In the backend, SODA leverages state-of-the-art high-level synthesis techniques to generate highly efficient accelerators, targeting both field programmable devices (FPGAs) and application-specific circuits (ASICs). In this paper, we describe how the SODA Synthesizer can also assemble the generated accelerators (based on the finite state machine with datapath model) in a custom system driven by a distributed controller, building a coarse-grained dataflow architecture that does not require a host processor to orchestrate parallel execution of multiple accelerators. We show the effectiveness of our approach by automatically generating ASIC accelerators for layers of popular deep neural networks (DNNs). Our high-level optimizations result in up to 74x speedup on isolated accelerators for individual DNN layers, and our dynamically scheduled architecture yields an additional 3x performance improvement when combining accelerators to handle streaming inputs. [ABSTRACT FROM AUTHOR]

Machine Learning (ML) workloads are increasingly deployed at the edge. Enabling efficient inference execution while considering model and system heterogeneity remains challenging, especially for ML tasks built with a network of deep neural networks (DNNs). The challenge is to maximize the utilization of all available resources on the multiprocessor system on a chip (MPSoC) at the same time. This becomes even more complicated because the optimal mapping for the network of DNNs can vary with input batch sizes and scene complexity. In this paper, a holistic hierarchical scheduling framework is presented to optimize the execution time for a network of DNN models on an edge MPSoC at runtime, considering varying input characteristics. The framework consists of a local and a global scheduler. The local scheduler maps individual DNNs in the inference pipeline to the best-performing hardware unit while the global scheduler customizes an Integer Linear Programming (ILP) solution to instantiate DNN remapping. To minimize scheduler runtime overhead, an imitation learning (IL) based scheduler is used that approximates the ILP solutions. The proposed scheduling framework (CAMDNN) was implemented on a Qualcomm Robotic RB5 platform. CAMDNN resulted in lower execution time of up to 32% than heterogeneous earliest finish time, and by factors of 6.67X, 5.6X and 2.17X than the CPU-only, GPU-only and Central Queue schedulers. [ABSTRACT FROM AUTHOR]

Deep neural networks have been tremendously successful at segmenting objects in images. However, it has been shown they still have limitations on challenging problems such as the segmentation of medical images. The main reason behind this lower success resides in the reduced size of the object in the image. In this paper we overcome this limitation through a cyclic collaborative framework, CyCoSeg. The proposed framework is based on a deep active shape model (D-ASM), which provides prior information about the shape of the object, and a semantic segmentation network (SSN). These two models collaborate to reach the desired segmentation by influencing each other: SSN helps D-ASM identify relevant keypoints in the image through an Expectation Maximization formulation, while D-ASM provides a segmentation proposal that guides the SSN. This cycle is repeated until both models converge. Extensive experimental evaluation shows CyCoSeg boosts the performance of the baseline models, including several popular SSNs, while avoiding major architectural modifications. The effectiveness of our method is demonstrated on the left ventricle segmentation on two benchmark datasets, where our approach achieves one of the most competitive results in segmentation accuracy. Furthermore, its generalization is demonstrated for lungs and kidneys segmentation in CT scans. [ABSTRACT FROM AUTHOR]

Classification of 3D objects – the selection of a category in which each object belongs – is of great interest in the field of machine learning. Numerous researchers use deep neural networks to address this problem, altering the network architecture and representation of the 3D shape used as an input. To investigate the effectiveness of their approaches, we conduct an extensive survey of existing methods and identify common ideas by which we categorize them into a taxonomy. Second, we evaluate 11 selected classification networks on two 3D object datasets, extending the evaluation to a larger dataset on which most of the selected approaches have not been tested yet. For this, we provide a framework for converting shapes from common 3D mesh formats into formats native to each network, and for training and evaluating different classification approaches on this data. Despite being partially unable to reach the accuracies reported in the original papers, we compare the relative performance of the approaches as well as their performance when changing datasets as the only variable to provide valuable insights into performance on different kinds of data. We make our code available to simplify running training experiments with multiple neural networks with different prerequisites. [ABSTRACT FROM AUTHOR]

Hyperdimensional computing (HDC) has emerged as an alternative lightweight learning solution to deep neural networks. A key characteristic of HDC is the great extent of parallelism that can facilitate hardware acceleration. However, previous hardware implementations of HDC seldom focus on GPU designs, which were also inefficient partly due to the complexity of accelerating HDC on GPUs. In this paper, we present OpenHD, a flexible and high-performance GPU-powered framework for automating the mapping of general HDC applications including classification and clustering to GPUs. OpenHD takes advantage of memory optimization strategies specialized for HDC, minimizing the access time to different memory subsystems, and removing redundant operations. We also propose a novel training method to enable data parallelism in HDC training. Our evaluation result shows that the proposed training rapidly achieves the target accuracy, reducing the required training epochs by 4×. With OpenHD, users can deploy GPU-accelerated HDC applications without domain expert knowledge. Compared to the state-of-the-art GPU-powered HDC implementation, our evaluation on NVIDIA Jetson TX2 shows that OpenHD is up to 10.5× and 314× faster for HDC-based classification and clustering, respectively. Compared with non-HDC classification and clustering on GPUs, OpenHD-based HDC is 11.7× and 53× faster at comparable accuracy. OpenHD is available at: https://github.com/UCSD-SEELab/openhd. [ABSTRACT FROM AUTHOR]

