The welding seams visual inspection is still manually operated by humans in different companies, so the result of the test is still highly subjective and expensive. At present, the integration of deep learning methods for welds classification is a research focus in engineering applications. This work intends to apprehend and emphasize the contribution of deep learning model explainability to the improvement of welding seams classification accuracy and reliability, two of the various metrics affecting the production lines and cost in the automotive industry. For this purpose, we implement a novel hybrid method that relies on combining the model prediction scores and visual explanation heatmap of the model in order to make a more accurate classification of welding seam defects and improve both its performance and its reliability. The results show that the hybrid model performance is relatively above our target performance and helps to increase the accuracy by at least 18%, which presents new perspectives to the developments of deep Learning explainability and interpretability.
Uncertainty quantification of machine learning and deep learning methods plays an important role in enhancing trust to the obtained result. In recent years, a numerous number of uncertainty quantification methods have been introduced. Monte Carlo dropout (MC-Dropout) is one of the most well-known techniques to quantify uncertainty in deep learning methods. In this study, we propose two new loss functions by combining cross entropy with Expected Calibration Error (ECE) and Predictive Entropy (PE). The obtained results clearly show that the new proposed loss functions lead to having a calibrated MC-Dropout method. Our results confirmed the great impact of the new hybrid loss functions for minimising the overlap between the distributions of uncertainty estimates for correct and incorrect predictions without sacrificing the model's overall performance.
Watershed models such as the Soil and Water Assessment Tool (SWAT) consist of high-dimensional physical and empirical parameters. These parameters need to be accurately calibrated for models to produce reliable predictions for streamflow, evapotranspiration, snow water equivalent, and nutrient loading. Existing parameter estimation methods are time-consuming, inefficient, and computationally intensive, with reduced accuracy when estimating high-dimensional parameters. In this paper, we present a fast, accurate, and reliable methodology to calibrate the SWAT model (i.e., 21 parameters) using deep learning (DL). We develop DL-enabled inverse models based on convolutional neural networks to ingest streamflow data and estimate the SWAT model parameters. Hyperparameter tuning is performed to identify the optimal neural network architecture and the nine next best candidates. We use ensemble SWAT simulations to train, validate, and test the above DL models. We estimated the actual parameters of the SWAT model using observational data. We test and validate the proposed DL methodology on the American River Watershed, located in the Pacific Northwest-based Yakima River basin. Our results show that the DL models-based calibration is better than traditional parameter estimation methods, such as generalized likelihood uncertainty estimation (GLUE). The behavioral parameter sets estimated by DL have narrower ranges than GLUE and produce values within the sampling range even under high relative observational errors. This narrow range of parameters shows the reliability of the proposed workflow to estimate sensitive parameters accurately even under noise. Due to its fast and reasonably accurate estimations of process parameters, the proposed DL workflow is attractive for calibrating integrated hydrologic models for large spatial-scale applications.
Singing voice synthesis (SVS) is a task that aims to generate audio signals according to musical scores and lyrics. With its multifaceted nature concerning music and language, producing singing voices indistinguishable from that of human singers has always remained an unfulfilled pursuit. Nonetheless, the advancements of deep learning techniques have brought about a substantial leap in the quality and naturalness of synthesized singing voice. This paper aims to review some of the state-of-the-art deep learning-driven SVS systems. We intend to summarize their deployed model architectures and identify the strengths and limitations for each of the introduced systems. Thereby, we picture the recent advancement trajectory of this field and conclude the challenges left to be resolved both in commercial applications and academic research.
Automatic speech recognition (ASR) is a capability which enables a program to process human speech into a written form. Recent developments in artificial intelligence (AI) have led to high-accuracy ASR systems based on deep neural networks, such as the recurrent neural network transducer (RNN-T). However, the core components and the performed operations of these approaches depart from the powerful biological counterpart, i.e., the human brain. On the other hand, the current developments in biologically-inspired ASR models, based on spiking neural networks (SNNs), lag behind in terms of accuracy and focus primarily on small scale applications. In this work, we revisit the incorporation of biologically-plausible models into deep learning and we substantially enhance their capabilities, by taking inspiration from the diverse neural and synaptic dynamics found in the brain. In particular, we introduce neural connectivity concepts emulating the axo-somatic and the axo-axonic synapses. Based on this, we propose novel deep learning units with enriched neuro-synaptic dynamics and integrate them into the RNN-T architecture. We demonstrate for the first time, that a biologically realistic implementation of a large-scale ASR model can yield competitive performance levels compared to the existing deep learning models. Specifically, we show that such an implementation bears several advantages, such as a reduced computational cost and a lower latency, which are critical for speech recognition applications.
