Machine Learning (ML) computation requires feeding input data for the models to ingest. Traditionally, input data processing happens on the same host as the ML computation. The input data processing can however become a bottleneck of the ML computation if there are insufficient resources to process data quickly enough. This slows down the ML computation and wastes valuable and scarce ML hardware (e.g. GPUs and TPUs) used by the ML computation. In this paper, we present tf.data service, a disaggregated input data processing service built on top of tf.data. Our work goes beyond describing the design and implementation of a new system which disaggregates preprocessing from ML computation and presents: (1) empirical evidence based on production workloads for the need of disaggregation, as well as quantitative evaluation of the impact disaggregation has on the performance and cost of production workloads, (2) benefits of disaggregation beyond horizontal scaling, (3) analysis of tf.data service's adoption at Google, the lessons learned during building and deploying the system and potential future lines of research opened up by our work. We demonstrate that horizontally scaling data processing using tf.data service helps remove input bottlenecks, achieving speedups of up to 110x and job cost reductions of up to 89x. We further show that tf.data service can support computation reuse through data sharing across ML jobs with identical data processing pipelines (e.g. hyperparameter tuning jobs), incurring no performance penalty and reducing overall resource cost. Finally, we show that tf.data service advanced features can benefit performance of non-input bound jobs; in particular, coordinated data reads through tf.data service can yield up to 2x speedups and job cost savings for NLP jobs.
相關內容
Processing 是一門開源編程語言和與之配套的集成開發環境（IDE）的名稱。Processing 在電子藝術和視覺設計社區被用來教授編程基礎，并運用于大量的新媒體和互動藝術作品中。
By transferring knowledge from large, diverse, task-agnostic datasets, modern machine learning models can solve specific downstream tasks either zero-shot or with small task-specific datasets to a high level of performance. While this capability has been demonstrated in other fields such as computer vision, natural language processing or speech recognition, it remains to be shown in robotics, where the generalization capabilities of the models are particularly critical due to the difficulty of collecting real-world robotic data. We argue that one of the keys to the success of such general robotic models lies with open-ended task-agnostic training, combined with high-capacity architectures that can absorb all of the diverse, robotic data. In this paper, we present a model class, dubbed Robotics Transformer, that exhibits promising scalable model properties. We verify our conclusions in a study of different model classes and their ability to generalize as a function of the data size, model size, and data diversity based on a large-scale data collection on real robots performing real-world tasks. The project's website and videos can be found at robotics-transformer.github.io
Fuzzy inference engine, as one of the most important components of fuzzy systems, can obtain some meaningful outputs from fuzzy sets on input space and fuzzy rule base using fuzzy logic inference methods. In multi-input-single-output (MISO) fuzzy systems, in order to enhance the computational efficiency of fuzzy inference engine, this paper aims mainly to investigate three MISO fuzzy hierarchical inference engines employed fuzzy implications satisfying the law of importation with aggregation functions (LIA). We firstly find some aggregation functions for well-known fuzzy implications such that they satisfy LIA. The fuzzy implication satisfying LIA with respect to a given aggregation function is then characterized. Finally, three fuzzy hierarchical inference engines in MISO fuzzy systems are constructed according to aforementioned theoretical developments. Three examples are also provided to illustrate our theoretical arguments.
This work addresses the problems of (a) designing utilization measurements of trained artificial intelligence (AI) models and (b) explaining how training data are encoded in AI models based on those measurements. The problems are motivated by the lack of explainability of AI models in security and safety critical applications, such as the use of AI models for classification of traffic signs in self-driving cars. We approach the problems by introducing theoretical underpinnings of AI model utilization measurement and understanding patterns in utilization-based class encodings of traffic signs at the level of computation graphs (AI models), subgraphs, and graph nodes. Conceptually, utilization is defined at each graph node (computation unit) of an AI model based on the number and distribution of unique outputs in the space of all possible outputs (tensor-states). In this work, utilization measurements are extracted from AI models, which include poisoned and clean AI models. In contrast to clean AI models, the poisoned AI models were trained with traffic sign images containing systematic, physically realizable, traffic sign modifications (i.e., triggers) to change a correct class label to another label in a presence of such a trigger. We analyze class encodings of such clean and poisoned AI models, and conclude with implications for trojan injection and detection.