This paper investigates the stochastic optimization problem focusing on developing scalable parallel algorithms for deep learning tasks. Our solution involves a reformation of the objective function for stochastic optimization in neural network models, along with a novel parallel computing strategy, coined the weighted aggregating stochastic gradient descent (WASGD). Following a theoretical analysis on the characteristics of the new objective function, WASGD introduces a decentralized weighted aggregating scheme based on the performance of local workers. Without any center variable, the new method automatically gauges the importance of local workers and accepts them by their contributions. Furthermore, we have developed an enhanced version of the method, WASGD+, by (1) implementing a designed sample order and (2) upgrading the weight evaluation function. To validate the new method, we benchmark our pipeline against several popular algorithms including the state-of-the-art deep neural network classifier training techniques (e.g., elastic averaging SGD). Comprehensive validation studies have been conducted on four classic datasets: CIFAR-100, CIFAR-10, Fashion-MNIST, and MNIST. Subsequent results have firmly validated the superiority of the WASGD scheme in accelerating the training of deep architecture. Better still, the enhanced version, WASGD+, is shown to be a significant improvement over its prototype. [ABSTRACT FROM AUTHOR]

Graph Neural Networks (GNNs) have led to state-of-the-art performance on a variety of machine learning tasks such as recommendation, node classification and link prediction. Graph neural network models generate node embeddings by merging nodes features with the aggregated neighboring nodes information. Most existing GNN models exploit a single type of aggregator (e.g., mean-pooling) to aggregate neighboring nodes information, and then add or concatenate the output of aggregator to the current representation vector of the center node. However, using only a single type of aggregator is difficult to capture the different aspects of neighboring information and the simple addition or concatenation update methods limit the expressive capability of GNNs. Not only that, existing supervised or semi-supervised GNN models are trained based on the loss function of the node label, which leads to the neglect of graph structure information. In this paper, we propose a novel graph neural network architecture, Graph Attention & Interaction Network (GAIN), for inductive learning on graphs. Unlike the previous GNN models that only utilize a single type of aggregation method, we use multiple types of aggregators to gather neighboring information in different aspects and integrate the outputs of these aggregators through the aggregator-level attention mechanism. Furthermore, we design a graph regularized loss to better capture the topological relationship of the nodes in the graph. Additionally, we first present the concept of graph feature interaction and propose a vector-wise explicit feature interaction mechanism to update the node embeddings. We conduct comprehensive experiments on two node-classification benchmarks and a real-world financial news dataset. The experiments demonstrate our GAIN model outperforms current state-of-the-art performances on all the tasks. [ABSTRACT FROM AUTHOR]

The efficiency of link adaptation in wireless communications relies greatly on the accuracy of channel knowledge and transmission mode selection. In this paper, a novel deep learning based link adaptation framework is proposed for the orthogonal frequency-division multiplexing (OFDM) systems with compressed-sensing-assisted index modulation, termed as OFDM-CSIM, communicating over millimeter-wave (mmWave) channels. To achieve link adaptation, a novel multi-layer sparse Bayesian learning (SBL) algorithm is proposed for accurately and instantaneously providing the required channel state information. Meanwhile, a deep neural networks (DNN)-assisted adaptive modulation algorithm is proposed to choose the best possible transmission mode to maximize the achievable throughput. Simulation results show that the proposed multi-layer SBL algorithm enables more accurate channel estimation than the conventional techniques. The DNN-based adaptive modulator is capable of achieving a higher throughput than the learning-assisted solution based on the $k$ nearest neighbor ($k$ -NN) algorithm, and also the classic average signal-to-noise ratio (SNR)-based solutions. Moreover, analysis shows that both the multi-layer SBL algorithm and the DNN-assisted adaptive modulator achieve better performance than their respective conventional counterparts while at a significantly lower computational complexity cost. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (DCNNs) are currently the method of choice both for generative, as well as for discriminative learning in computer vision and machine learning. The success of DCNNs can be attributed to the careful selection of their building blocks (e.g., residual blocks, rectifiers, sophisticated normalization schemes, to mention but a few). In this paper, we propose $\Pi$ Π -Nets, a new class of function approximators based on polynomial expansions. $\Pi$ Π -Nets are polynomial neural networks, i.e., the output is a high-order polynomial of the input. The unknown parameters, which are naturally represented by high-order tensors, are estimated through a collective tensor factorization with factors sharing. We introduce three tensor decompositions that significantly reduce the number of parameters and show how they can be efficiently implemented by hierarchical neural networks. We empirically demonstrate that $\Pi$ Π -Nets are very expressive and they even produce good results without the use of non-linear activation functions in a large battery of tasks and signals, i.e., images, graphs, and audio. When used in conjunction with activation functions, $\Pi$ Π -Nets produce state-of-the-art results in three challenging tasks, i.e., image generation, face verification and 3D mesh representation learning. The source code is available at https://github.com/grigorisg9gr/polynomial_nets. [ABSTRACT FROM AUTHOR]

With the popularity of location-aware devices (e.g., smart phones), large amounts of location-based social media data such as check-ins are generated. This stimulates plenty of studies for POI recommendations by applying machine learning techniques. However, most of the existing studies focus on POI recommendations in the same city or region, and fail to recommend POIs for users when they travel to a new city. In this paper, we propose a novel deep neural network, named as ST-TransRec, for crossing-city POI recommendations. It integrates the deep neural network, transfer learning technique, and density-based resampling method into a unified framework. In this model, the deep neural network is used to capture users’ preferences for POIs and learn the embeddings of POIs. Besides, the transfer learning technique is employed to bridge the gap between cities that results from the city-dependent features. As the distributions over POIs are imbalanced, we design a density-based spatial resampling model which enables POIs to be well matched across cities. We conduct extensive experiments on two real-world datasets. The experimental results show the advantages of ST-TransRec over the state-of-the-art methods for crossing-city POI recommendations. [ABSTRACT FROM AUTHOR]

