Research Projects

The field of artificial intelligence has seen significant progress in recent years, particularly with regards to the use of pretrained models, which have yielded impressive results across a range of tasks. Tabular data is a widely used format in the financial sector for presenting and organizing information, but it has been relatively neglected in the context of AI research. This project aims to address this gap by developing a novel pretraining methodology for tabular data in the domain of finance, which is particularly challenging due to the complex and diverse table schemas involved. Our research questions will focus on adapting to heterogeneous table structures, establishing a universal pretraining protocol, ensuring generalizability and transferability of learned knowledge, and incorporating incremental columns over time. To tackle these challenges, we will propose a pioneering tabular GPT method that utilizes a module called TabUnit to represent basic table elements and a Transformer encoder to refine the representation. We will also utilize free-form prompts to facilitate pretraining and finetuning. Our methodology will be extensively tested and analyzed under various scenarios to validate its effectiveness.

The concept of a metaverse has been gaining popularity in recent years, and it is no longer just a science fiction idea. With the latest technological advancements, it is now possible to create a virtual world where people can interact and connect with each other. However, the very few existing metaverse platforms are typically designed for specific purposes, like gaming, or require specialized hardware such as VR headsets, making it difficult for users to engage easily. This lack of flexibility inhibits the growth of the metaverse as sustainable user engagement is crucial for its success.This project aims to develop a metaverse with core functionalities that enable engaging user experiences through user-centered design. To showcase the functionalities of the metaverse, a social networking application allowing users at different physical locations to take selfies together will be built. A public launch will also be held to inspire demand for further exciting applications such as business promotion and FinTech services to run on the platform. This project is the first step towards establishing a far-reaching metaverse that advances technologies like blockchain and supports studies on safety, ethical and legal practices in the metaverse.

Construction progress payments are critical to successful project delivery and stakeholders’ financial well-being. However, the construction industry has long struggled with delayed and missed payments. Although increased access to digital progress data offers potential benefits for payment automation, it still frequently relies on accurate valuation of work packages (i.e., deliverables) and trustworthy documentation. Furthermore, a theoretical dilemma exists between paying smaller work packages to enhance the cash flows of payees and larger work packages for economies of scale of payors. For example, configuring smaller work packages increases the granularity of the progress payment and is more precise about how much disbursement should be paid to each completed work. However, it increases the administrative burden and dispute frequency. Paying larger work packages can achieve economies of scale (e.g., less financial cost and accounting frequency), but it increases late payment risks, reduces trust, and discourages productivity. This project, thus, proposes a blockchain multi-modal adaptive work package (BC-MAWP) approach to resolve the dilemma. We first model the multi-modal progress data as adaptive graphs to form MAWP for payment automation. Second, studying Blockchainextends the MAWP model from a local system to stakeholder networks for incentive collaboration. Finally, BC-MAWP also offers blockchain oracles and plugins for mainstream financial management systems so that BC-MAWP can work in a trustworthy, seamless, and non-disruptive way.

One unique feature of non-fungible tokens (NFTs) is that they allow creators to embed royalties via smart contract execution. NFTs can use royalties to incentive talent creators and signal higher underlying value of NFTs. However, critics argue that rational buyers may price in the cost of royalties and reduce the trading activities. Since the royalty setting is controversial among NFT communities, this project provides novel evidence for the economics of creator royalties in NFT market.First, this project aggerates comprehensive on-chain NFT transaction and royalty payment data to examine the impact of creator royalties on the market price of individual NFT items. Second, the project calculates NFT collection characteristics to examine the collection-level impact of creator royalties via price discrimination and trading facilizing. By retrieving social media textual data of NFT ownership posts, this project uses machine learning architectures for ownership sentiment scoring of NFT buyers. The project then tests the relationship between creator royalties and ownership sentiment of NFT buyers.This project contributes to the knowledge base of NFT literature, facilitate decision-making of NFT investors, and enhance understanding and utilization of NFTs as an alternative asset in the investment landscape.

In this project, we will design and develop an incentivized blockchain and federated learning (FL) framework that allows financial institutions with heterogeneous resources to co-train financial machine learning models in a robust and privacy-preserving manner. This blockchain-FL framework considers financial institutions with heterogeneous resources, including computation resources, amount and quality of data, and reliability. Our incentivized blockchain-FL framework will encourage financial institutions to participate in distributed machine learning training and blockchain validation processes. Contribution evaluation and incentive allocation will be conducted to give reliable participating financial institutions with higher rewards. The project will lead to a powerful and robust global model for financial institutions to perform their financial services, such as credit risk assessment and customer portfolio analysis.

