PyTorchKR Papers

최신 논문

총 8,922편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2108.13880공식 코드 · Papers with Code

Using a one dimensional parabolic model of the full-batch loss to estimate learning rates during training

A fundamental challenge in Deep Learning is to find optimal step sizes for stochastic gradient descent automatically. In traditional optimization, line searches are a commonly used method to determine step sizes. One problem in Deep Learning is that finding appropriate step sizes on the full-batch loss is unfeasibly expensive. Therefore, classical line search approaches, designed for losses without inherent noise, are usually not applicable. Recent empirical findings suggest, inter alia, that the full-batch loss behaves locally parabolically in the direction of noisy update step directions. Furthermore, the trend of the optimal update step size changes slowly. By exploiting these and more findings, this work introduces a line-search method that approximates the full-batch loss with a parabola estimated over several mini-batches. Learning rates are derived from such parabolas during training. In the experiments conducted, our approach is on par with SGD with Momentum tuned with a piece-wise constant learning rate schedule and often outperforms other line search approaches for Deep Learning across models, datasets, and batch sizes on validation and test accuracy. In addition, our approach is the first line search approach for Deep Learning that samples a larger batch size over multiple inferences to still work in low-batch scenarios.

arXiv:2108.13732공식 코드 · Papers with Code

Deep Learning on Edge TPUs

Computing at the edge is important in remote settings, however, conventional hardware is not optimized for utilizing deep neural networks. The Google Edge TPU is an emerging hardware accelerator that is cost, power and speed efficient, and is available for prototyping and production purposes. Here, I review the Edge TPU platform, the tasks that have been accomplished using the Edge TPU, and which steps are necessary to deploy a model to the Edge TPU hardware. The Edge TPU is not only capable of tackling common computer vision tasks, but also surpasses other hardware accelerators, especially when the entire model can be deployed to the Edge TPU. Co-embedding the Edge TPU in cameras allows a seamless analysis of primary data. In summary, the Edge TPU is a maturing system that has proven its usability across multiple tasks.

arXiv:2108.13653

Explaining Classes through Word Attribution

In recent years, several methods have been proposed for explaining individual predictions of deep learning models, yet there has been little study of how to aggregate these predictions to explain how such models view classes as a whole in text classification tasks. In this work, we propose a method for explaining classes using deep learning models and the Integrated Gradients feature attribution technique by aggregating explanations of individual examples in text classification to general descriptions of the classes. We demonstrate the approach on Web register (genre) classification using the XML-R model and the Corpus of Online Registers of English (CORE), finding that the method identifies plausible and discriminative keywords characterizing all but the smallest class.

arXiv:2108.13640

Module-Power Prediction from PL Measurements using Deep Learning

The individual causes for power loss of photovoltaic modules are investigated for quite some time. Recently, it has been shown that the power loss of a module is, for example, related to the fraction of inactive areas. While these areas can be easily identified from electroluminescense (EL) images, this is much harder for photoluminescence (PL) images. With this work, we close the gap between power regression from EL and PL images. We apply a deep convolutional neural network to predict the module power from PL images with a mean absolute error (MAE) of 4.4% or 11.7WP. Furthermore, we depict that regression maps computed from the embeddings of the trained network can be used to compute the localized power loss. Finally, we show that these regression maps can be used to identify inactive regions in PL images as well.

arXiv:2108.13465

Full-Cycle Energy Consumption Benchmark for Low-Carbon Computer Vision

The energy consumption of deep learning models is increasing at a breathtaking rate, which raises concerns due to potential negative effects on carbon neutrality in the context of global warming and climate change. With the progress of efficient deep learning techniques, e.g., model compression, researchers can obtain efficient models with fewer parameters and smaller latency. However, most of the existing efficient deep learning methods do not explicitly consider energy consumption as a key performance indicator. Furthermore, existing methods mostly focus on the inference costs of the resulting efficient models, but neglect the notable energy consumption throughout the entire life cycle of the algorithm. In this paper, we present the first large-scale energy consumption benchmark for efficient computer vision models, where a new metric is proposed to explicitly evaluate the full-cycle energy consumption under different model usage intensity. The benchmark can provide insights for low carbon emission when selecting efficient deep learning algorithms in different model usage scenarios.