With the fast development of various positioning techniques such as Global Position System (GPS), mobile devices and remote sensing, spatio-temporal data has become increasingly available nowadays. Mining valuable knowledge from spatio-temporal data is critically important to many real-world applications including human mobility understanding, smart transportation, urban planning, public safety, health care and environmental management. As the number, volume and resolution of spatio-temporal data increase rapidly, traditional data mining methods, especially statistics-based methods for dealing with such data are becoming overwhelmed. Recently deep learning models such as recurrent neural network (RNN) and convolutional neural network (CNN) have achieved remarkable success in many domains due to the powerful ability in automatic feature representation learning, and are also widely applied in various spatio-temporal data mining (STDM) tasks such as predictive learning, anomaly detection and classification. In this paper, we provide a comprehensive review of recent progress in applying deep learning techniques for STDM. We first categorize the spatio-temporal data into five different types, and then briefly introduce the deep learning models that are widely used in STDM. Next, we classify existing literature based on the types of spatio-temporal data, the data mining tasks, and the deep learning models, followed by the applications of deep learning for STDM in different domains including transportation, on-demand service, climate & weather analysis, human mobility, location-based social network, crime analysis, and neuroscience. Finally, we conclude the limitations of current research and point out future research directions. [ABSTRACT FROM AUTHOR]

Condition monitoring of power transformers, which are key components of electrical power systems, is essential to identify incipient faults and avoid catastrophic failures. In this paper machine learning algorithms, i.e., nonlinear autoregressive neural networks and support vector machines, are proposed to model the transformer thermal behavior for the purpose of monitoring. The thermal models are developed based on the historical measurements from nine transformers comprised of two 180-MVA units, four 240-MVA units and three 1000-MVA units. The data consist of load profile, tap position, winding indicator temperature (WTI) measurement, ambient temperature, wind speed and solar radiation. The results are validated against field measurements, and it is clearly demonstrated that the alternative algorithms surpass the IEEE Annex G thermal model. An incipient thermal fault identification algorithm is then proposed and successfully used to identify an issue using measurements taken in the field. This algorithm could be used to alert the operator and plan intervention accordingly. [ABSTRACT FROM AUTHOR]

A sum-product network (SPN) is a probabilistic model, based on a rooted acyclic directed graph, in which terminal nodes represent probability distributions and non-terminal nodes represent convex sums (weighted averages) and products of probability distributions. They are closely related to probabilistic graphical models, in particular to Bayesian networks with multiple context-specific independencies. Their main advantage is the possibility of building tractable models from data, i.e., models that can perform several inference tasks in time proportional to the number of edges in the graph. They are somewhat similar to neural networks and can address the same kinds of problems, such as image processing and natural language understanding. This paper offers a survey of SPNs, including their definition, the main algorithms for inference and learning from data, several applications, a brief review of software libraries, and a comparison with related models. [ABSTRACT FROM AUTHOR]

In this paper, we focus on X-ray images (X-radiographs) of paintings with concealed sub-surface designs (e.g., deriving from reuse of the painting support or revision of a composition by the artist), which therefore include contributions from both the surface painting and the concealed features. In particular, we propose a self-supervised deep learning-based image separation approach that can be applied to the X-ray images from such paintings to separate them into two hypothetical X-ray images. One of these reconstructed images is related to the X-ray image of the concealed painting, while the second one contains only information related to the X-ray image of the visible painting. The proposed separation network consists of two components: the analysis and the synthesis sub-networks. The analysis sub-network is based on learned coupled iterative shrinkage thresholding algorithms (LCISTA) designed using algorithm unrolling techniques, and the synthesis sub-network consists of several linear mappings. The learning algorithm operates in a totally self-supervised fashion without requiring a sample set that contains both the mixed X-ray images and the separated ones. The proposed method is demonstrated on a real painting with concealed content, Do na Isabel de Porcel by Francisco de Goya, to show its effectiveness. [ABSTRACT FROM AUTHOR]

Unsupervised domain adaptation (UDA) enables a learning machine to adapt from a labeled source domain to an unlabeled target domain under the distribution shift. Thanks to the strong representation ability of deep neural networks, recent remarkable achievements in UDA resort to learning domain-invariant features. Intuitively, the goal is that a good feature representation and the hypothesis learned from the source domain can generalize well to the target domain. However, the learning processes of domain-invariant features and source hypotheses inevitably involve domain-specific information that would degrade the generalizability of UDA models on the target domain. The lottery ticket hypothesis proves that only partial parameters are essential for generalization. Motivated by it, we find in this paper that only partial parameters are essential for learning domain-invariant information. Such parameters are termed transferable parameters that can generalize well in UDA. In contrast, the rest parameters tend to fit domain-specific details and often cause the failure of generalization, which are termed untransferable parameters. Driven by this insight, we propose Transferable Parameter Learning (TransPar) to reduce the side effect of domain-specific information in the learning process and thus enhance the memorization of domain-invariant information. Specifically, according to the distribution discrepancy degree, we divide all parameters into transferable and untransferable ones in each training iteration. We then perform separate update rules for the two types of parameters. Extensive experiments on image classification and regression tasks (keypoint detection) show that TransPar outperforms prior arts by non-trivial margins. Moreover, experiments demonstrate that TransPar can be integrated into the most popular deep UDA networks and be easily extended to handle any data distribution shift scenarios. [ABSTRACT FROM AUTHOR]

