Contributions in Merlin

One unsolved sub-task of document analysis is mathematical formula detection (MFD). Research by ourselves and others has shown that existing MFD datasets with inline and display formula labels are small and have insufficient labeling quality. There is therefore an urgent need for datasets with better quality labeling for future research in the MFD field, as they have a high impact on the performance of the models trained on them. We present an advanced labeling pipeline and a new dataset called FormulaNet in this paper. At over 45k pages, we believe that FormulaNet is the largest MFD dataset with inline formula labels. Our experiments demonstrate substantially improved labeling quality for inline and display formulae detection over existing datasets. Additionally, we provide a math formula detection baseline for FormulaNet with an mAP of 0.754. Our dataset is intended to help address the MFD task and may enable the development of new applications, such as making mathematical formulae accessible in PDFs for visually impaired screen reader users.

null

We contribute to the literature on statistical robustness of risk measures by computing the index of qualitative robustness for risk measures based on utility functions. This problem is intimately related to finding the natural domain of finiteness and continuity of such risk measures.

The Belt and Road initiative is a road of win-win cooperation to promote joint development, achieve shared prosperity, and a path of peace and friendship to enhance understanding and communications. Since the proposal was put forward, it has been widely supported by the international community and has achieved fruitful results. This paper selects the stock markets of Italy, New Zealand, Austria, Australia, China, South Korea, India, and Vietnam as the research objects and takes the participation, withdrawal, or refusal of the Belt and Road initiative as events, and applies the event study method to study the short-term impact of the Belt and Road initiative on the stock markets of participating countries. The results show that the Chinese and Italian markets in the participating countries have obtained positive abnormal returns, while the negative returns in the Korean market during the event window show a negative attitude towards the Belt and Road. The results of other countries are not significant. Australia and India, which withdrew or refused to participate in the Belt and Road initiative, have achieved significant negative abnormal returns after the event day, which indicates that the market believes that rejecting the Belt and Road initiative is a strong negative signal. In addition, this paper also finds that the impact of the Belt and Road policy on the stock market has noticeable industry effects, and the response of the same industry to the Belt and Road policy is quite different in different markets.

Aspect extraction (AE) and aspect-based sentiment analysis (ABSA) are tasks that aim to identify ”topics” and sentiment polarity of these topics in the text corpus. In industry, aspects in the review texts are the categories/topics talked about in the review text and are usually the features of a product/business. Obtaining feedback/opinion on these features is an essential part of feature improvement and, in turn, customer satisfaction for businesses. Consequently, a mechanism to read and extract information from the plethora of feedback available becomes an essential task. Extracting the sentiment polarities of these categories is done by a more complicated method than sentiment analysis known as aspect-based sentiment analysis (ABSA). Aspect-based sentiment analysis aims at extracting the sentiment of each of the categories/topics from the text. Aspect-based sentiment analysis is a sub-task of sentiment analysis (SA) but is especially challenging since there can be varying numbers of aspects mentioned and each with a different sentiment polarity. Moreover, the availability of labelled industrial data is limited and restricts the use of various supervised machine learning algorithms. High accuracy is also expected in industries to be able to make business decisions from model predictions. In order to solve this issue, in this paper, we aim to extract these aspects and their sentiment polarity using BERTBASE model using a weakly supervised technique and obtain high accuracy such that it can be used by businesses to make data driven business decisions for feature improvements in their product. BERTBASE uses unsupervised masked (Masked Language model) training and next sentence prediction (NSP) for learning. We propose a method to modify the unsupervised masked language model, training of the model to make it more category aware. We then, fine tune the weights of this pre-train the model using the auxiliary sentence pairs to extract aspects and predict its sentiment polarity to get state-of-the-art results.

Cyberattacks are posing a threat to the continuity and success of organisations worldwide and the costs associated with cybersecurity skyrocketed in the last years, reaching USD 6.9 billion in 2021. Therefore, the need for appropriate cybersecurity has never been higher. While malicious parties are teaming up to collaborate and cooperate on attacks, it is only natural that defenders should also cooperate or collaborate with each other in order to detect vulnerabilities or mitigate attacks. Firstly, this work gives an overview of the current state of cybersecurity and important concepts, and frameworks are discussed. Subsequently, definitions for collaborative cybersecurity and cooperative cybersecurity are proposed. A taxonomy is presented that could serve as a good categorisation basis for the existing research. A survey of the literature is performed, leveraging the proposed taxonomy. Finally, key limitations, trends and challenges are identified. As shown, current approaches focus on vulnerabilities sharing, intrusion detection and access control. While challenges remain with regards to the right amount of transparency in cybersecurity and to public-private cybersecurity collaboration and cooperation.