Data is the lifeblood of smart cities. However, the collection or sharing of personal data faces significant challenges due to privacy concerns and a lack of incentives. As a result, there is a pressing need to establish an automated data trading market enabling data owners to sell data while maintaining control over potential privacy breaches. By leveraging advancements in financial technology, such as auction and blockchain mechanisms, our objective is to develop privacy-aware auction systems that encourage data owners to sell their data with customized privacy guarantees. Specifically, we aim to design privacy-aware auction mechanisms for both platform-based data trading and peer-to-peer data trading. In platform-based trading scenarios, we will develop incentive-compatible reverse auction mechanisms. For peer-to-peer trading, we plan to utilize blockchain technology for decentralized implementations, thereby guaranteeing secure and transparent trading without needing a trusted platform. Additionally, we will extend these scenarios to include collaborative machine learning by motivating data owners to participate in model training for intelligence extraction.

In this project, our goal is to address the challenges faced by online financial services by developing robust, interpretable, and efficient methodologies and application tools for customer behavior modeling. We aim to empower online financial service providers to better understand the customer behavior dynamics of financial transactions and make sound decisions from noisy and sparse financial data. Our approach will involve leveraging advanced graph analytics techniques to model complex interactions and dependencies between various financial transactions and customer behavior dependencies. We will perform comprehensive experiments to validate the effectiveness and efficiency of our proposed learning solutions in various online financial service scenarios, such as fraud detection, credit scoring, market trend prediction, and risk management. With the development of robust and interpretable graph learning methodologies, we aim to provide decision-makers of online financial services with actionable insights and support decision-making to make better-informed decisions and provide more accurate and efficient financial services to their customers. Our systems will be designed to be user-friendly and easy to implement, ensuring that they can be seamlessly integrated into any online financial services platform. Ultimately, our project aims to create a more transparent and reliable financial ecosystem that benefits both financial service providers and their customers.

NFT art is a form of digital art that is authenticated and traded on a blockchain network. The recent success of diffusion models for image generation has empowered AI for NFT art creation. In recent years, several AI-created NFT artworks have been sold for high prices at auctions and marketplaces. However, the AI-created NFT art still faces many challenges, such as the difficulty of aligning the generated content with the market demand, and the limited control and personalization options for the users. In this project, our main objectives are: (1) to propose an algorithm to efficiently align the diffusion-based generative model with the NFT art pricing market; (2) to introduce a unified framework for fine-grained controllable and personalized NFT art generation based on the backbone of Stable Diffusion; and (3) to extend the proposed algorithms to NFT gif or video creation and NFT 3D character creation. Our project will have significant impacts for both academia and industry, as it will push the boundary of NFT art research and generative AI research, as well as create numerous applications and opportunities for various sectors, such as gaming industry, designing products, and entertainment.

The field of artificial intelligence has seen significant progress in recent years, particularly with regards to the use of pretrained models, which have yielded impressive results across a range of tasks. Tabular data is a widely used format in the financial sector for presenting and organizing information, but it has been relatively neglected in the context of AI research. This project aims to address this gap by developing a novel pretraining methodology for tabular data in the domain of finance, which is particularly challenging due to the complex and diverse table schemas involved. Our research questions will focus on adapting to heterogeneous table structures, establishing a universal pretraining protocol, ensuring generalizability and transferability of learned knowledge, and incorporating incremental columns over time. To tackle these challenges, we will propose a pioneering tabular GPT method that utilizes a module called TabUnit to represent basic table elements and a Transformer encoder to refine the representation. We will also utilize free-form prompts to facilitate pretraining and finetuning. Our methodology will be extensively tested and analyzed under various scenarios to validate its effectiveness.

The concept of a metaverse has been gaining popularity in recent years, and it is no longer just a science fiction idea. With the latest technological advancements, it is now possible to create a virtual world where people can interact and connect with each other. However, the very few existing metaverse platforms are typically designed for specific purposes, like gaming, or require specialized hardware such as VR headsets, making it difficult for users to engage easily. This lack of flexibility inhibits the growth of the metaverse as sustainable user engagement is crucial for its success.This project aims to develop a metaverse with core functionalities that enable engaging user experiences through user-centered design. To showcase the functionalities of the metaverse, a social networking application allowing users at different physical locations to take selfies together will be built. A public launch will also be held to inspire demand for further exciting applications such as business promotion and FinTech services to run on the platform. This project is the first step towards establishing a far-reaching metaverse that advances technologies like blockchain and supports studies on safety, ethical and legal practices in the metaverse.

