Publications

Lossy Light Field Compression Using Modern Deep Learning and Domain Randomization Techniques Permalink

Published in YorkSpace, 2022

Lossy data compression is a particular type of informational encoding utilizing approximations in order to efficiently tradeoff accuracy in favour of smaller file sizes. The transmission and storage of images is a typical example of this in the modern digital world. However the reconstructed images often suffer from degradation and display observable visual artifacts. Convolutional Neural Networks (CNNs) have garnered much attention in all corners of Computer Vision (CV), including the tasks of image compression and artifact reduction. We study how lossy compression can be extended to higher dimensional images with varying viewpoints, known as light fields. Domain Randomization (DR) is explored in detail, and used to generate the largest light field dataset we are aware of, to be used as training data. We formulate the task of compression under the frameworks of neural networks and calculate a quantization tensor for the 4-D Discrete Cosine Transform (DCT) coefficients of the light fields. In order to accurately train the network, a high degree approximation to the rounding operation is introduced. In addition, we present a multi-resolution convolutional-based light field enhancer, producing average gains of 0.854 db in Peak Signal-to-Noise Ratio (PSNR), and 0.0338 in Structual Similarity Index Measure (SSIM) over the base model, across a wide range of bitrates.

Valtchev, S.Z. and Wu, J. (2022). "Lossy Light Field Compression Using Modern Deep Learning and Domain Randomization Techniques", YorkSpace. 2022-12-14. http://zarkonium.github.io/files/Lossy_Light_Field_Compression_Using_Modern_Deep_Learning_and_Domain_Randomization_Techniques.pdf

Managing SARS-CoV-2 Testing in Schools with an Artificial Intelligence Model and Application Developed by Simulation Data

Published in MDPI Electronics, 2021

Research on SARS-CoV-2 and its social implications have become a major focus to interdisciplinary teams worldwide. As interest in more direct solutions, such as mass testing and vaccination grows, several studies appear to be dedicated to the operationalization of those solutions, leveraging both traditional and new methodologies, and, increasingly, the combination of both. This research examines the challenges anticipated for preventative testing of SARS-CoV-2 in schools and proposes an artificial intelligence (AI)-powered agent-based model crafted specifically for school scenarios. This research shows that in the absence of real data, simulation-based data can be used to develop an artificial intelligence model for the application of rapid assessment of school testing policies.

Valtchev, S.Z., Asgary, A, Chen, M., Cronemberger, F.B., Najafabadi, M., Cojocaru, M.G. and Wu, J. (2021). "Managing SARS-CoV-2 Testing in Schools with an Artificial Intelligence Model and Application Developed by Simulation Data", MDPI Electronics. 10(14). http://zarkonium.github.io/files/Managing_SARS-CoV-2_Testing_in_Schools_with_an_Art.pdf

Domain Randomization for Neural Network Classification Permalink

Published in SpringerOpen Journal of Big Data, 2021

Large data requirements are often the main hurdle in training neural networks. Synthetic data is a cheap and efficient solution to assemble such large datasets. Using domain randomization, we show that a sufficiently well generated synthetic image dataset can be used to train a neural network classifier, achieving accuracy levels as high as 88% on 2 category classification. We show that the most important domain randomization parameter is a large variety of subjects, while secondary parameters such as lighting and textures are not. Based on our results, there is reason to believe that models trained on domain randomized images transfer to new domains better than those trained on real photos. Model performance seems to diminish slightly as the number of categories increases.