With a major part of the Swiss cheese produced being exported, counterfeit cheeses selling under their protected name abroad pose a serious economic threat to the Swiss cheese industry. Consequently, a group of Swiss private and federal entities have teamed up to build the CheeseChain, a blockchain-based solution to increase transparency and trust along the Tˆete-de-Moine (a Swiss cheese) value chain, as well as provide proof-of-origin using a PCR-based system and publish the results to the public Blockchain (BC). This thesis presents a smart contract (SC) solution to track a cheese through its supply chain including a frontend to facilitate interactions with the SC and a server that allows data from the existing systems to be written to the BC via network requests. The SC abstracts a supply chain into supply chain participants, production batches, and individual production steps that are appended to a batch, which makes the SC applicable to supply chain tracking use cases other than cheese with only minimal modifications of the source code. This is achieved using custom data structures, called structs, for production batches and steps, where every batch contains the timestamp of creation, further information about the product, and a pointer to the last step performed within the supply chain and registered on the batch. Each step itself points to its predecessor and stores a description of the step, the participant who performed the step, the timestamp, and the coordinates alongside. This forms an immutable and append-only backward traceable linked list of steps that can be inspected via the custom frontend designed for manual interaction with the SC by supply chain participants and the retrieval of supply chain history for a specific cheese by the customer allowing for verification of a product’s authenticity. Further, the SC implements an access control mechanism that guards all functions that change the SC’s state. All functions are only callable by registered supply chain participants, which are managed by the system administrator, who is at the same time the deployer of the contract and the only entity that is allowed to call all available functions. The server provides an abstraction over the SC and facilitates injecting data available in a private system to be written into the BC automatically through an API. Evaluating the system has shown that choosing the best suitable BC for a specific application has to be done diligently and carefully since it directly implies the system’s performance and operational costs, which depend on the selected BC’s block time, respectively token and gas price. Further, the presented SC is evaluated to carry a low risk from a SC security standpoint, due to the access control mechanism in place and the absence of monetary incentives to exploit the system. The biggest security risk is introduced by the mismanagement and leakage of private keys by the registered supply chain participants or the administrator, which gives the holder the ability to write faulty information into the SC.

The analysis of Bitcoin blockchain data, especially the transaction data, is interesting for both blockchain and economic researchers, in specific behavioural economic researchers. The blockchain serves as an accounting system that is managed in a decentralised manner and involves many participants. Blockchain makes use of the distributed ledger technology (DLT) but goes further and is characterised by the following five features in particular: (i) Cryptography, (ii) peer-to-peer (“P2P”) network, (iii) Consensus Mechanism, (iv) ledger, and (v) the Validity Rules. In this respect, the blockchain facilitates the transfer of assets and data and does not require a trusted central authority. However, interest in blockchain technology was triggered or increased by the emergence of the cryptocurrency Bitcoin. The cryptocurrency Bitcoin enables the exchange of valuable goods and services without a central authority. Thus, it fulfils the function of cash. With its cryptographic structure, it enables the user to stay anonymous while doing transfers. Furthermore, the blockchain infrastructure offers the possibility of analysing data of economic transactions between users and thus recognising economic patterns. Previous research has focused on the pure financial or technical aspect of Bitcoin, but not on the structural properties of the network of Bitcoin users. Thus, the dynamics of the Bitcoin network remain unconsidered. However, research into the structural properties of the Bitcoin network can draw conclusions about how the behaviour of users can be influenced by other users. In this paper, this possibility is taken into account by examining the consequences of two major events, the bankruptcy of the crypto bank Mt. Gox in 2014 and the ongoing Covid-19 pandemic, respectively its emergency in March 2020. The comparison of these two events should contribute to a better understanding of the behaviour of Bitcoin users and the associated price formation mechanism.

The results for the calculation of the energy consumption of streaming are currently very far apart. The difficulty here is that many assumptions have to be made and the streaming process is very complex. The aim of this work is to determine the relevant parameters for the energy consumption of streaming and also to show ways to effectively reduce energy consumption using these parameters. To achieve this goal, available literature was used to analyze the streaming process to find out where energy is consumed during video streaming. In the second part, it was worked out which parameters have a decisive influence on the level of energy consumption. It turned out that the relevant parameters for the energy consumption of streaming are the utilization and energy efficiency of the servers in the data centers, the file size of the videos, the efficiency of file transfer and the screen size of the end device. Effective ways to reduce energy consumption with these parameters are the purchase of energy-efficient servers, server virtualization, the use of so-called content delivery networks (CDN), the expansion of fiber optic networks and the 5G infrastructure, the compression of video files and the choice of smaller dispaying devices.

This study sought to calculate the environmental impacts of streaming video in terms of energy use and carbon dioxide emissions, based on technical and geographical parameters. The resulting data was used to create an online carbon dioxide calculator, which allows users to estimate the environmental impact of their own video consumption. The calculator can be accessed online at https://streaming-co2-rechner.ch/. To calculate the environmental impacts, the energy consumption of the various components involved in the distribution and playback of the video content was taken into account: the data centers, the communication networks and the end devices. In an engagement with current studies, it was found that, on average, streaming video emits 62 g CO2e per hour. This is equivalent to driving a car for 430 meters. However, this value can not be applied indiscriminately to any streaming scenario, as there is inherent variability in the estimates depending on various factors. The calculator allows users to identify the factors that most significantly affect the results. These include the country-specific emissions factor for the electricity grid, the video resolution, the choice of end device and the choice of network connection. The study also identified a need for further research into the use of appropriate allocation rules in order to calculate the impacts of streaming video in a more precise and consistent way. It was found that the use of energy consumption values per unit of data is not suitable for streaming and other data-intensive services. This leads to significant uncertainty in calculating the energy consumption of network components, as there is currently a lack of quantitative data and appropriate calculation methods need to be further validated and refined.