Construction progress payments are critical to successful project delivery and stakeholders’ financial well-being. However, the construction industry has long struggled with delayed and missed payments. Although increased access to digital progress data offers potential benefits for payment automation, it still frequently relies on accurate valuation of work packages (i.e., deliverables) and trustworthy documentation. Furthermore, a theoretical dilemma exists between paying smaller work packages to enhance the cash flows of payees and larger work packages for economies of scale of payors. For example, configuring smaller work packages increases the granularity of the progress payment and is more precise about how much disbursement should be paid to each completed work. However, it increases the administrative burden and dispute frequency. Paying larger work packages can achieve economies of scale (e.g., less financial cost and accounting frequency), but it increases late payment risks, reduces trust, and discourages productivity. This project, thus, proposes a blockchain multi-modal adaptive work package (BC-MAWP) approach to resolve the dilemma. We first model the multi-modal progress data as adaptive graphs to form MAWP for payment automation. Second, studying Blockchainextends the MAWP model from a local system to stakeholder networks for incentive collaboration. Finally, BC-MAWP also offers blockchain oracles and plugins for mainstream financial management systems so that BC-MAWP can work in a trustworthy, seamless, and non-disruptive way.

One unique feature of non-fungible tokens (NFTs) is that they allow creators to embed royalties via smart contract execution. NFTs can use royalties to incentive talent creators and signal higher underlying value of NFTs. However, critics argue that rational buyers may price in the cost of royalties and reduce the trading activities. Since the royalty setting is controversial among NFT communities, this project provides novel evidence for the economics of creator royalties in NFT market.First, this project aggerates comprehensive on-chain NFT transaction and royalty payment data to examine the impact of creator royalties on the market price of individual NFT items. Second, the project calculates NFT collection characteristics to examine the collection-level impact of creator royalties via price discrimination and trading facilizing. By retrieving social media textual data of NFT ownership posts, this project uses machine learning architectures for ownership sentiment scoring of NFT buyers. The project then tests the relationship between creator royalties and ownership sentiment of NFT buyers.This project contributes to the knowledge base of NFT literature, facilitate decision-making of NFT investors, and enhance understanding and utilization of NFTs as an alternative asset in the investment landscape.

In this project, we will design and develop an incentivized blockchain and federated learning (FL) framework that allows financial institutions with heterogeneous resources to co-train financial machine learning models in a robust and privacy-preserving manner. This blockchain-FL framework considers financial institutions with heterogeneous resources, including computation resources, amount and quality of data, and reliability. Our incentivized blockchain-FL framework will encourage financial institutions to participate in distributed machine learning training and blockchain validation processes. Contribution evaluation and incentive allocation will be conducted to give reliable participating financial institutions with higher rewards. The project will lead to a powerful and robust global model for financial institutions to perform their financial services, such as credit risk assessment and customer portfolio analysis.

Data is the lifeblood of smart cities. However, the collection or sharing of personal data faces significant challenges due to privacy concerns and a lack of incentives. As a result, there is a pressing need to establish an automated data trading market enabling data owners to sell data while maintaining control over potential privacy breaches. By leveraging advancements in financial technology, such as auction and blockchain mechanisms, our objective is to develop privacy-aware auction systems that encourage data owners to sell their data with customized privacy guarantees. Specifically, we aim to design privacy-aware auction mechanisms for both platform-based data trading and peer-to-peer data trading. In platform-based trading scenarios, we will develop incentive-compatible reverse auction mechanisms. For peer-to-peer trading, we plan to utilize blockchain technology for decentralized implementations, thereby guaranteeing secure and transparent trading without needing a trusted platform. Additionally, we will extend these scenarios to include collaborative machine learning by motivating data owners to participate in model training for intelligence extraction.

In this project, our goal is to address the challenges faced by online financial services by developing robust, interpretable, and efficient methodologies and application tools for customer behavior modeling. We aim to empower online financial service providers to better understand the customer behavior dynamics of financial transactions and make sound decisions from noisy and sparse financial data. Our approach will involve leveraging advanced graph analytics techniques to model complex interactions and dependencies between various financial transactions and customer behavior dependencies. We will perform comprehensive experiments to validate the effectiveness and efficiency of our proposed learning solutions in various online financial service scenarios, such as fraud detection, credit scoring, market trend prediction, and risk management. With the development of robust and interpretable graph learning methodologies, we aim to provide decision-makers of online financial services with actionable insights and support decision-making to make better-informed decisions and provide more accurate and efficient financial services to their customers. Our systems will be designed to be user-friendly and easy to implement, ensuring that they can be seamlessly integrated into any online financial services platform. Ultimately, our project aims to create a more transparent and reliable financial ecosystem that benefits both financial service providers and their customers.