With the rapid growth in the number of devices of the Internet of Things (IoT), the volume and types of stream data are rapidly increasing in the real world. Unfortunately, the stream data has the characteristics of infinite and periodic volatility in the real world, which cause problems with the inefficient stream processing tasks. In this study, we report our recent efforts on this issue, with a focus on simulating stream data. Firstly, we explore the characteristics of the real-world stream data of the IoT, which helps us to understand the stream data in the real world. Secondly, the pipeline of simulating stream data is proposed, which can accurately and efficiently simulate the characteristics of the stream data to improve efficiency for specific tasks. Finally, we design and implement a novel framework that can simulate various stream data for related stream processing tasks. To verify the validity of the proposed framework, we apply this framework to stream processing task running in the stream processing system. The experimental results reveal that the related stream processing task is accelerated by at least 24 times using our proposed simulation framework with the premise of ensuring volatility and trends of stream data.
We propose a monitoring strategy for efficient and robust estimation of disease prevalence and case numbers within closed and enumerated populations such as schools, workplaces, or retirement communities. The proposed design relies largely on voluntary testing, notoriously biased (e.g., in the case of COVID-19) due to non-representative sampling. The approach yields unbiased and comparatively precise estimates with no assumptions about factors underlying selection of individuals for voluntary testing, building on the strength of what can be a small random sampling component. This component unlocks a previously proposed "anchor stream" estimator, a well-calibrated alternative to classical capture-recapture (CRC) estimators based on two data streams. We show here that this estimator is equivalent to a direct standardization based on "capture", i.e., selection (or not) by the voluntary testing program, made possible by means of a key parameter identified by design. This equivalency simultaneously allows for novel two-stream CRC-like estimation of general means (e.g., of continuous variables such as antibody levels or biomarkers). For inference, we propose adaptations of a Bayesian credible interval when estimating case counts and bootstrapping when estimating means of continuous variables. We use simulations to demonstrate significant precision benefits relative to random sampling alone.
The classification of sleep stages plays a crucial role in understanding and diagnosing sleep pathophysiology. Sleep stage scoring relies heavily on visual inspection by an expert that is time consuming and subjective procedure. Recently, deep learning neural network approaches have been leveraged to develop a generalized automated sleep staging and account for shifts in distributions that may be caused by inherent inter/intra-subject variability, heterogeneity across datasets, and different recording environments. However, these networks ignore the connections among brain regions, and disregard the sequential connections between temporally adjacent sleep epochs. To address these issues, this work proposes an adaptive product graph learning-based graph convolutional network, named ProductGraphSleepNet, for learning joint spatio-temporal graphs along with a bidirectional gated recurrent unit and a modified graph attention network to capture the attentive dynamics of sleep stage transitions. Evaluation on two public databases: the Montreal Archive of Sleep Studies (MASS) SS3; and the SleepEDF, which contain full night polysomnography recordings of 62 and 20 healthy subjects, respectively, demonstrates performance comparable to the state-of-the-art (Accuracy: 0.867;0.838, F1-score: 0.818;0.774 and Kappa: 0.802;0.775, on each database respectively). More importantly, the proposed network makes it possible for clinicians to comprehend and interpret the learned connectivity graphs for sleep stages.