This paper presents a deep learning-based audio-in-image watermarking scheme. Audio-in-image watermarking is the process of covertly embedding and extracting audio watermarks on a cover-image. Using audio watermarks can open up possibilities for different downstream applications. For the purpose of implementing an audio-in-image watermarking that adapts to the demands of increasingly diverse situations, a neural network architecture is designed to automatically learn the watermarking process in an unsupervised manner. In addition, a similarity network is developed to recognize the audio watermarks under distortions, therefore providing robustness to the proposed method. Experimental results have shown high fidelity and robustness of the proposed blind audio-in-image watermarking scheme.
Much of the success of deep learning is drawn from building architectures that properly respect underlying symmetry and structure in the data on which they operate - a set of considerations that have been united under the banner of geometric deep learning. Often problems in the physical sciences deal with relatively small sets of points in two- or three-dimensional space wherein translation, rotation, and permutation equivariance are important or even vital for models to be useful in practice. In this work, we present rotation- and permutation-equivariant architectures for deep learning on these small point clouds, composed of a set of products of terms from the geometric algebra and reductions over those products using an attention mechanism. The geometric algebra provides valuable mathematical structure by which to combine vector, scalar, and other types of geometric inputs in a systematic way to account for rotation invariance or covariance, while attention yields a powerful way to impose permutation equivariance. We demonstrate the usefulness of these architectures by training models to solve sample problems relevant to physics, chemistry, and biology.
With the continuous improvement in various aspects in the field of artificial intelligence, the momentum of artificial intelligence with deep learning capabilities into the field of music is coming. The research purpose of this paper is to design a Bach style music authoring system based on deep learning. We use a LSTM neural network to train serialized and standardized music feature data. By repeated experiments, we find the optimal LSTM model which can generate imitation of Bach music. Finally the generated music is comprehensively evaluated in the form of online audition and Turing test. The repertoires which the music generation system constructed in this article are very close to the style of Bach's original music, and it is relatively difficult for ordinary people to distinguish the musics Bach authored and AI created.
Multiplex networks are complex graph structures in which a set of entities are connected to each other via multiple types of relations, each relation representing a distinct layer. Such graphs are used to investigate many complex biological, social, and technological systems. In this work, we present a novel semi-supervised approach for structure-aware representation learning on multiplex networks. Our approach relies on maximizing the mutual information between local node-wise patch representations and label correlated structure-aware global graph representations to model the nodes and cluster structures jointly. Specifically, it leverages a novel cluster-aware, node-contextualized global graph summary generation strategy for effective joint-modeling of node and cluster representations across the layers of a multiplex network. Empirically, we demonstrate that the proposed architecture outperforms state-of-the-art methods in a range of tasks: classification, clustering, visualization, and similarity search on seven real-world multiplex networks for various experiment settings.
Heterogeneous tabular data are the most commonly used form of data and are essential for numerous critical and computationally demanding applications. On homogeneous data sets, deep neural networks have repeatedly shown excellent performance and have therefore been widely adopted. However, their adaptation to tabular data for inference or data generation tasks remains challenging. To facilitate further progress in the field, this work provides an overview of state-of-the-art deep learning methods for tabular data. We categorize these methods into three groups: data transformations, specialized architectures, and regularization models. For each of these groups, our work offers a comprehensive overview of the main approaches. Moreover, we discuss deep learning approaches for generating tabular data, and we also provide an overview over strategies for explaining deep models on tabular data. Thus, our first contribution is to address the main research streams and existing methodologies in the mentioned areas, while highlighting relevant challenges and open research questions. Our second contribution is to provide an empirical comparison of traditional machine learning methods with eleven deep learning approaches across five popular real-world tabular data sets of different sizes and with different learning objectives. Our results, which we have made publicly available as competitive benchmarks, indicate that algorithms based on gradient-boosted tree ensembles still mostly outperform deep learning models on supervised learning tasks, suggesting that the research progress on competitive deep learning models for tabular data is stagnating. To the best of our knowledge, this is the first in-depth overview of deep learning approaches for tabular data; as such, this work can serve as a valuable starting point to guide researchers and practitioners interested in deep learning with tabular data.