NFT art is a form of digital art that is authenticated and traded on a blockchain network. The recent success of diffusion models for image generation has empowered AI for NFT art creation. In recent years, several AI-created NFT artworks have been sold for high prices at auctions and marketplaces. However, the AI-created NFT art still faces many challenges, such as the difficulty of aligning the generated content with the market demand, and the limited control and personalization options for the users. In this project, our main objectives are: (1) to propose an algorithm to efficiently align the diffusion-based generative model with the NFT art pricing market; (2) to introduce a unified framework for fine-grained controllable and personalized NFT art generation based on the backbone of Stable Diffusion; and (3) to extend the proposed algorithms to NFT gif or video creation and NFT 3D character creation. Our project will have significant impacts for both academia and industry, as it will push the boundary of NFT art research and generative AI research, as well as create numerous applications and opportunities for various sectors, such as gaming industry, designing products, and entertainment.

An NFT (Non-fungible token) is a non-interchangeable unit of data stored on a blockchain, often combined with digital files such as photos, music, films, and games. Since 2017, NFTs have raised public awareness and experienced enormous growth. However, there is limited research on NFT pricing and risk management, and our project fills the gap between research and practice. We aggregate data from different NFT platforms and blockchains and create NFT benchmark indices to quantify investors’ sentiment and different types of risks (liquidity, frauds, bubble crash, etc.) Our research aims to tackle three issues: 1. We construct quantitative metrics for the rapidly-evolving NFT market. 2. This project aggregates NFT transaction data from different data sources and compiles a comprehensive NFT database. 3. We use machine learning to construct NFT-specific factors and estimate asset pricing models for NFTs. The project will generate social impact in the following dimensions: Researchers can also benefit from the aggregated data to conduct NFT studies and cite the benchmark indices. NFT creators can use the data to assess the market condition. Our research enables investors to construct their portfolios efficiently and mitigate investment risk.

Bitcoin is gaining ever increasing popularity. Meanwhile, due to the unique characteristics, e.g., decentralization, and pseudo-anonymity, bitcoin has also attracted the attention of criminals as safe means of settlement. Therefore, governments still treat bitcoin in a cautious attitude for the stark lack of effective regulation technologies and start to realize that it is necessary to monitor and analyze suspicious bitcoin transactions and wallets. Hence, intuitive software tools for bitcoin wallet investigation are important for regulators. Nevertheless, for the best of our knowledge, there are few existing works can well tackle the problem. Therefore, we propose a visual analytical system to offer such help. Note that this proposal is an extension of our previous proposal “A visualization assisted abnormal trading detection system for multi-crypto currencies”. Comparing to the previous proposal, this proposal presents a system that provides a much richer set of functions and intuitive visual interfaces for regulators to effectively visualize and analyze the transactions of a bitcoin wallet, to track bitcoins flow, and to identify wallet correlation. Besides, we design new visual techniques for presenting bitcoin transactions information and introduce the novel connection diagram and bitcoin flow map as new ways of analyzing, tracking, and monitoring the trading activities of bitcoin wallets.

Accommodating the high penetration of renewable energy effectively reduces carbon emissions. However, renewable energy shows large randomness and fluctuations. It thus threatens the stability of power and energy systems where the energy generated and consumed have to be in balance in real-time. Multi-energy systems bring together electric power, heat, and gas systems. Flexibility can be explored by making use of the complementary characteristics of each energy system to mitigate renewable energy fluctuations. To this end, a growing number of distributed energy resources (DERs) with small sizes, such as microgrids and building prosumers, should be adaptively coordinated to provide flexibility with a high requirement of privacy protection. As classical centralized markets fail in this situation, more efforts should be made to establish a peer-to-peer (P2P) market with extended security and flexibility support considerations. Fortunately, financial and technological (FinTech) innovation opens up various new opportunities for modernizing the existing energy systems. This project is committed to developing a novel, trustful, efficient, and incentive-compatible peer-to-peer market for multi-energy prosumers using blockchain technology. How to realize the organic integration of distributed energy market and blockchain technology will be investigated, and a demonstration platform will be developed to verify the proposed methods.