Representation learning has been proven to play an important role in the unprecedented success of machine learning models in numerous tasks, such as machine translation, face recognition and recommendation. The majority of existing representation learning approaches often require a large number of consistent and noise-free labels. However, due to various reasons such as budget constraints and privacy concerns, labels are very limited in many real-world scenarios. Directly applying standard representation learning approaches on small labeled data sets will easily run into over-fitting problems and lead to sub-optimal solutions. Even worse, in some domains such as education, the limited labels are usually annotated by multiple workers with diverse expertise, which yields noises and inconsistency in such crowdsourcing settings. In this paper, we propose a novel framework which aims to learn effective representations from limited data with crowdsourced labels. Specifically, we design a grouping based deep neural network to learn embeddings from a limited number of training samples and present a Bayesian confidence estimator to capture the inconsistency among crowdsourced labels. Furthermore, to expedite the training process, we develop a hard example selection procedure to adaptively pick up training examples that are misclassified by the model. Extensive experiments conducted on three real-world data sets demonstrate the superiority of our framework on learning representations from limited data with crowdsourced labels, comparing with various state-of-the-art baselines. In addition, we provide a comprehensive analysis on each of the main components of our proposed framework and also introduce the promising results it achieved in our real production to fully understand the proposed framework. To encourage reproducible results, we make our code available online at https://github.com/tal-ai/RECLE. [ABSTRACT FROM AUTHOR]

In this paper, we consider a real-time deep learning-based symbol detection approach for MIMO-OFDM systems. To exploit the temporal correlation of the wireless channel and the time-frequency structure of OFDM signals, a recurrent neural network (RNN) with deep feedforward output layers is introduced, where the recurrent layers and feedforward output layers are designed to process time-domain and frequency-domain information respectively. Reservoir computing (RC), a special type of RNN, and extreme learning machine (ELM), a special type of feedforward neural network, are chosen as the corresponding building blocks to facilitate over-the-air training. An online training loss objective is introduced to recursively update the neural weights in real-time. We believe this is the first work in the literature to realize real-time machine learning for MIMO-OFDM symbol detection, i.e., conducting NN-based symbol detection on an OFDM symbol basis. We demonstrate that (1) the IEEE standardized WiFi training sequence can be directly applied as the real-time training sequence (2) the symbol detection performance can be further improved by using our theoretically derived pilot pattern. Evaluation results show that our RC-ELM-based symbol detection method outperforms traditional model-based techniques as well as state-of-the-art learning-based approaches in highly dynamic channel environments for real-time symbol detection. [ABSTRACT FROM AUTHOR]

Neural Network Coding and Representation (NNR) is the first international standard for efficient compression of neural networks (NNs). The standard is designed as a toolbox of compression methods, which can be used to create coding pipelines. It can be either used as an independent coding framework (with its own bitstream format) or together with external neural network formats and frameworks. For providing the highest degree of flexibility, the network compression methods operate per parameter tensor in order to always ensure proper decoding, even if no structure information is provided. The NNR standard contains compression-efficient quantization and deep context-adaptive binary arithmetic coding (DeepCABAC) as core encoding and decoding technologies, as well as neural network parameter pre-processing methods like sparsification, pruning, low-rank decomposition, unification, local scaling and batch norm folding. NNR achieves a compression efficiency of more than 97% for transparent coding cases, i.e. without degrading classification quality, such as top-1 or top-5 accuracies. This paper provides an overview of the technical features and characteristics of NNR. [ABSTRACT FROM AUTHOR]

User churn is an important issue in online services that threatens the health and profitability of services. Most of the previous works on churn prediction convert the problem into a binary classification task where the users are labeled as churned and non-churned. More recently, some works have tried to convert the user churn prediction problem into the prediction of user return time. In this approach which is more realistic in real world online services, at each time-step the model predicts the user return time instead of predicting a churn label. However, the previous works in this category suffer from lack of generality and require high computational complexity. In this paper, we introduce ChOracle, an oracle that predicts the user churn by modeling the user return times to service by utilizing a combination of Temporal Point Processes and Recurrent Neural Networks. Moreover, we incorporate latent variables into the proposed recurrent neural network to model the latent user loyalty to the system. We also develop an efficient approximate variational inference algorithm for learning parameters of the proposed RNN by using back propagation through time. Finally, we demonstrate the superior performance of ChOracle on a wide variety of real world datasets. [ABSTRACT FROM AUTHOR]

Artificial Neural networks (ANNs) are widely used to solve classification problems for many machine learning applications. When errors occur in the computational units of an ANN implementation due to for example radiation effects, the result of an arithmetic operation can be changed, and therefore, the predicted classification class may be erroneously affected. This is not acceptable when ANNs are used in many safety-critical applications, because the incorrect classification may result in a system failure. Existing error-tolerant techniques usually rely on physically replicating parts of the ANN implementation or incurring in a significant computation overhead. Therefore, efficient protection schemes are needed for ANNs that are run on a processor and used in resource-limited platforms. A technique referred to as Selective Neuron Re-Computation (SNRC), is proposed in this paper. As per the ANN structure and algorithmic properties, SNRC can identify the cases in which the errors have no impact on the outcome; therefore, errors only need to be handled by re-computation when the classification result is detected as unreliable. Compared with existing temporal redundancy-based protection schemes, SNRC saves more than 60 percent of the re-computation (more than 90 percent in many cases) overhead to achieve complete error protection as assessed over a wide range of datasets. Different activation functions are also evaluated. [ABSTRACT FROM AUTHOR]