Accurate weather prediction is essential for many aspects of life, notably the early warning of extreme weather events such as rainstorms. Short-term predictions of these events rely on forecasts from numerical weather models, in which, despite much improvement in the past decades, outstanding issues remain concerning model uncertainties, and increasing demands for computation and storage resources. In recent years, the advance of deep learning offers a viable alternative approach. Here, we show that a 3D convolutional neural network using a single frame of meteorology fields as input is capable of predicting the precipitation spatial distribution. The network is developed based on 39-years (1980-2018) data of meteorology and daily precipitation over the contiguous United States. The results bring fundamental advancements in weather prediction. First, the trained network alone outperforms the state-of-the-art weather models in predicting daily total precipitation, and the superiority of the network extends to forecast leads up to 5 days. Second, combining the network predictions with the weather-model forecasts significantly improves the accuracy of model forecasts, especially for heavy-precipitation events. Third, the millisecond-scale inference time of the network facilitates large ensemble predictions for further accuracy improvement. These findings strongly support the use of deep-learning in short-term weather predictions.
Transfer learning (TL) for medical image segmentation helps deep learning models achieve more accurate performances when there are scarce medical images. This study focuses on completing segmentation of the ribs from lung ultrasound images and finding the best TL technique with U-Net, a convolutional neural network for precise and fast image segmentation. Two approaches of TL were used, using a pre-trained VGG16 model to build the U-Net (V-Unet) and pre-training U-Net network with grayscale natural salient object dataset (X-Unet). Visual results and dice coefficients (DICE) of the models were compared. X-Unet showed more accurate and artifact-free visual performances on the actual mask prediction, despite its lower DICE than V-Unet. A partial-frozen network fine-tuning (FT) technique was also applied to X-Unet to compare results between different FT strategies, which FT all layers slightly outperformed freezing part of the network. The effect of dataset sizes was also evaluated, showing the importance of the combination between TL and data augmentation.
Distributed stochastic gradient descent (SGD) approach has been widely used in large-scale deep learning, and the gradient collective method is vital to ensure the training scalability of the distributed deep learning system. Collective communication such as AllReduce has been widely adopted for the distributed SGD process to reduce the communication time. However, AllReduce incurs large bandwidth resources while most gradients are sparse in many cases since many gradient values are zeros and should be efficiently compressed for bandwidth saving. To reduce the sparse gradient communication overhead, we propose Sparse-Sketch Reducer (S2 Reducer), a novel sketch-based sparse gradient aggregation method with convergence guarantees. S2 Reducer reduces the communication cost by only compressing the non-zero gradients with count-sketch and bitmap, and enables the efficient AllReduce operators for parallel SGD training. We perform extensive evaluation against four state-of-the-art methods over five training models. Our results show that S2 Reducer converges to the same accuracy, reduces 81\% sparse communication overhead, and achieves 1.8$ \times $ speedup compared to state-of-the-art approaches.
Many critical applications rely on cameras to capture video footage for analytical purposes. This has led to concerns about these cameras accidentally capturing more information than is necessary. In this paper, we propose a deep learning approach towards protecting privacy in camera-based systems. Instead of specifying specific objects (e.g. faces) are privacy sensitive, our technique distinguishes between salient (visually prominent) and non-salient objects based on the intuition that the latter is unlikely to be needed by the application.
Modern image files are usually progressively transmitted and provide a preview before downloading the entire image for improved user experience to cope with a slow network connection. In this paper, with a similar goal, we propose a progressive transmission framework for deep learning models, especially to deal with the scenario where pre-trained deep learning models are transmitted from servers and executed at user devices (e.g., web browser or mobile). Our progressive transmission allows inferring approximate models in the middle of file delivery, and quickly provide an acceptable intermediate outputs. On the server-side, a deep learning model is divided and progressively transmitted to the user devices. Then, the divided pieces are progressively concatenated to construct approximate models on user devices. Experiments show that our method is computationally efficient without increasing the model size and total transmission time while preserving the model accuracy. We further demonstrate that our method can improve the user experience by providing the approximate models especially in a slow connection.