Bitcoin technology is a distributed, peer-to-peer system, and Bitcoin users communicate with each other using the Bitcoin protocol. When spending Bitcoin, the current Bitcoin owner presents his/her public key and digital signature in a Bitcoin transaction to spend those Bitcoin. A Bitcoin mixer service is a cryptocurrency service that allows users to “anonymise” their Bitcoin by eliminating any possible connection between their original deposited Bitcoin and the mixed Bitcoin that they withdraw later from the service. It is not possible to trace the flow of transactions in a mixer as the owner of the deposit address is not the one moving them after the deposit. The anonymous nature of Bitcoin addresses and the existing of Bitcoin mixer make Bitcoin the perfect currency for money laundering. Many criminals therefore use Bitcoin together with the mixing services to help make their “dirty” money into “clean” money. Tracing the flow of Bitcoin through the mixing services is a challenge to the law enforcement. We propose here to do research in finding traces that may left behind in the Bitcoin transactions that can help investigators to determine the flow of Bitcoin into and out of a mixer, with the purpose to determine relationship between incoming and outgoing Bitcoin of a mixer. We plan to research heuristic rules that can help us to determine 1. A group of addresses that may belong to a mixer 2. The Bitcoin incoming addresses and the Bitcoin outgoing addresses belong to the same user or related users 3. Potential money laundering activities with respect to a group of transactions

Convergence trading was made popular by the hedge fund Long-Term Capital Management (LTCM), which benefits from the phenomenon that the price of a portfolio fluctuates around a certain level. Since deviations from this level are temporary, investors can build appropriate trading strategies accordingly. Ideally, convergence trading is market-neutral and investors will always make profits if this statistical arbitrage happens. In practice, however, the expected convergence may not happen, or it may diverge before converging. The near-collapse of LTCM in 1998 demonstrates this serious risk. Moreover, investors prefer a sparse portfolio since sparsity means fewer transaction costs. Many methods have been developed to study convergence trading, including its existence and convergence speed. However, many methods become expensive for large-scale problems. In this project, we develop novel deep learning methods to address these issues. Specifically, we will construct sparse mean-reverting portfolios and study the cointegration of financial time series. We will evaluate the performance of our trading strategies using real financial data in the Hong Kong stock market. Due to its powerful approximation ability, we expect that deep learning methods can efficiently solve these two problems. Thus, the proposed project is promising in generating broader impacts in the financial engineering community.

The edge computing and big-data computing paradigms are pushing more and more applications to deploy in edge datacenters (e.g., AWS Region and Azure Edge) in order to enable ultra low-latency data access for end users. Recently, the famous Gartner company predicted that, in around 2025, about 75% of enterprise data will be created and processed in edge datacenters — outside traditional core data centers (clouds). To conduct real-time actions and global-scale decisions, a real-world edge application (e.g., edge financial trading, smart- city scheduling robotics, and edge supply chain) usually desires stringent client-perceived latency (e.g., 99% tail latency) on processing both OLTP (OnLine Transactional Processing) transactions and real-time OLAP (OnLine Analytical Processing, or big-data processing) analytical queries. Furthermore, since edge devices (mobile phones and hosts in edge datacenters) are often owned or managed by diverse mutually untrusted enterprises, it is crucial to tolerate the reliability issues of both hardware failures and malicious behaviors (e.g., issuing forged transactions) in these devices. Therefore, this proposal aims to enhance performance and tackle both these two reliability issues as a whole. This proposal will pursue three substantial objectives using a bottom-then-up methodology and create FINO (Freshness, Performance Isolation, Non-blocking, and One-round commit), the first high-performance and reliable HTAP system for edge computing. Dr Cui (the PI)’s preliminary results of this proposal have led to: (1) publications in international best-tiered (China Computer Federation, or CCF, A-class journals and conferences) academic venues (including ACM SOSP 2021, IEEE TDSC 2021, and Usenix ATC 2022); (2) an ITF Platform grant award (Dr. Cui is the PI, HK $3.3 million) in May 2022; and (3) commercial releases of Dr Cui’s UTEE (an HKU-invented big-data security system) on Huawei Clouds (see UTEE in https://www.huaweicloud.com/product/tics.html) and Huawei’s official commercialization acknowledgement letter (https://hemingcui.github.io/doc/utee-ack.pdf). This letter confirms that our HKU’s UTEE big-data security system is already usable by Huawei Clouds’ 3 million users in 170 countries.