Many real–world domains involve information naturally represented by graphs, where nodes denote basic patterns while edges stand for relationships among them. The graph neural network (GNN) is a machine learning model capable of directly managing graph–structured data. In the original framework, GNNs are inductively trained, adapting their parameters based on a supervised learning environment. However, GNNs can also take advantage of transductive learning, thanks to the natural way they make information flow and spread across the graph, using relationships among patterns. In this paper, we propose a mixed inductive–transductive GNN model, study its properties and introduce an experimental strategy that allows us to understand and distinguish the role of inductive and transductive learning. The preliminary experimental results show interesting properties for the mixed model, highlighting how the peculiarities of the problems and the data can impact on the two learning strategies. [ABSTRACT FROM AUTHOR]

Significance: The performance of traditional approaches to decoding movement intent from electromyograms (EMGs) and other biological signals commonly degrade over time. Furthermore, conventional algorithms for training neural network based decoders may not perform well outside the domain of the state transitions observed during training. The work presented in this paper mitigates both these problems, resulting in an approach that has the potential to substantially improve the quality of life of the people with limb loss. Objective: This paper presents and evaluates the performance of four decoding methods for volitional movement intent from intramuscular EMG signals. Methods: The decoders are trained using the dataset aggregation (DAgger) algorithm, in which the training dataset is augmented during each training iteration based on the decoded estimates from previous iterations. Four competing decoding methods, namely polynomial Kalman filters (KFs), multilayer perceptron (MLP) networks, convolutional neural networks (CNN), and long short-term memory (LSTM) networks, were developed. The performances of the four decoding methods were evaluated using EMG datasets recorded from two human volunteers with transradial amputation. Short-term analyses, in which the training and cross-validation data came from the same dataset, and long-term analyses, in which the training and testing were done in different datasets, were performed. Results: Short-term analyses of the decoders demonstrated that CNN and MLP decoders performed significantly better than KF and LSTM decoders, showing an improvement of up to 60% in the normalized mean-square decoding error in cross-validation tests. Long-term analyses indicated that the CNN, MLP, and LSTM decoders performed significantly better than a KF-based decoder at most analyzed cases of temporal separations (0–150 days) between the acquisition of the training and testing datasets. Conclusion: The short-term and long-term performances of MLP- and CNN-based decoders trained with DAgger demonstrated their potential to provide more accurate and naturalistic control of prosthetic hands than alternate approaches. [ABSTRACT FROM AUTHOR]

In order to further improve the productivity of manufacturing equipment, it is indispensable to monitor the conditions of all the manufacturing equipment and not just the processing chambers. In this paper, we present a robust machine learning based deterioration diagnosis technology with harmonic sensor, which has frequency characteristics that have high sensitivity to deterioration and low sensitivity to torque. The robust deterioration diagnosis using a small amount of data is possible by limiting the input data to be trained and inferred to signals and motion patterns, and inferring with neural network models that do not require developments for manufacturing equipment. The example of wafer transfer robots in the ion implanter and the Low-Pressure Chemical Vapor Deposition (LP-CVD) and the coater/developer show that wear deterioration of machine components can be detected from the level of deterioration output and it is possible to prevent errors of the wafer transfer robots because of maintenance based on increases in the level of deterioration. [ABSTRACT FROM AUTHOR]

A general deep learning (DL) mechanism for a multiple hidden layer feed-forward neural network contains two parts, i.e., 1) an unsupervised greedy layer-wise training and 2) a supervised fine-tuning which is usually an iterative process. Although this mechanism has been demonstrated in many fields to be able to significantly improve the generalization of neural network, there is no clear evidence to show which one of the two parts plays the essential role for the generalization improvement, resulting in an argument within the DL community. Focusing on this argument, this paper proposes a new DL approach to train multilayer feed-forward neural networks. This approach uses restricted Boltzmann machine (RBM) as the layer-wise training and uses the generalized inverse of a matrix as the supervised fine-tuning. Different from the general deep training mechanism like back-propagation (BP), the proposed approach does not need to iteratively tune the weights, and therefore, has many advantages such as quick training, better generalization, and high understandability, etc. Experimentally, the proposed approach demonstrates an excellent performance in comparison with BP-based DL and the traditional training method for multilayer random weight neural networks. To a great extent, this paper demonstrates that the supervised part plays a more important role than the unsupervised part in DL, which provides some new viewpoints for exploring the essence of DL. [ABSTRACT FROM AUTHOR]

This paper proposes energy-efficient approximate multipliers based on the Mitchell's log multiplication, optimized for performing inferences on convolutional neural networks (CNN). Various design techniques are applied to the log multiplier, including a fully-parallel LOD, efficient shift amount calculation, and exact zero computation. Additionally, the truncation of the operands is studied to create the customizable log multiplier that further reduces energy consumption. The paper also proposes using the one's complements to handle negative numbers, as an approximation of the two's complements that had been used in the prior works. The viability of the proposed designs is supported by the detailed formal analysis as well as the experimental results on CNNs. The experiments also provide insights into the effect of approximate multiplication in CNNs, identifying the importance of minimizing the range of error.The proposed customizable design at $w$w = 8 saves up to 88 percent energy compared to the exact fixed-point multiplier at 32 bits with just a performance degradation of 0.2 percent for the ImageNet ILSVRC2012 dataset. [ABSTRACT FROM AUTHOR]