Deep learning has powered recent successes of artificial intelligence (AI). However, the deep neural network, as the basic model of deep learning, has suffered from issues such as local traps and miscalibration. In this paper, we provide a new framework for sparse deep learning, which has the above issues addressed in a coherent way. In particular, we lay down a theoretical foundation for sparse deep learning and propose prior annealing algorithms for learning sparse neural networks. The former has successfully tamed the sparse deep neural network into the framework of statistical modeling, enabling prediction uncertainty correctly quantified. The latter can be asymptotically guaranteed to converge to the global optimum, enabling the validity of the down-stream statistical inference. Numerical result indicates the superiority of the proposed method compared to the existing ones.
Graph Neural Networks (GNN) have shown a strong potential to be integrated into commercial products for network control and management. Early works using GNN have demonstrated an unprecedented capability to learn from different network characteristics that are fundamentally represented as graphs, such as the topology, the routing configuration, or the traffic that flows along a series of nodes in the network. In contrast to previous solutions based on Machine Learning (ML), GNN enables to produce accurate predictions even in other networks unseen during the training phase. Nowadays, GNN is a hot topic in the Machine Learning field and, as such, we are witnessing great efforts to leverage its potential in many different fields (e.g., chemistry, physics, social networks). In this context, the Graph Neural Networking challenge 2021 brings a practical limitation of existing GNN-based solutions for networking: the lack of generalization to larger networks. This paper approaches the scalability problem by presenting a GNN-based solution that can effectively scale to larger networks including higher link capacities and aggregated traffic on links.
Here, we develop a deep learning algorithm for solving Principal-Agent (PA) mean field games with market-clearing conditions -- a class of problems that have thus far not been studied and one that poses difficulties for standard numerical methods. We use an actor-critic approach to optimization, where the agents form a Nash equilibria according to the principal's penalty function, and the principal evaluates the resulting equilibria. The inner problem's Nash equilibria is obtained using a variant of the deep backward stochastic differential equation (BSDE) method modified for McKean-Vlasov forward-backward SDEs that includes dependence on the distribution over both the forward and backward processes. The outer problem's loss is further approximated by a neural net by sampling over the space of penalty functions. We apply our approach to a stylized PA problem arising in Renewable Energy Certificate (REC) markets, where agents may rent clean energy production capacity, trade RECs, and expand their long-term capacity to navigate the market at maximum profit. Our numerical results illustrate the efficacy of the algorithm and lead to interesting insights into the nature of optimal PA interactions in the mean-field limit of these markets.
Destination prediction is known as an important problem for many location based services (LBSs). Existing solutions generally apply probabilistic models or neural network models to predict destinations over a subtrajectory, and adopt the standard attention mechanism to improve the prediction accuracy. However, the standard attention mechanism uses fixed feature representations, and has a limited ability to represent distinct features of locations. Besides, existing methods rarely take the impact of spatial and temporal characteristics of the trajectory into account. Their accuracies in fine-granularity prediction are always not satisfactory due to the data sparsity problem. Thus, in this paper, a carefully designed deep learning model called LATL model is presented. It not only adopts an adaptive attention network to model the distinct features of locations, but also implements time gates and distance gates into the Long Short-Term Memory (LSTM) network to capture the spatial-temporal relation between consecutive locations. Furthermore, to better understand the mobility patterns in different spatial granularities, and explore the fusion of multi-granularity learning capability, a hierarchical model that utilizes tailored combination of different neural networks under multiple spatial granularities is further proposed. Extensive empirical studies verify that the newly proposed models perform effectively and settle the problem nicely.
Line of sight satellite systems, unmanned aerial vehicles, high-altitude platforms, and microwave links that operate on frequency bands such as Ka-band or higher are extremely susceptible to rain. Thus, rain fade forecasting for these systems is critical because it allows the system to switch between ground gateways proactively before a rain fade event to maintain seamless service. Although empirical, statistical, and fade slope models can predict rain fade to some extent, they typically require statistical measurements of rain characteristics in a given area and cannot be generalized to a large scale system. Furthermore, such models typically predict near-future rain fade events but are incapable of forecasting far into the future, making proactive resource management more difficult. In this paper, a deep learning (DL)-based architecture is proposed that forecasts future rain fade using satellite and radar imagery data as well as link power measurements. Furthermore, the data preprocessing and architectural design have been thoroughly explained and multiple experiments have been conducted. Experiments show that the proposed DL architecture outperforms current state-of-the-art machine learning-based algorithms in rain fade forecasting in the near and long term. Moreover, the results indicate that radar data with weather condition information is more effective for short-term prediction, while satellite data with cloud movement information is more effective for long-term predictions.