arXiv:2108.13461

Time Series Prediction using Deep Learning Methods in Healthcare

Traditional machine learning methods face two main challenges in dealing with healthcare predictive analytics tasks. First, the high-dimensional nature of healthcare data needs labor-intensive and time-consuming processes to select an appropriate set of features for each new task. Second, these methods depend on feature engineering to capture the sequential nature of patient data, which may not adequately leverage the temporal patterns of the medical events and their dependencies. Recent deep learning methods have shown promising performance for various healthcare prediction tasks by addressing the high-dimensional and temporal challenges of medical data. These methods can learn useful representations of key factors (e.g., medical concepts or patients) and their interactions from high-dimensional raw or minimally-processed healthcare data. In this paper we systematically reviewed studies focused on advancing and using deep neural networks to leverage patients structured time series data for healthcare prediction tasks. To identify relevant studies, MEDLINE, IEEE, Scopus and ACM digital library were searched for studies published up to February 7th 2021. We found that researchers have contributed to deep time series prediction literature in ten research streams: deep learning models, missing value handling, irregularity handling, patient representation, static data inclusion, attention mechanisms, interpretation, incorporating medical ontologies, learning strategies, and scalability. This study summarizes research insights from these literature streams, identifies several critical research gaps, and suggests future research opportunities for deep learning in patient time series data.

arXiv:2108.13989

DeepTaskAPT: Insider APT detection using Task-tree based Deep Learning

APT, known as Advanced Persistent Threat, is a difficult challenge for cyber defence. These threats make many traditional defences ineffective as the vulnerabilities exploited by these threats are insiders who have access to and are within the network. This paper proposes DeepTaskAPT, a heterogeneous task-tree based deep learning method to construct a baseline model based on sequences of tasks using a Long Short-Term Memory (LSTM) neural network that can be applied across different users to identify anomalous behaviour. Rather than applying the model to sequential log entries directly, as most current approaches do, DeepTaskAPT applies a process tree based task generation method to generate sequential log entries for the deep learning model. To assess the performance of DeepTaskAPT, we use a recently released synthetic dataset, DARPA Operationally Transparent Computing (OpTC) dataset and a real-world dataset, Los Alamos National Laboratory (LANL) dataset. Both of them are composed of host-based data collected from sensors. Our results show that DeepTaskAPT outperforms similar approaches e.g. DeepLog and the DeepTaskAPT baseline model demonstrate its capability to detect malicious traces in various attack scenarios while having high accuracy and low false-positive rates. To the best of knowledge this is the very first attempt of using recently introduced OpTC dataset for cyber threat detection.

arXiv:2108.13203

Feature Importance in a Deep Learning Climate Emulator

We present a study using a class of post-hoc local explanation methods i.e., feature importance methods for "understanding" a deep learning (DL) emulator of climate. Specifically, we consider a multiple-input-single-output emulator that uses a DenseNet encoder-decoder architecture and is trained to predict interannual variations of sea surface temperature (SST) at 1, 6, and 9 month lead times using the preceding 36 months of (appropriately filtered) SST data. First, feature importance methods are employed for individual predictions to spatio-temporally identify input features that are important for model prediction at chosen geographical regions and chosen prediction lead times. In a second step, we also examine the behavior of feature importance in a generalized sense by considering an aggregation of the importance heatmaps over training samples. We find that: 1) the climate emulator's prediction at any geographical location depends dominantly on a small neighborhood around it; 2) the longer the prediction lead time, the further back the "importance" extends; and 3) to leading order, the temporal decay of "importance" is independent of geographical location. An ablation experiment is adopted to verify the findings. From the perspective of climate dynamics, these findings suggest a dominant role for local processes and a negligible role for remote teleconnections at the spatial and temporal scales we consider. From the perspective of network architecture, the spatio-temporal relations between the inputs and outputs we find suggest potential model refinements. We discuss further extensions of our methods, some of which we are considering in ongoing work.

arXiv:2108.12768공식 코드 · Papers with Code

CrossedWires: A Dataset of Syntactically Equivalent but Semantically Disparate Deep Learning Models