This paper proposes a novel intelligent fault diagnosis method to automatically identify different health conditions of wind turbine (WT) gearbox. Unlike traditional approaches, where feature extraction and classification are separately designed and performed, this paper aims to automatically learn effective fault features directly from raw vibration signals while classify the type of faults in a single framework, thus providing an end-to-end learning-based fault diagnosis system for WT gearbox without additional signal processing and diagnostic expertise. Considering the multiscale characteristics inherent in vibration signals of a gearbox, a new multiscale convolutional neural network (MSCNN) architecture is proposed to perform multiscale feature extraction and classification simultaneously. The proposed MSCNN incorporates multiscale learning into the traditional CNN architecture, which has two merits: 1) high-level fault features can be effectively learned by the hierarchical learning structure with multiple pairs of convolutional and pooling layers; and 2) multiscale learning scheme can capture complementary and rich diagnosis information at different scales. This greatly improves the feature learning ability and enables better diagnosis performance. The proposed MSCNN approach is evaluated through experiments on a WT gearbox test rig. Experimental results and comprehensive comparison analysis with respect to the traditional CNN and traditional multiscale feature extractors have demonstrated the superiority of the proposed method. [ABSTRACT FROM AUTHOR]

Linear spectral unmixing is the practice of decomposing the mixed pixel into a linear combination of the constituent endmembers and the estimated abundances. This paper focuses on unsupervised spectral unmixing where the endmembers are unknown a priori. Conventional approaches use either geometrical- or statistical-based approaches. In this paper, we address the challenges of spectral unmixing with unsupervised deep learning models, in specific, the autoencoder models, where the decoder serves as the endmembers and the hidden layer output serves as the abundances. In several recent attempts, part-based autoencoders have been designed to solve the unsupervised spectral unmixing problem. However, the performance has not been satisfactory. In this paper, we first discuss some important findings we make on issues with part-based autoencoders. By proof of counterexample, we show that all existing part-based autoencoder networks with nonnegative and tied encoder and decoder are inherently defective by making these inappropriate assumptions on the network structure. As a result, they are not suitable for solving the spectral unmixing problem. We propose a so-called untied denoising autoencoder with sparsity, in which the encoder and decoder of the network are independent, and only the decoder of the network is enforced to be nonnegative. Furthermore, we make two critical additions to the network design. First, since denoising is an essential step for spectral unmixing, we propose to incorporate the denoising capacity into the network optimization in the format of a denoising constraint rather than cascading another denoising preprocessor in order to avoid the introduction of additional reconstruction error. Second, to be more robust to the inaccurate estimation of a number of endmembers, we adopt an $l_{21}$ -norm on the encoder of the network to reduce the redundant endmembers while decreasing the reconstruction error simultaneously. The experimental results demonstrate that the proposed approach outperforms several state-of-the-art methods, especially for highly noisy data. [ABSTRACT FROM AUTHOR]

This paper proposes a novel method for voltage dip classification using deep convolutional neural networks. The main contributions of this paper include: 1) to propose a new effective deep convolutional neural network architecture for automatically learning voltage dip features, rather than extracting hand-crafted features; 2) to employ the deep learning in an effective two-dimensional (2-D) transform domain, under space-phasor model (SPM), for efficient learning of dip features; 3) to characterize voltage dips by 2-D SPM-based deep learning, which leads to voltage dip features independent of the duration and sampling frequency of dip recordings; and 4) to develop robust automatically-extracted features that are insensitive to training and test datasets measured from different countries/regions. Experiments were conducted on datasets containing about 6000 measured voltage dips spread over seven classes measured from several different countries. Results have shown good performance of the proposed method: average classification rate is about 97% and false alarm rate is about 0.50%. The test results from the proposed method are compared with the results from two existing dip classification methods. The proposed method is shown to outperform these existing methods. [ABSTRACT FROM AUTHOR]

This paper studies the optimized setup in the design-of-experiment (DoE) method to efficiently construct precise artificial neural network (ANN) model for high-speed channel. The accuracy of an ANN model is in general determined by the number of the training sets used in constructing the model. The more the training sets, the better the accuracy is. However, a large number of training sets can significantly increase the effort in obtaining these data (by time-consuming full-wave simulations in the case of channel analysis). Therefore, improving the accuracy while maintaining the same number of training sets by optimizing how the training sets are obtained is critical to develop an efficient ANN construction method. In this paper, different setups in a DoE method, which is used for training data set generation, are studied. Based on error analysis, an optimized setup is proposed. Finally, the performance of the proposed method is validated by the S-parameter simulations with arbitrarily-selected channel parameters within the ranges of interest. [ABSTRACT FROM AUTHOR]

This paper presents an adaptive neural control approach for nonstrict-feedback nonlinear systems in presence of unmodeled dynamics, unknown control directions and input dead-zone nonlinearity. To handle the difficulty due to uncertain control directions, Nussbaum gain functions are applied. Based on the structural characteristic of radial basis function neural networks, a backstepping-based adaptive neural control algorithm is developed. The main contributions of this paper lie in the fact that a backstepping-based neural control algorithm is developed for nonstrict-feedback nonlinear systems with unmodeled dynamics, unknown control directions and actuator dead-zone, and the total number of adaptive laws is not greater than the order of control system. As a beneficial result, the controller is much easier to be implemented in practice with less computational burden. A simulation example is given to reveal the viability of the presented approach. It is demonstrated by both theoretical analysis and simulation study that the presented control strategy ensures the semiglobally uniform ultimate boundedness of all closed-loop system signals. [ABSTRACT FROM AUTHOR]