Project Abstract: The proposed research aims to accelerate high-frequency trading with new in-memory computing hardware based on emerging memristors. High-frequency trading is a set of strategies to trade securities in the financial markets based on quantitative models that run on a computer. Usually, the one who can execute the orders faster can make more profit than the others. The requirement has limited the available model complexity that can be used for high-frequency trading. Therefore, financial firms are actively building more powerful computing hardware, to expedite the speed of processing market information, but the conventional hardware is ill-suited to perform data-intensive tasks. Memristor-based compute-in-memory hardware has been proposed to accelerate various data-intensive computing workloads, owing to its massive parallelism and co-located computing and memory units. With the hardware, the latency to process the market information and execute the order is independent of model size, promising at least three orders of magnitude improvement compared to conventional hardware. The research in our group has suggested the hardware can predict time-series data with minimum latency. In this research, we plan to apply the technology to process market information. We expect the hardware enables more capable models for high-frequency trading, leading to more profit.

We propose a sentiment-based view on examining investment returns of cryptocurrencies. Cryptocurrency sentiment refers to the emotions and opinions expressed in cryptocurrency- related texts. We investigate whether sentiment-based strategies, especially from social media, generate economically sizable and statistically significant returns, which are not explained by the emerging factor models in the cryptocurrency market. We then explore whether exogenous sentiment shocks affect user growth and if such effect, in turn, contrasts or reinforces the user base feedback effect that underpins blockchain network theories. Blockchain network theories broadly suggest that user network growth positively reinforces itself as existing users are motivated to promote the network and benefit from its growth. We investigate the role of sentiment in the flywheel of user growth and examine if sentiment affects cryptocurrency returns via the channel of user growth. To improve the measurement of the sentiment premium in the cryptocurrency market, we employ three common machine learning methods, which allow us to capture the complicated dependence between different sentiments, which sheds more light on our understanding of how various sentiments affect the cryptocurrency market. Overall, our project aims to provide the first evidence of the nexus between cryptocurrency markets, machine learning methods, and behavioral finance insights.

Money laundering is the process of concealing the origin of money, often obtained from crimes such as drug dealing and human trafficking, by converting it into a legitimate source. Money laundering results in around 2-5% of global GDP (1.7-4 trillion euros) being laundered annually (Lannoo and Parlour 2021). Anti-money laundering (AML) refers to the legal controls that require financial institutions and other regulated entities to prevent, detect, and report money laundering activities. The AML systems deployed by financial institutions typically comprise rules aligned with regulatory frameworks. Human investigators review the alerts and report suspicious cases. Such systems suffer from high false-positive rates, undermining their effectiveness and resulting in high operational costs. In this project, we investigate graph neural networks (GNNs) for AML. GNNs have shown great promise in learning from relational data for a wide range of predictive tasks. GNN as a universal approximator can reduce the burden of manually designing rules. The goal of this project is to design new GNNs models that are not only accurate in predicting money laundering, but also self-explainable via graph intervention and counterfactual inference.

Ethereum has established itself as the most actively used blockchain network by providing integrity, resilience, capability, transparency, etc. These intrinsic features are specified by the official consensus protocol, which is implemented into programs called Ethereum clients. Nevertheless, as Ethereum is new and complex, not all developers have a systematic and comprehensive understanding of these properties, which widens the gap between the specifications and the implementations. In fact, there is an emerging trend for attackers to exploit implementation vulnerabilities in Ethereum clients to launch attacks with damaging consequences. This project aims to narrow the gap between Ethereum’s specifications and implementations by building model checking systems to verify the consistency between different implementations and the correctness of the desired properties. Specifically, we will first build a system to map the code of different Ethereum clients that implements the same functionality. Then, we will build a system that uses bounded model checking to check the consistency of the mapped code and generate counterexamples for any inconsistencies. Furthermore, we will manually check the specifications to define the desired properties in a formal language and build a system to check the correctness of the properties for each Ethereum implementation.

Personal genome promised to make medicine more precise and efficient but can only be realized by aggregating and analyzing people’s health and genomic data at scale. However, losing data sovereignty has become the primary reason hindering more personal genomes from being generated and studied. Moreover, new privacy laws such as GDPR and CCPA now cover personal genome and require it to be protected with “privacy by technology design”. Blockchain technology is a promising candidate for the problem, but a few hurdles remain to be understood and overcome. In this project, we study on-chain storage of personal genome. At the intersection of privacy laws, genomics data, and blockchain technologies, we will study why on-chain storage is a must. Limited by the scarcity of on- chain space, we will study the balance between what personal genomics data to be stored and what genomics operations can be achieved. We will study the options of level-2 blockchain solutions that rollup to public chains but provide more cost-efficient storage. Eventually, we will provide a solution to facilitate genome-wide association study in clinical trial using on-chain personal genomes.