The training of neural networks using different deep learning frameworks may lead to drastically differing accuracy levels despite the use of the same neural network architecture and identical training hyperparameters such as learning rate and choice of optimization algorithms. Currently, our ability to build standardized deep learning models is limited by the availability of a suite of neural network and corresponding training hyperparameter benchmarks that expose differences between existing deep learning frameworks. In this paper, we present a living dataset of models and hyperparameters, called CrossedWires, that exposes semantic differences between two popular deep learning frameworks: PyTorch and Tensorflow. The CrossedWires dataset currently consists of models trained on CIFAR10 images using three different computer vision architectures: VGG16, ResNet50 and DenseNet121 across a large hyperparameter space. Using hyperparameter optimization, each of the three models was trained on 400 sets of hyperparameters suggested by the HyperSpace search algorithm. The CrossedWires dataset includes PyTorch and Tensforflow models with test accuracies as different as 0.681 on syntactically equivalent models and identical hyperparameter choices. The 340 GB dataset and benchmarks presented here include the performance statistics, training curves, and model weights for all 1200 hyperparameter choices, resulting in 2400 total models. The CrossedWires dataset provides an opportunity to study semantic differences between syntactically equivalent models across popular deep learning frameworks. Further, the insights obtained from this study can enable the development of algorithms and tools that improve reliability and reproducibility of deep learning frameworks. The dataset is freely available at https://github.com/maxzvyagin/crossedwires through a Python API and direct download link.

arXiv:2108.12489

Using Graph Neural Networks to model the performance of Deep Neural Networks

With the unprecedented proliferation of machine learning software, there is an ever-increasing need to generate efficient code for such applications. State-of-the-art deep-learning compilers like TVM and Halide incorporate a learning-based performance model to search the space of valid implementations of a given deep learning algorithm. For a given application, the model generates a performance metric such as the run time without executing the application on hardware. Such models speed up the compilation process by obviating the need to benchmark an enormous number of candidate implementations, referred to as schedules, on hardware. Existing performance models employ feed-forward networks, recurrent networks, or decision tree ensembles to estimate the performance of different implementations of a neural network. Graphs present a natural and intuitive way to model deep-learning networks where each node represents a computational stage or operation. Incorporating the inherent graph structure of these workloads in the performance model can enable a better representation and learning of inter-stage interactions. The accuracy of a performance model has direct implications on the efficiency of the search strategy, making it a crucial component of this class of deep-learning compilers. In this work, we develop a novel performance model that adopts a graph representation. In our model, each stage of computation represents a node characterized by features that capture the operations performed by the stage. The interaction between nodes is achieved using graph convolutions. Experimental evaluation shows a 7:75x and 12x reduction in prediction error compared to the Halide and TVM models, respectively.

arXiv:2108.12061

Using GAN-based models to sentimental analysis on imbalanced datasets in education domain

While the whole world is still struggling with the COVID-19 pandemic, online learning and home office become more common. Many schools transfer their courses teaching to the online classroom. Therefore, it is significant to mine the students' feedback and opinions from their reviews towards studies so that both schools and teachers can know where they need to improve. This paper trains machine learning and deep learning models using both balanced and imbalanced datasets for sentiment classification. Two SOTA category-aware text generation GAN models: CatGAN and SentiGAN, are utilized to synthesize text used to balance the highly imbalanced dataset. Results on three datasets with different imbalance degree from distinct domains show that when using generated text to balance the dataset, the F1-score of machine learning and deep learning model on sentiment classification increases 2.79% ~ 9.28%. Also, the results indicate that the average growth degree for CR100k is higher than CR23k, the average growth degree for deep learning is more increased than machine learning algorithms, and the average growth degree for more complex deep learning models is more increased than simpler deep learning models in experiments.

arXiv:2108.12290공식 코드 · Papers with Code

Music Composition with Deep Learning: A Review

Generating a complex work of art such as a musical composition requires exhibiting true creativity that depends on a variety of factors that are related to the hierarchy of musical language. Music generation have been faced with Algorithmic methods and recently, with Deep Learning models that are being used in other fields such as Computer Vision. In this paper we want to put into context the existing relationships between AI-based music composition models and human musical composition and creativity processes. We give an overview of the recent Deep Learning models for music composition and we compare these models to the music composition process from a theoretical point of view. We have tried to answer some of the most relevant open questions for this task by analyzing the ability of current Deep Learning models to generate music with creativity or the similarity between AI and human composition processes, among others.