With only coarse labels, weakly supervised learning typically uses top-down attention maps generated by back-propagating gradients as priors for tasks such as object localization and semantic segmentation. While these attention maps are intuitive and informative explanations of deep neural network, there is no effective mechanism to manipulate the network attention during learning process. In this paper, we address three shortcomings of previous approaches in modeling such attention maps in one common framework. First, we make attention maps a natural and explicit component in the training pipeline such that they are end-to-end trainable. Moreover, we provide self-guidance directly on these maps by exploring supervision from the network itself to improve them towards specific target tasks. Lastly, we proposed a design to seamlessly bridge the gap between using weak and extra supervision if available. Despite its simplicity, experiments on the semantic segmentation task demonstrate the effectiveness of our methods. Besides, the proposed framework provides a way not only explaining the focus of the learner but also feeding back with direct guidance towards specific tasks. Under mild assumptions our method can also be understood as a plug-in to existing convolutional neural networks to improve their generalization performance. [ABSTRACT FROM AUTHOR]

Although deep features have achieved the state-of-the-art performance in action recognition recently, the hand-crafted shallow features still play a critical role in characterizing human actions for taking advantage of visual contents in an intuitive way such as edge features. Therefore, the shallow features can serve as auxiliary visual cues supplementary to deep representations. In this paper, we propose a discriminative subspace learning model (DSLM) to explore the complementary properties between the hand-crafted shallow feature representations and the deep features. As for the RGB action recognition, this is the first work attempting to mine multi-level feature complementaries by the multi-view subspace learning scheme. To sufficiently capture the complementary information among heterogeneous features, we construct the DSLM by integrating the multi-view reconstruction error and classification error into an unified objective function. To be specific, we first use Fisher Vector to encode improved dense trajectories (iDT+FV) for shallow representations and two-stream convolutional neural network models (T-CNN) for generating deep features. Moreover, the presented DSLM algorithm projects multi-level features onto a shared discriminative subspace with the complementary information and discriminating capacity simultaneously incorporated. Finally, the action types of test samples are identified by the margins from the learned compact representations to the decision boundary. The experimental results on three datasets demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

The distribution system state estimation problem seeks to determine the network state from available measurements. Widely used Gauss-Newton approaches are very sensitive to the initialization and often not suitable for real-time estimation. Learning approaches are very promising for real-time estimation, as they shift the computational burden to an offline training stage. Prior machine learning approaches to power system state estimation have been electrical model-agnostic, in the sense that they did not exploit the topology and physical laws governing the power grid to design the architecture of the learning model. In this paper, we propose a novel learning model that utilizes the structure of the power grid. The proposed neural network architecture reduces the number of coefficients needed to parameterize the mapping from the measurements to the network state by exploiting the separability of the estimation problem. This prevents overfitting and reduces the complexity of the training stage. We also propose a greedy algorithm for phasor measuring units placement that aims at minimizing the complexity of the neural network required for realizing the state estimation mapping. Simulation results show superior performance of the proposed method over the Gauss-Newton approach. [ABSTRACT FROM AUTHOR]

In this paper, an $H_{\infty }$ optimal tracking controller for completely unknown discrete-time nonlinear systems with control constraints is obtained by using an iterative adaptive learning algorithm. An augmented system is established by integrating the tracking error system and the reference trajectory. As an identifier of the unknown systems, a neural network (NN) is introduced with asymptotic stability of the estimation error. An action–disturbance–critic NN structure is proposed to implement the iterative dual heuristic programming algorithm with convergence guarantee of the costate function and the control policy. Simulation results and comparisons are provided to illustrate the superior performance of the designed optimal tracking controller. [ABSTRACT FROM AUTHOR]

The extreme learning machine (ELM), which was originally proposed for “generalized” single-hidden layer feedforward neural networks, provides efficient unified learning solutions for the applications of regression and classification. Although, it provides promising performance and robustness and has been used for various applications, the single-layer architecture possibly lacks the effectiveness when applied for natural signals. In order to over come this shortcoming, the following work indicates a new architecture based on multilayer network framework. The significant contribution of this paper are as follows: 1) unlike existing multilayer ELM, in which hidden nodes are obtained randomly, in this paper all hidden layers with invertible functions are calculated by pulling the network output back and putting it into hidden layers. Thus, the feature learning is enriched by additional information, which results in better performance; 2) in contrast to the existing multilayer network methods, which are usually efficient for classification applications, the proposed architecture is implemented for dimension reduction and image reconstruction; and 3) unlike other iterative learning-based deep networks (DL), the hidden layers of the proposed method are obtained via four steps. Therefore, it has much better learning efficiency than DL. Experimental results on 33 datasets indicate that, in comparison to the other existing dimension reduction techniques, the proposed method performs competitively better with fast training speeds. [ABSTRACT FROM AUTHOR]