A private set intersection (PSI) protocol is a protocol to get the intersection of two sets, each of which belongs to one party, without disclosing extra information of each party’s set to the other party. For example, after installed an e-banking app, a user wants to check which friends in his contact list have also installed the same app (so that he can make payment later). The user does not want to reveal his entire contact list due to privacy concern, and the bank cannot reveal its customer list too. PSI is a suitable solution for this application. A number of PSI protocols have been proposed in the literature. We focus on the special case of unbalanced PSI: one party has a much larger set (the bank) than the other party (the user). However, there are not much implementation available and hence it is difficult to compare their performance. In this project, we propose new optimization techniques by using new cryptographic protocols as well as new form of data structure. We provide efficient implementation for unbalanced PSI protocol and compare it with existing schemes.

The project aims to design and develop a secure and privacy-preserving blockchain-empowered federated learning framework for FinTech applications. The proposed Federated Learning (FL) framework preserves data privacy by enabling collaborative machine learning for distributed financial organizations without disclosing their private datasets. The FL framework will be fully integrated with blockchain technology to ensure data integrity and secure data aggregation in model training. The local training updates from different organizations will be aggregated to build global models supporting Fintech applications, such as credit evaluation and fraud prediction. We will implement and evaluate the performance of our blockchain-empowered federated learning framework and demonstrate a Fintech application for credit approval.

The development of information technology has reformed many industries. Social media, as a representative of the Web 2.0 era, reshaped the way people obtain financial information with online investment communities (OICs), such as Seeking Alpha and StockTwits. Existing research has suggested that the contents in OICs own prediction value for stock returns. However, since the content volume is huge and the prediction qualities are varying, investors will suffer losses if they are misguided by valueless content. Over the long term, valueless content poses threats to the development of platforms and the financial market. To date, limited research has focused on valuable content screening in OICs. To fill this research gap, we will leverage SeekingAlpha as our research context, where contributors share stock opinions and recommendations and readers consume these contents. Based on signaling theory, we propose that contributors’ disclosed information will serve as a quality signal that helps investors screen content with high prediction value. Specifically, we will examine the moderating effects of contributors’ disclosed information on the relationships between stock recommendations and short-/long-term cumulative abnormal returns. Moreover, the outcomes of investment strategies based on our findings will be illustrated to show their effectiveness and practical relevance.

Artificial intelligence (AI) and other emerging financial technology (“FinTech”) applications such as blockchain and the decentralized autonomous organizations (DAOs) associated with decentralized finance (“DeFi”) protocols are expected to profoundly change the way commerce is conducted within society. While great changes can represent a major windfall to society, they are not without risks. Our goal is to produce empirical research that helps answering critical questions for both policymakers and technologists about the effect of these emerging technologies on the workforce. For example, policymakers fear artificial intelligence (AI) will disrupt labor markets, especially for high-skilled workers. We want to investigate this concern using novel, task-specific data for workers. Our goal is to understand if AI improves product quality, how employees reallocate their efforts, what new tasks are created as AI is adopted, and what the labor market dynamics will look like as talent is reassigned to new opportunities and career path experimentation occurs. We plan to hire research assistants and associates to help with the data collection and statistical analyses that we plan to conduct for this research project.

The predictability of stock market has been studied for a long time in financial economics. To some degree, stock market is predictable based on its historical behavior and other market information. Studies have shown that information from social media can be used to predict stock price movement to some extent. In this project, we propose a novel deep learning model that aims to improve of prediction of stock price movement using social media data compared with existing models. Our proposed model has three main components, namely (1) bidirectional long short-term memory network, (2) attention mechanism, and (3) sample weight adjustment based on social media engagement data. To our best knowledge, we are the first to propose using social media engagement data to adjust sample weights in deep learning models. We will evaluate the performance of the proposed model using real world social media and financial data in the U.S. stock market. We will also propose and evaluate trading strategies based on our model. The results of our findings and applications will be useful for both researchers and practitioners.

Nowadays policymakers are promoting lending to small and medium enterprises (SMEs) given their important roles in creating jobs and driving economic growth. However, the limited of information availability to creditors has largely constrained SME credit access. By incorporating high-frequency, high-dimensional, and high-volume big data and by utilizing the cutting-edge dynamic machine learning models into the business credit risk assessment, the information asymmetry problem can be substantially alleviated. The improvement in SME credit risk assessment is valuable to financial institutions, to SMEs, as well as to policy makers. For the financial institutions that originate SME loans, being able to better assess the credit risk can help them lower the risk exposure and reduce potential losses. For SMEs, a better credit risk management by lenders can increase credit supply and alleviate their credit constraints. For policy makers, the improvement in credit risk assessment can help them better monitor risk in the financial system.