arXiv:2108.11730

Deep learning based dictionary learning and tomographic image reconstruction

This work presents an approach for image reconstruction in clinical low-dose tomography that combines principles from sparse signal processing with ideas from deep learning. First, we describe sparse signal representation in terms of dictionaries from a statistical perspective and interpret dictionary learning as a process of aligning distribution that arises from a generative model with empirical distribution of true signals. As a result we can see that sparse coding with learned dictionaries resembles a specific variational autoencoder, where the decoder is a linear function and the encoder is a sparse coding algorithm. Next, we show that dictionary learning can also benefit from computational advancements introduced in the context of deep learning, such as parallelism and as stochastic optimization. Finally, we show that regularization by dictionaries achieves competitive performance in computed tomography (CT) reconstruction comparing to state-of-the-art model based and data driven approaches.

arXiv:2108.11833

DeepGene Transformer: Transformer for the gene expression-based classification of cancer subtypes

Cancer and its subtypes constitute approximately 30% of all causes of death globally and display a wide range of heterogeneity in terms of clinical and molecular responses to therapy. Molecular subtyping has enabled the use of precision medicine to overcome these challenges and provide significant biological insights to predict prognosis and improve clinical decision-making. Over the past decade, conventional machine learning (ML) and deep learning (DL) algorithms have been widely espoused for the classification of cancer subtypes from gene expression datasets. However, these methods are potentially biased toward the identification of cancer biomarkers. Hence, an end-to-end deep learning approach, DeepGene Transformer, is proposed which addresses the complexity of high-dimensional gene expression with a multi-head self-attention module by identifying relevant biomarkers across multiple cancer subtypes without requiring feature selection as a pre-requisite for the current classification algorithms. Comparative analysis reveals that the proposed DeepGene Transformer outperformed the commonly used traditional and state-of-the-art classification algorithms and can be considered an efficient approach for classifying cancer and its subtypes, indicating that any improvement in deep learning models in computational biologists can be reflected well in this domain as well.

arXiv:2108.11371

Understanding the Generalization of Adam in Learning Neural Networks with Proper Regularization

Adaptive gradient methods such as Adam have gained increasing popularity in deep learning optimization. However, it has been observed that compared with (stochastic) gradient descent, Adam can converge to a different solution with a significantly worse test error in many deep learning applications such as image classification, even with a fine-tuned regularization. In this paper, we provide a theoretical explanation for this phenomenon: we show that in the nonconvex setting of learning over-parameterized two-layer convolutional neural networks starting from the same random initialization, for a class of data distributions (inspired from image data), Adam and gradient descent (GD) can converge to different global solutions of the training objective with provably different generalization errors, even with weight decay regularization. In contrast, we show that if the training objective is convex, and the weight decay regularization is employed, any optimization algorithms including Adam and GD will converge to the same solution if the training is successful. This suggests that the inferior generalization performance of Adam is fundamentally tied to the nonconvex landscape of deep learning optimization.

arXiv:2108.11057

Opportunistic Emulation of Computationally Expensive Simulations via Deep Learning

With the underlying aim of increasing efficiency of computational modelling pertinent for managing & protecting the Great Barrier Reef, we perform a preliminary investigation on the use of deep neural networks for opportunistic model emulation of APSIM models by repurposing an existing large dataset containing outputs of APSIM model runs. The dataset has not been specifically tailored for the model emulation task. We employ two neural network architectures for the emulation task: densely connected feed-forward neural network (FFNN), and gated recurrent unit feeding into FFNN (GRU-FFNN), a type of a recurrent neural network. Various configurations of the architectures are trialled. A minimum correlation statistic is used to identify clusters of APSIM scenarios that can be aggregated to form training sets for model emulation. We focus on emulating 4 important outputs of the APSIM model: runoff, soil_loss, DINrunoff, Nleached. The GRU-FFNN architecture with three hidden layers and 128 units per layer provides good emulation of runoff and DINrunoff. However, soil_loss and Nleached were emulated relatively poorly under a wide range of the considered architectures; the emulators failed to capture variability at higher values of these two outputs. While the opportunistic data available from past modelling activities provides a large and useful dataset for exploring APSIM emulation, it may not be sufficiently rich enough for successful deep learning of more complex model dynamics. Design of Computer Experiments may be required to generate more informative data to emulate all output variables of interest. We also suggest the use of synthetic meteorology settings to allow the model to be fed a wide range of inputs. These need not all be representative of normal conditions, but can provide a denser, more informative dataset from which complex relationships between input and outputs can be learned.

arXiv:2108.10828

Physics-Informed Deep Learning: A Promising Technique for System Reliability Assessment

