Quantifying Uncertainty in Discrete-Continuous and Skewed Data with Bayesian Deep Learning

Vandal, Thomas; Kodra, Evan; Dy, Jennifer G.; Ganguly, Sangram; Nemani, Ramakrishna R.; Ganguly, Auroop R.

doi:10.1145/3219819.3219996

Cited by 43 publications

(47 citation statements)

References 41 publications

(45 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, the proposed method was applied with the MODIS image with the resolution of about 500 m and the Landsat TM image with the resolution of 30 m in the experiments. For both of them, there are spectral bands with finer spatial resolutions, including the first and second bands of MODIS that have a spatial resolution about 250 m and the panchromatic band of Landsat with the spatial resolution of 15 m. Incorporating these finer‐resolution bands is expected to increase the accuracy of a SRM analysis (Li et al, ; Nguyen et al, ; Vandal et al, ). Second, the proposed method could be applied on other remotely sensed images, such as the newly launched Sentinel‐2 imagery that has a finest spatial resolution of 10 m. For remotely sensed imagery with the spatial resolution of several meters or finer than 1 m, the proposed method can also be applied to further improve the RWW estimated.…”

Section: Discussionmentioning

confidence: 99%

Measuring River Wetted Width From Remotely Sensed Imagery at the Subpixel Scale With a Deep Convolutional Neural Network

Ling

Boyd

et al. 2019

Water Resources Research

View full text Add to dashboard Cite

River wetted width (RWW) is an important variable in the study of river hydrological and biogeochemical processes. Presently, RWW is often measured from remotely sensed imagery, and the accuracy of RWW estimation is typically low when coarse spatial resolution imagery is used because river boundaries often run through pixels that represent a region that is a mixture of water and land. Thus, when conventional hard classification methods are used in the estimation of RWW, the mixed pixel problem can become a large source of error. To address this problem, this paper proposes a novel approach to measure RWW at the subpixel scale. Spectral unmixing is first applied to the imagery to obtain a water fraction image that indicates the proportional coverage of water in image pixels. A fine spatial resolution river map from which RWW may be estimated is then produced from the water fraction image by superresolution mapping (SRM). In the SRM analysis, a deep convolutional neural network is used to eliminate the negative effects of water fraction errors and reconstruct the geographical distribution of water. The proposed approach is assessed in two experiments, with the results demonstrating that the convolutional neural network‐based SRM model can effectively estimate subpixel scale details of rivers and that the accuracy of RWW estimation is substantially higher than that obtained from the use of a conventional hard image classification. The improvement shows that the proposed method has great potential to derive more accurate RWW values from remotely sensed imagery.

show abstract

Section: Discussionmentioning

confidence: 99%

Measuring River Wetted Width From Remotely Sensed Imagery at the Subpixel Scale With a Deep Convolutional Neural Network

Ling

Boyd

et al. 2019

Water Resources Research

View full text Add to dashboard Cite

show abstract

“…Distributed computing is becoming ever faster [11,12,13]. Another important trend is that CNN deep learning becomes better [14,15,16,17].…”

Section: Ai-based Big Data Processing: Theorymentioning

confidence: 99%

Spark-Based Big Data Processing: AI to Decide on Opening a SiU9 Contract Position

Lomakin¹,

Kravcheny²,

Maramygin³

et al. 2020

Proceedings of the International Scientific Conference "Far East Con" (ISCFEC 2020)

View full text Add to dashboard Cite

This paper presents the theory behind using AI for big data processing. It considers an artificial neural network designed to make decisions whether to open a position on a SiU9 futures contract; to that end, Spark-based big data processing is applied. It is hypothesized and proven herein that the futures closing price can be predicted by the developed AI trained on a dataset containing web-parsed digitized 'news-related' fluctuations as well as time-series Japanese candlesticks of a SiU9 futures contract on a 15-minute timeframe. Research has shown that in the modern world, the ever-larger bulk of stock transactions are done by using trading systems (trading robots). More and more of them use AI. The novelty hereof is that the futures contract price is predicted by an AI trained not only on quantities and Japanese candlestick parameters, but also on the digitized news-related fluctuations collected by Skraper from websites. The dataset generated that was was fed to a perceptron that had a 305-parameter input layer, two hidden layers having (100 parameters and 10 parameters), and an output layer that gave the predicted price. The perceptron was created and run on the Deductor platform, while Skraper was a Python program in the Spark framework. The research team further analyzed how accurately the ANN could predict the price of a financial instrument, the SiU9 futures contract, on a Russian stock exchange over a 15-minute timeframe. The study has effectively produced an ANN-based trading algorithm for opening long and short SiU9 positions in the fixed-term stock market; the algorithm has a good predictive accuracy.

show abstract

“…Faster distributed computation is another trend in the developing area [11,12,13]. Improved deep learning of convolution neural network is another important direction of development [14,15,16,17].…”

Section: Research Of Assessment Theoretical Basis and Financial Risk mentioning

confidence: 99%

AI-System of Stock Exchange Trading Robot for Financial Risk Hedging

Lomakin¹,

Popov²,

Shohneh³

et al. 2020

Proceedings of the International Scientific Conference "Far East Con" (ISCFEC 2020)

View full text Add to dashboard Cite

The article considered the theoretical basis for the calculation of financial risks, their hedging with the help of artificial intelligence. The article presents a neural network for the assessment of financial risk for atime sequence of VAR-methods with a view to mitigate it when engaging in stock exchange trade. A hypothesis was formulated and proved that with the help of the developed AI-system that had been taught on the basis of a combined data sample with digitilized "new fluctuations" information from web-sites and time sequence candlestick charts for SiU9 US Dollar futures contracts in a 15-minute timeframe, it is possible to increase the accuracy of a futures price forecast and ensure the assessment of financial risks using the VAR-method for hedging an uncovered position with a PUT option. As is shown by scientific research, today the majority of stock operations are conducted using trade systems, or trade robots, and their number is continuously increasing. Among trade systems, we can single out the ones using artificial intelligence (AI). The novelty of the research results from the fact that in order to forecast the price of a futures contract, both candlestick charts parameters and size and digitilized "news fluctuations" received by Skraper programme from web-sites were used to teach the neural network. The created data set was used in the perceptron with 305 parameters in the input layer, 2 hidden layers by 100 10 parameters correspondingly and an output layer with one parameter, the target price. The perceptron was created and used on the Deductor platform, and the Python-based Skraper programme was put into Spark framework and was employed with the help of parallel calculations. The results yielded by the neural network were analysed concerning the accuracy of the forecast of the price of a financial instrument, a SiU9 futures contract at MoEx, a Russian stock exchange, when working in a 15-minute timeframe. The hedging mechanism for an uncovered long position of a SiU9 futures was activate by buying two PUT Si-9 options with an increasing risk parameter that had been calculated using the VaR-method. As a result of the conducted research, a neural network trade algorithm was created for speculative trading at the stock exchange with a SiU9 USD futures contract with high accuracy and minimal loss risks.

show abstract

Quantifying Uncertainty in Discrete-Continuous and Skewed Data with Bayesian Deep Learning

Cited by 43 publications

References 41 publications

Measuring River Wetted Width From Remotely Sensed Imagery at the Subpixel Scale With a Deep Convolutional Neural Network

Measuring River Wetted Width From Remotely Sensed Imagery at the Subpixel Scale With a Deep Convolutional Neural Network

Spark-Based Big Data Processing: AI to Decide on Opening a SiU9 Contract Position

AI-System of Stock Exchange Trading Robot for Financial Risk Hedging

Contact Info

Product

Resources

About