Unsupervised domain adaptation has recently emerged as an effective paradigm for generalizing deep neural networks to new target domains. However, there is still enormous potential to be tapped to reach the fully supervised performance. In this paper, we present a novel active learning strategy to assist knowledge transfer in the target domain, dubbed active domain adaptation. We start from an observation that energy-based models exhibit free energy biases when training (source) and test (target) data come from different distributions. Inspired by this inherent mechanism, we empirically reveal that a simple yet efficient energy-based sampling strategy sheds light on selecting the most valuable target samples than existing approaches requiring particular architectures or computation of the distances. Our algorithm, Energy-based Active Domain Adaptation (EADA), queries groups of targe data that incorporate both domain characteristic and instance uncertainty into every selection round. Meanwhile, by aligning the free energy of target data compact around the source domain via a regularization term, domain gap can be implicitly diminished. Through extensive experiments, we show that EADA surpasses state-of-the-art methods on well-known challenging benchmarks with substantial improvements, making it a useful option in the open world. Code is available at //github.com/BIT-DA/EADA.
Graph Neural Networks (GNNs) have been shown to be effective models for different predictive tasks on graph-structured data. Recent work on their expressive power has focused on isomorphism tasks and countable feature spaces. We extend this theoretical framework to include continuous features - which occur regularly in real-world input domains and within the hidden layers of GNNs - and we demonstrate the requirement for multiple aggregation functions in this context. Accordingly, we propose Principal Neighbourhood Aggregation (PNA), a novel architecture combining multiple aggregators with degree-scalers (which generalize the sum aggregator). Finally, we compare the capacity of different models to capture and exploit the graph structure via a novel benchmark containing multiple tasks taken from classical graph theory, alongside existing benchmarks from real-world domains, all of which demonstrate the strength of our model. With this work, we hope to steer some of the GNN research towards new aggregation methods which we believe are essential in the search for powerful and robust models.
Existing few-shot learning (FSL) methods assume that there exist sufficient training samples from source classes for knowledge transfer to target classes with few training samples. However, this assumption is often invalid, especially when it comes to fine-grained recognition. In this work, we define a new FSL setting termed few-shot fewshot learning (FSFSL), under which both the source and target classes have limited training samples. To overcome the source class data scarcity problem, a natural option is to crawl images from the web with class names as search keywords. However, the crawled images are inevitably corrupted by large amount of noise (irrelevant images) and thus may harm the performance. To address this problem, we propose a graph convolutional network (GCN)-based label denoising (LDN) method to remove the irrelevant images. Further, with the cleaned web images as well as the original clean training images, we propose a GCN-based FSL method. For both the LDN and FSL tasks, a novel adaptive aggregation GCN (AdarGCN) model is proposed, which differs from existing GCN models in that adaptive aggregation is performed based on a multi-head multi-level aggregation module. With AdarGCN, how much and how far information carried by each graph node is propagated in the graph structure can be determined automatically, therefore alleviating the effects of both noisy and outlying training samples. Extensive experiments show the superior performance of our AdarGCN under both the new FSFSL and the conventional FSL settings.
Recent advances in 3D fully convolutional networks (FCN) have made it feasible to produce dense voxel-wise predictions of volumetric images. In this work, we show that a multi-class 3D FCN trained on manually labeled CT scans of several anatomical structures (ranging from the large organs to thin vessels) can achieve competitive segmentation results, while avoiding the need for handcrafting features or training class-specific models. To this end, we propose a two-stage, coarse-to-fine approach that will first use a 3D FCN to roughly define a candidate region, which will then be used as input to a second 3D FCN. This reduces the number of voxels the second FCN has to classify to ~10% and allows it to focus on more detailed segmentation of the organs and vessels. We utilize training and validation sets consisting of 331 clinical CT images and test our models on a completely unseen data collection acquired at a different hospital that includes 150 CT scans, targeting three anatomical organs (liver, spleen, and pancreas). In challenging organs such as the pancreas, our cascaded approach improves the mean Dice score from 68.5 to 82.2%, achieving the highest reported average score on this dataset. We compare with a 2D FCN method on a separate dataset of 240 CT scans with 18 classes and achieve a significantly higher performance in small organs and vessels. Furthermore, we explore fine-tuning our models to different datasets. Our experiments illustrate the promise and robustness of current 3D FCN based semantic segmentation of medical images, achieving state-of-the-art results. Our code and trained models are available for download: //github.com/holgerroth/3Dunet_abdomen_cascade.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191