Machine learning models are growing, with some large language models reaching a scale of billions of trainable parameters. Training these models has since become one of the most data-hungry and computation-heavy tasks. Efforts to distribute the training task mostly follow a federated approach, where a central server oversees the training process. This approach: 1) raises concerns about data privacy; and 2) creates a single point of failure. Current proposals for a fully decentralized approach often rely on costly broadcasts to disseminate model updates and do not tolerate heterogeneity in the training data, as it makes detecting Byzantine contributions harder. We propose BLADE, a generalized fully decentralized (and asynchronous) Byzantine fault-tolerant machine learning algorithm. BLADE was designed to be configurable and adapt to harsh environments, and significantly reduces the communication overhead compared to the state of the art. We performed a comprehensive empirical evaluation, and results confirm models trained with BLADE can achieve an accuracy comparable to a centralized training instance, even if the data distribution among peers is heterogeneous, and robustly aggregate model updates in the presence of Byzantine attacks, and even against sporadic Byzantine majorities.
Decentralized exchanges are one of the most remarkable revolutionary inventions in cryptocurrency trading. Decentralized exchanges present a way to set prices mathematically without maintaining the order book-based trading system. The order book-based trading system has been a widely-known method to determine prices in the stock markets for over 400 years. Recently, a new type of decentralized exchange has devised a constant function market maker to determine prices mathematically. Among them, the method of constant product market makers is the most widely used one. In addition, several constant function market makers have been proposed. However, there was little discussion about the desirable properties of a constant function market maker regarding price and impermanent loss. In this paper, we discuss the desirable price and its effects on impermanent loss. This study considers two types of prices: the reference price and the actual price. We show that impermanent gain can be achievable under certain conditions and prove it mathematically. Two examples are provided to show that impermanent gain is achievable. One question is whether there is agreement on applying actual price when calculating value gains.
Web applications are widely used to access and manage services, including databases to store and retrieve data. However, these applications have suffered from vulnerability issues in their source code, which represents a serious threat to information protection, such as the exploitation of the SQL injection vulnerability that allows access unduly to data from databases. This paper addresses the web vulnerability problem by using Natural Language Processing (NLP) models to detect vulnerabilities in an Intermediate Language (IL) representation of web application source code. The focus on using an IL is to represent the code in a generic language that can be used to train different NLP models, such as sequential (e.g. Hidden Markov Models) and neural network (e.g. Transformers), with the goal of detecting vulnerabilities in several manners and explaining their existence. A vulnerability detection tool will combine these models to check if the code is absent of vulnerabilities or not, stating where they are posed in the code. Preliminary results show that neural network models are able to interpret IL code as being original code and detect vulnerabilities through it.
Integrating air and rail transportation systems offers a significant opportunity to enhance global mobility and operational efficiency. This study explores the role of Artificial Intelligence (AI) in addressing passenger behaviour and interoperability in air-rail networks. Advanced AI techniques are employed to analyse passenger behaviour patterns and optimize multimodal integration. Findings emphasize the importance of tailored solutions for diverse passenger groups, as universal approaches prove inadequate. Methodologically, this paper employs a systematic literature review to synthesize insights and identify trends, focusing on AI applications in air-rail systems. Ethical and technical challenges, including data integration, algorithmic bias, and privacy concerns, are addressed, highlighting the need for robust governance mechanisms. The study advocates for advancements in joint ticketing, baggage handling, AI-driven personalized services, and global standards for multimodal transportation. The findings underscore the potential of AI to revolutionize air-rail networks, offering actionable insights for stakeholders to improve passenger experience and operational efficiency.
As Large Language Models (LLMs) become increasingly integrated into real-world applications and are made more accessible to the general public, significant concerns regarding their security and the safety of the content they output arise. Recent research demonstrates that these models are vulnerable to adversarial techniques, which can manipulate them into generating harmful or biased outputs. These risks highlight the need for robust evaluation methods to assess the resilience of LLMs against these threats. While some efforts have been done to benchmark LLMs' robustness against adversarial prompt-based attacks, these approaches are often highly context-specific and lack the comprehensive components required for a robust and systematic evaluation. In this work we present a benchmarking framework for systematically evaluating the adversarial robustness of LLMs' built-in security and safety measures against text-based prompt attacks during inference time. The proposed framework is composed of the key components that make up a reliable benchmark, including scenarios, workloads, metrics and attack loads and is designed to ensure its representativeness, usefulness and adaptability to different contexts.