Valtchev, S.Z. and Wu, J. (2021). "Domain Randomization for Neural Network Classification", SpringerOpen Journal of Big Data. 8:94. http://zarkonium.github.io/files/Domain_Randomization_for_Neural_Network_Classification_Published.pdf

Artificial Intelligence Model of Drive-Through Vaccination Simulation

Published in MDPI International Journal of Environmental and Public Health, 2020

Planning for mass vaccination against SARS-Cov-2 is ongoing in many countries considering that vaccine will be available for the general public in the near future. Rapid mass vaccination while a pandemic is ongoing requires the use of traditional and new temporary vaccination clinics. Use of drive-through has been suggested as one of the possible effective temporary mass vaccinations among other methods. In this study, we present a machine learning model that has been developed based on a big dataset derived from 125K runs of a drive-through mass vaccination simulation tool. The results show that the model is able to reasonably well predict the key outputs of the simulation tool. Therefore, the model has been turned to an online application that can help mass vaccination planners to assess the outcomes of different types of drive-through mass vaccination facilities much faster.

Asgary, A., Valtchev, S.Z., Chen, M., Najafabadi, M. and Wu, J. (2020). "Artificial Intelligence Model of Drive-Through Vaccination Simulation", MDPI International Journal of Environmental and Public Health. 18(1):268. http://zarkonium.github.io/files/Artificial_Intelligence_Model_of_Drive-Through_Vac.pdf

A Correlation Measure for Multi-Dimensional Signals

Similarity measures are largely needed for a variety of tasks such as anomaly detection, classification and forecasting. In this paper we explore the shortcomings of current multi-dimensional correlations measures, namely the RV coefficient and the normalized RV coefficient. When a particular dimension is positively correlated between 2 samples, while another is negatively correlated, these methods can provide undesired results. We propose a new measure, the DwC (Deviation-weighted Correlation) coefficient, which overcomes these limitations. Our measure correctly incorporates positive and negative correlation on a dimension-by-dimension basis, ultimately providing a more intuitive and useful measure that generalizes to higher dimensions for the comparison of arbitrary matrices. The measure also holds some scaling properties which become useful in the presence of noise. Lastly we provide an example using accelerometer data, to classify common human activities based on maximum DwC between predefined templates and the data.

Valtchev, S.Z. and Wu, J. (2019). "A Correlation Measure for Multi-Dimensional Signals", Manuscript. http://zarkonium.github.io/files/A_Similarity_Measure_for_Multi_Dimensional_Signals.pdf

Identifying Volatility Regimes in Bitcoin Prices

We apply clustering techniques to volatility and market sentiment measurements of historical Bitcoin prices in the aim of identifying hidden structural patterns separating different regions of our data. Using these regimes should help a fund allocate between trading strategies depending on market conditions, based on relatively simple market measurements. K-means, Ward, complete-linkage Agglomerative and Birch Clustering was used to separate the data. The resulting clusters were used as state vectors in a Markov Chain model, and used to predict the market’s state in the short-term horizon. Lastly, we provide an improvement to our one time-step-ahead forecast using historical results and the conditional probability framework.

Valtchev, S.Z., Sadiku, J. and Wu, J. (2019). "Identifying Volatility Regimes in Bitcoin Prices", Manuscript. http://zarkonium.github.io/files/Identifying_Volatility_Regimes_in_Bitcoin_Prices.pdf

Convolutional Autoencoders for Lossy Light Field Compression

Expansion and reduction of a neural network’s width has well known properties in terms of the entropy of the propagating information. When carefully stacked on top of one another, an encoder network and a decoder network produce an autoencoder, often used in compression. Using this architecture, we develop an efficient method of encoding and decoding 4D Light Field data, with a substantial compression factor at a minimal loss in quality. Our best results managed to achieve a compression of 48.6x, with a PSNR of 29.46 dB and a SSIM of 0.8104. Computations of the encoder and decoder can be run in real time, with average computation times of 1.62s and 1.81s respectively, and the entire network occupies a reasonable 584MB by today’s storage standards.

Valtchez, S.Z. and Wu, J. (2019). "Convolutional Autoencoders for Lossy Light Field Compression", IEEE Journal of Selected Topics in Signal Processing, *in revision* http://zarkonium.github.io/files/Convolutional_Autoencoders_for_Light_Field_Compression.pdf