An official website of the United States government
Spatial data volumes have grown exponentially over the past several years. The number of domains that spatial data are extensively leveraged include atmospheric sciences, environmental monitoring, ecological modeling, epidemiology, sociology, commerce, and social media among others. These data are often used to understand phenomena and inform decision-making by fitting models to them. In this study, we present our methodology to fit models at scale over spatial data. Our methodology encompasses segmentation, spatial similarity based on the dataset(s) under consideration, and transfer learning schemes that are informed by the spatial similarity to train models faster while utilizing fewer resources. We consider several model fitting algorithms and execution within containerized environments as we profile the suitability of our methodology. Our benchmarks validate the suitability of our methodology to facilitate faster, resource-efficient training of models over spatial data.
more »
« less
Alleviating Resource Requirements for Spatial Deep Learning Workloads
Spatial data volumes have increased exponentially over the past couple of decades. This growth has been fueled by networked observational devices, remote sensing sources such as satellites, and simulations that characterize spatiotemporal dynamics of phenomena (e.g., climate). Manual inspection of these data becomes unfeasible at such scales. Fitting models to the data offer an avenue to extract patterns from the data, make predictions, and leverage them to understand phenomena and decision-making. Innovations in deep learning and their ability to capture non-linear interactions between features make them particularly relevant for spatial datasets. However, deep learning workloads tend to be resource-intensive. In this study, we design and contrast transfer learning schemes to substantively alleviate resource requirements for training deep learning models over spatial data at scale. We profile the suitability of our methodology using deep networks built over satellite datasets and gridded data. Empirical benchmarks demonstrate that our spatiotemporally aligned transfer learning scheme ensures ~2.87-5.3 fold reduction in completion times for each model without sacrificing on the accuracy of the models.
more »
« less
- Award ID(s):
- 1931363
- PAR ID:
- 10352213
- Date Published:
- Journal Name:
- 22nd IEEE International Symposium on Cluster, Cloud and Internet Computing (CCGrid)
- Page Range / eLocation ID:
- 452 to 462
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Modern NLP applications have enjoyed a great boost utilizing neural networks models. Such deep neural models, however, are not applicable to most human languages due to the lack of annotated training data for various NLP tasks. Cross-lingual transfer learning (CLTL) is a viable method for building NLP models for a low-resource target language by leveraging labeled data from other (source) languages. In this work, we focus on the multilingual transfer setting where training data in multiple source languages is leveraged to further boost target language performance. Unlike most existing methods that rely only on language-invariant features for CLTL, our approach coherently utilizes both language invariant and language-specific features at instance level. Our model leverages adversarial networks to learn language-invariant features, and mixture-of-experts models to dynamically exploit the similarity between the target language and each individual source language1. This enables our model to learn effectively what to share between various languages in the multilingual setup. Moreover, when coupled with unsupervised multilingual embeddings, our model can operate in a zero-resource setting where neither target language training data nor cross-lingual resources are available. Our model achieves significant performance gains over prior art, as shown in an extensive set of experiments over multiple text classification and sequence tagging.more » « less
-
In the age of Big Genomics Data, institutions such as the National Human Genome Research Institute (NHGRI) are challenged in their efforts to share volumes of data between researchers, a process that has been plagued by unreliable transfers and slow speeds. These occur due to throughput bottlenecks of traditional transfer technologies. Two factors that affect the effciency of data transmission are the channel bandwidth and the amount of data. Increasing the bandwidth is one way to transmit data effciently, but might not always be possible due to resource limitations. Another way to maximize channel utilization is by decreasing the bits needed for transmission of a dataset. Traditionally, transmission of big genomic data between two geographical locations is done using general-purpose protocols, such as hypertext transfer protocol (HTTP) and file transfer protocol (FTP) secure. In this paper, we present a novel deep learning-based data minimization algorithm that 1) minimizes the datasets during transfer over the carrier channels; 2) protects the data from the man-in-the-middle (MITM) and other attacks by changing the binary representation (content-encoding) several times for the same dataset: we assign different codewords to the same character in different parts of the dataset. Our data minimization strategy exploits the alphabet limitation of DNA sequences and modifies the binary representation (codeword) of dataset characters using deep learning-based convolutional neural network (CNN) to ensure a minimum of code word uses to the high frequency characters at different time slots during the transfer time. This algorithm ensures transmission of big genomic DNA datasets with minimal bits and latency and yields an effcient and expedient process. Our tested heuristic model, simulation, and real implementation results indicate that the proposed data minimization algorithm is up to 99 times faster and more secure than the currently used content-encoding scheme used in HTTP of the HTTP content-encoding scheme and 96 times faster than FTP on tested datasets. The developed protocol in C# will be available to the wider genomics community and domain scientists.more » « less
-
The growing number of AI-driven applications in mobile devices has led to solutions that integrate deep learning models with the available edge-cloud resources. Due to multiple benefits such as reduction in on-device energy consumption, improved latency, improved network usage, and certain privacy improvements, split learning, where deep learning models are split away from the mobile device and computed in a distributed manner, has become an extensively explored topic. Incorporating compression-aware methods (where learning adapts to compression level of the communicated data) has made split learning even more advantageous. This method could even offer a viable alternative to traditional methods, such as federated learning techniques. In this work, we develop an adaptive compression-aware split learning method (“deprune”) to improve and train deep learning models so that they are much more network-efficient, which would make them ideal to deploy in weaker devices with the help of edge-cloud resources. This method is also extended (“prune”) to very quickly train deep learning models through a transfer learning approach, which tradesoff little accuracy for much more network-efficient inference abilities. We show that the “deprune” method can reduce network usage by 4× when compared with a split-learning approach (that does not use our method) without loss of accuracy, while also improving accuracy over compression-aware split-learning by up to 4 percent. Lastly, we show that the “prune” method can reduce the training time for certain models by up to 6× without affecting the accuracy when compared against a compression-aware split-learning approach.more » « less
-
null (Ed.)The prevalence of deep learning has drawn attention to the privacy protection of sensitive data. Various privacy threats have been presented, where an adversary can steal model owners' private data. Meanwhile, countermeasures have also been introduced to achieve privacy-preserving deep learning. However, most studies only focused on data privacy during training, and ignored privacy during inference. In this paper, we devise a new set of attacks to compromise the inference data privacy in collaborative deep learning systems. Specifically, when a deep neural network and the corresponding inference task are split and distributed to different participants, one malicious participant can accurately recover an arbitrary input fed into this system, even if he has no access to other participants' data or computations, or to prediction APIs to query this system. We evaluate our attacks under different settings, models and datasets, to show their effectiveness and generalization. We also study the characteristics of deep learning models that make them susceptible to such inference privacy threats. This provides insights and guidelines to develop more privacy-preserving collaborative systems and algorithms.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/CCGrid54584.2022.00055