Supervised approaches classify input data using a set of representative samples for each class, known as training samples. The collection of such samples is expensive and time demanding. Hence, unsupervised feature learning, which has a quick access to arbitrary amounts of unlabeled data, is conceptually of high interest. In this paper, we propose a novel network architecture, fully Conv-Deconv network, for unsupervised spectral-spatial feature learning of hyperspectral images, which is able to be trained in an end-to-end manner. Specifically, our network is based on the so-called encoder-decoder paradigm, i.e., the input 3-D hyperspectral patch is first transformed into a typically lower dimensional space via a convolutional subnetwork (encoder), and then expanded to reproduce the initial data by a deconvolutional subnetwork (decoder). However, during the experiment, we found that such a network is not easy to be optimized. To address this problem, we refine the proposed network architecture by incorporating: 1) residual learning and 2) a new unpooling operation that can use memorized max-pooling indexes. Moreover, to understand the "black box," we make an in-depth study of the learned feature maps in the experimental analysis. A very interesting discovery is that some specific "neurons" in the first residual block of the proposed network own good description power for semantic visual patterns in the object level, which provide an opportunity to achieve "free" object detection. This paper, for the first time in the remote sensing community, proposes an end-to-end fully Conv-Deconv network for unsupervised spectral-spatial feature learning. Moreover, this paper also introduces an in-depth investigation of learned features. Experimental results on two widely used hyperspectral data, Indian Pines and Pavia University, demonstrate competitive performance obtained by the proposed methodology compared with other studied approaches. [ABSTRACT FROM AUTHOR]

This paper deals with semantic segmentation of high-resolution (aerial) images where a semantic class label is assigned to each pixel via supervised classification as a basis for automatic map generation. Recently, deep convolutional neural networks (CNNs) have shown impressive performance and have quickly become the de-facto standard for semantic segmentation, with the added benefit that task-specific feature design is no longer necessary. However, a major downside of deep learning methods is that they are extremely data hungry, thus aggravating the perennial bottleneck of supervised classification, to obtain enough annotated training data. On the other hand, it has been observed that they are rather robust against noise in the training labels. This opens up the intriguing possibility to avoid annotating huge amounts of training data, and instead train the classifier from existing legacy data or crowd-sourced maps that can exhibit high levels of noise. The question addressed in this paper is: can training with large-scale publicly available labels replace a substantial part of the manual labeling effort and still achieve sufficient performance? Such data will inevitably contain a significant portion of errors, but in return virtually unlimited quantities of it are available in larger parts of the world. We adapt a state-of-the-art CNN architecture for semantic segmentation of buildings and roads in aerial images, and compare its performance when using different training data sets, ranging from manually labeled pixel-accurate ground truth of the same city to automatic training data derived from OpenStreetMap data from distant locations. We report our results that indicate that satisfying performance can be obtained with significantly less manual annotation effort, by exploiting noisy large-scale training data. [ABSTRACT FROM PUBLISHER]

Predicting the millimeter wave (mmWave) beams and blockages using sub-6 GHz channels has the potential of enabling mobility and reliability in scalable mmWave systems. Prior work has focused on extracting spatial channel characteristics at the sub-6 GHz band and then use them to reduce the mmWave beam training overhead. This approach still requires beam refinement at mmWave and does not normally account for the different dielectric properties at the different bands. In this paper, we first prove that under certain conditions, there exist mapping functions that can predict the optimal mmWave beam and blockage status directly from the sub-6 GHz channel. These mapping functions, however, are hard to characterize analytically which motivates exploiting deep neural network models to learn them. For that, we prove that a large enough neural network can predict mmWave beams and blockages with success probabilities that can be made arbitrarily close to one. Then, we develop a deep learning model and empirically evaluate its beam/blockage prediction performance using a publicly available dataset. The results show that the proposed solution can predict the mmWave blockages with more than 90% success probability and can predict the optimal mmWave beams to approach the upper bounds while requiring no beam training overhead. [ABSTRACT FROM AUTHOR]

The long short-term memory (LSTM) is a widely used neural network model for dealing with time-varying data. To reduce the memory requirement, pruning is often applied to the weight matrix of the LSTM, which makes the matrix sparse. In this paper, we present a new sparse matrix format, named rearranged compressed sparse column (RCSC), to maximize the inference speed of the LSTM hardware accelerator. The RCSC format speeds up the inference by: 1) evenly distributing the computation loads to processing elements (PEs) and 2) reducing the input vector load miss within the local buffer. We also propose a hardware architecture adopting hierarchical input buffer to further reduce the pipeline stalls which cannot be handled by the RCSC format alone. The simulation results for various datasets show that combined use of the RSCS format and the proposed hardware requires $2\times $ smaller inference runtime on average compared to the previous work. [ABSTRACT FROM AUTHOR]

Vascular tree disentanglement and vessel type classification are two crucial steps of the graph-based method for retinal artery-vein (A/V) separation. Existing approaches treat them as two independent tasks and mostly rely on ad hoc rules (e.g. change of vessel directions) and hand-crafted features (e.g. color, thickness) to handle them respectively. However, we argue that the two tasks are highly correlated and should be handled jointly since knowing the A/V type can unravel those highly entangled vascular trees, which in turn helps to infer the types of connected vessels that are hard to classify based on only appearance. Therefore, designing features and models isolatedly for the two tasks often leads to a suboptimal solution of A/V separation. In view of this, this paper proposes a multi-task siamese network which aims to learn the two tasks jointly and thus yields more robust deep features for accurate A/V separation. Specifically, we first introduce Convolution Along Vessel (CAV) to extract the visual features by convolving a fundus image along vessel segments, and the geometric features by tracking the directions of blood flow in vessels. The siamese network is then trained to learn multiple tasks: i) classifying A/V types of vessel segments using visual features only, and ii) estimating the similarity of every two connected segments by comparing their visual and geometric features in order to disentangle the vasculature into individual vessel trees. Finally, the results of two tasks mutually correct each other to accomplish final A/V separation. Experimental results demonstrate that our method can achieve accuracy values of 94.7%, 96.9%, and 94.5% on three major databases (DRIVE, INSPIRE, WIDE) respectively, which outperforms recent state-of-the-arts. [ABSTRACT FROM AUTHOR]