Considerable research has been devoted to deep learning-based predictive models for system prognostics and health management in the reliability and safety community. However, there is limited study on the utilization of deep learning for system reliability assessment. This paper aims to bridge this gap and explore this new interface between deep learning and system reliability assessment by exploiting the recent advances of physics-informed deep learning. Particularly, we present an approach to frame system reliability assessment in the context of physics-informed deep learning and discuss the potential value of physics-informed generative adversarial networks for the uncertainty quantification and measurement data incorporation in system reliability assessment. The proposed approach is demonstrated by three numerical examples involving a dual-processor computing system. The results indicate the potential value of physics-informed deep learning to alleviate computational challenges and combine measurement data and mathematical models for system reliability assessment.

arXiv:2108.10682

Hybrid deep learning methods for phenotype prediction from clinical notes

Identifying patient cohorts from clinical notes in secondary electronic health records is a fundamental task in clinical information management. However, with the growing number of clinical notes, it becomes challenging to analyze the data manually for phenotype detection. Automatic extraction of clinical concepts would helps to identify the patient phenotypes correctly. This paper proposes a novel hybrid model for automatically extracting patient phenotypes using natural language processing and deep learning models to determine the patient phenotypes without dictionaries and human intervention. The model is based on a neural bidirectional sequence model (BiLSTM or BiGRU) and a CNN layer for phenotypes identification. An extra CNN layer is run parallel to the hybrid model to extract more features related to each phenotype. We used pre-trained embeddings such as FastText and Word2vec separately as the input layers to evaluate other embedding's performance. Experimental results using MIMIC III database in internal comparison demonstrate that the proposed model achieved significant performance improvement over existing models. The enhanced version of our model with an extra CNN layer obtained a relatively higher F1-score than the original hybrid model. We also showed that BiGRU layer with FastText embedding had better performance than BiLSTM layer to identify patient phenotypes.

arXiv:2108.10573

The staircase property: How hierarchical structure can guide deep learning

This paper identifies a structural property of data distributions that enables deep neural networks to learn hierarchically. We define the "staircase" property for functions over the Boolean hypercube, which posits that high-order Fourier coefficients are reachable from lower-order Fourier coefficients along increasing chains. We prove that functions satisfying this property can be learned in polynomial time using layerwise stochastic coordinate descent on regular neural networks -- a class of network architectures and initializations that have homogeneity properties. Our analysis shows that for such staircase functions and neural networks, the gradient-based algorithm learns high-level features by greedily combining lower-level features along the depth of the network. We further back our theoretical results with experiments showing that staircase functions are also learnable by more standard ResNet architectures with stochastic gradient descent. Both the theoretical and experimental results support the fact that staircase properties have a role to play in understanding the capabilities of gradient-based learning on regular networks, in contrast to general polynomial-size networks that can emulate any SQ or PAC algorithms as recently shown.

arXiv:2108.10335공식 코드 · Papers with Code

edge-SR: Super-Resolution For The Masses

Classic image scaling (e.g. bicubic) can be seen as one convolutional layer and a single upscaling filter. Its implementation is ubiquitous in all display devices and image processing software. In the last decade deep learning systems have been introduced for the task of image super-resolution (SR), using several convolutional layers and numerous filters. These methods have taken over the benchmarks of image quality for upscaling tasks. Would it be possible to replace classic upscalers with deep learning architectures on edge devices such as display panels, tablets, laptop computers, etc.? On one hand, the current trend in Edge-AI chips shows a promising future in this direction, with rapid development of hardware that can run deep-learning tasks efficiently. On the other hand, in image SR only few architectures have pushed the limit to extreme small sizes that can actually run on edge devices at real-time. We explore possible solutions to this problem with the aim to fill the gap between classic upscalers and small deep learning configurations. As a transition from classic to deep-learning upscaling we propose edge-SR (eSR), a set of one-layer architectures that use interpretable mechanisms to upscale images. Certainly, a one-layer architecture cannot reach the quality of deep learning systems. Nevertheless, we find that for high speed requirements, eSR becomes better at trading-off image quality and runtime performance. Filling the gap between classic and deep-learning architectures for image upscaling is critical for massive adoption of this technology. It is equally important to have an interpretable system that can reveal the inner strategies to solve this problem and guide us to future improvements and better understanding of larger networks.