23 June 2025 in Chemical Reviews
Generative artificial intelligence (AI) techniques are advancing rapidly and are becoming increasingly challenging to implement. Researchers, practitioners, and enthusiasts alike now require an understanding of complex concepts far beyond the scope of simple feed-forward neural networks to implement the current state-of-the-art methods for their research interests. In contrast, while data augmentation methods may not perform at the same level, they are easier to understand and implement, and are well demonstrated. For these reasons, this review aims to bridge the knowledge gap between the sciences of chemometrics and generative AI and provide a starting point for new researchers. In the context of spectroscopy, this work collects, categorizes, and describes the most popular preprocessing techniques and the state-of-the-art in generative AI and data augmentation, spanning over 104 peer-reviewed journals and proceedings across 32 publishers and organisations. We provide intuitive explanations of the methods, highlighting their strengths and weaknesses, and we include graphical and practical examples of their applications.
6 March 2020 in Journal of Chemical Information and Modeling
1 August 2021 in Journal of Information Security and Applications
11 October 2023 in Journal of Chemical Information and Modeling
Raman spectra are examples of high dimensional data that can often be limited in the number of samples. This is a primary concern when Deep Learning frameworks are developed for tasks such as chemical species identification, quantification, and diagnostics. Open-source data are difficult to obtain and often sparse; furthermore, the collecting and curating of new spectra require expertise and resources. Deep generative modeling utilizes Deep Learning architectures to approximate high dimensional distributions and aims to generate realistic synthetic data. The evaluation of the data and the performance of the deep models is usually conducted on a per-task basis and provides no indication of an increase to robustness, or generalization, on a wider scale. In this study, we compare the benefits and limitations of a standard statistical approach to data synthesis (weighted blending) with a popular deep generative model, the Variational Autoencoder. Two binary data sets are divided into 3-fold to simulate small, limited samples. Synthetic data distributions are created per fold using the two methods and then augmented into the training of two Deep Learning algorithms, a Convolutional Neural Network and a Fully-Connected Neural Network. The goal of this study is to observe the trends in learning as synthetic data are continually augmented to the training data in increasing batches. To determine the impact of each synthetic method, Principal Component Analysis and the discrete Fréchet distance are implemented to visualize and measure the distance between the source and synthetic distributions along with the Machine Learning metric balanced accuracy for evaluating performance on imbalanced data.
19 August 2022 in Journal of Intelligent Information Systems
The intentional targeting of components in a cloud based application, in order to artificially inflate usage bills, is an issue application owners have faced for many years. This has occurred under many guises, such as: Economic Denial of Sustainability (EDoS), Click Fraud and even secondary effects of Denial of Service (DoS) attacks. With the advent of commercial offerings of serverless computing circa 2015, a variant of the EDoS attack has emerged, termed, Denial-of-Wallet (DoW). We describe our development of a simulation tool as safe means to research these attacks as well as to generate datasets for the training of future mitigation systems to combat DoW. We believe that DoW may become increasingly prevalent as applications further utilise services based on a pay-per-invocation cost model. Given that the damage caused is purely financial, such attacks may not be disclosed as application users are not directly effected. As such, we believe that the development of an attack simulator and specific testing of security measures against this niche attack will be able to provide previously unavailable data and insights for the research community. We have developed a prototype DoW simulator that can emulate multiple months worth of API calls in a matter of hours for ease of training data generation. Our aspiration for the future of this work is to provide a system and starting point for research on this form of attack. We present our work on such a system Denial-of-Wallet Test Simulator (DoWTS) - a system that allows for safe testing of theorised DoW attacks against serverless applications via synthetic data generation. We also expand upon prior research on DoW and provide an analysis on the lack of specific safety measures for DoW.
1 January 2024 in Journal of Cybersecurity
Serverless computing is an ever-growing programming paradigm being adopted by developers all over the world. Its highly scalable, automatic load balancing, and pay for what you use design is a powerful tool that can also greatly reduce operational costs. However, these advantages also leave serverless computing open to a unique threat, Denial-of-Wallet (DoW). It is the intentional targeting of serverless function endpoints with request traffic in order to artificially raise the usage bills for the application owner. A subset of these attacks are leeches. They perform DoW at a rate that could go undetected as it is not a sudden violent influx of requests. We devise a means of detecting such attacks by utilizing a novel approach of representing request traffic as heat maps and training an image classification algorithm to distinguish between normal and malicious traffic behaviour. Our classifier utilizes convolutional neural networks and achieves 97.98% accuracy. We then design a system for the implementation of this model that would allow application owners to monitor their traffic in real time for suspicious behaviour.