The cryptocurrency Bitcoin was created in 2008 and has received lots of attention recently. The current price of 1 Bitcoin is over US$ 35,000. Bitcoin is based on strong cryptography, which enable users to trade Bitcoin anonymously. The anonymous nature of Bitcoin makes it particularly suitable for crime, such as ransom for ransomware, money laundering by organized crime, etc. Recent development in deterministic wallet allows Bitcoin to be traded with new and unused Bitcoin addresses for every new transaction, which make it almost impossible to identify owners of Bitcoin addresses. Many Bitcoin address clustering algorithms were proposed in the past that have been successfully grouped “unrelated” Bitcoin addresses together and linked to suspected criminals. On the other hand, none of these clustering algorithms have been successfully admitted at the Court of Law because all of them are heuristic in nature. According to Daubert’s principle, to have an “algorithm” be admissible, it needs to have a known error rate. No clustering algorithm was able to report an error rate. We propose here to build a computational model which can be used to validate the validity of the clustering algorithms and to measure the corresponding error rates. With these results, the clustering algorithms will have a higher chance be admitted at the Court of Law. The computational model will be a Bitcoin transaction simulator with built-in traces. The simulator will simulate the real-world Bitcoin transactions and collect statistics at the same time. As it is a simulator, we can always identify the actual owners behind the Bitcoin addresses, and therefore able to measure the error rate of the Bitcoin clustering algorithms.

Environmental, Social and Governance (ESG) index is designed to measure corporations’ ESG performance, and help investors to navigate around ESG risks. Although various ESG indices have been developed by many professional agencies, the ESG data authentication, consistency, and transparency have been largely unexplored. This research proposes a crypto-token incentive mechanism to construct ESG index. Specifically, we begin with designing the fact-, consistency-, and transparency-token incentive mechanism in ESG report preparation, generation, and publication stages. Game theory based operational rules are created to assign the tokens. Then we devise smart contract to execute these token assignment rules in the consortium blockchain for building ESG index. Finally, the fact-, consistency-, and transparency-tokens are aggregated to construct ESG index, by means of developing the sophisticated aggregation schemes. A case study will be conducted using the ESG practices of Hong Kong apparel industry to demonstrate the effectiveness of this research.

Cryptocurrencies have created a new market with numerous opportunities. With a total market capitalization of over 1.7T USD, the security and privacy of cryptocurrencies attracted numerous attention lately. The goal of this project is to investigate new techniques to enhance the security and privacy of cryptocurrencies. To be more specific, we will investigate techniques to properly protect private keys which ultimately control the ownership of the currency units. Our idea is to utilize the latest development from threshold cryptography which allows the sensitive key to be generated and stored in a distributed manner so that the number of compromised server is below a certain threshold, the key remains secure. Furthermore, under no condition do we need to pool these key pieces together so that there will be no single-point-of-failure. We believe this is one of the most promising ways forward to solve the problem of key compromise, and is a robust way to offer full wallet security. In addition, we will consider not just traditional attackers but also attackers equipped with the all-powerful quantum computers. The core idea is to rely on lattice-based cryptography (LBC), a new class of cryptographic techniques from a mathematical structure known as lattices. LBC can be deployed in traditional computers yet offers security against quantum computers. This will ensure that existing cryptocurrencies can adopt our techniques (without requiring any quantum device) yet they will be future-proof. We plan to apply our techniques to the previous HKU coin project to build the first prototype offering both post-quantum security, privacy and enhanced wallet security.

In the innovation-driven economy, the labor market must respond efficiently and effectively to support our economic life. Every day, new jobs appear, and new skills are added to the scope of existing job profiles. Some skills that were once assumed to be “must-haves” are no longer requested, and some jobs are becoming obsolete. What new skills are required for the labor force? Besides the new skill requirements, spatial changes in the economy also need a geographically redistributed labor force to match. Does a specific economic area, such as Hong Kong, have a suitable labor force composition? In this research proposal, we aim to use large quantities of firms’ online job posting information to generate real-time knowledge of the labor demand of the economy. Such information could have many useful applications. For example, the overall temporal evolution of the labor demand can be used to understand industrial upgrading. An individual firm’s labor demand information can be used to evaluate its investment activities in human capital and predict its future performance. Specifically, we propose to first homogenize the definition of jobs and firms by applying NLP word embedding learning and graph neural network on the big data of job posts. Second, no existing explicit model describes the spatial and temporal evolution of the job market. We will use data-driven dynamics identification tools such as dynamic mode decomposition to extract the model from data for further prediction and decision making.

This study combines detailed data on digital finance (from Alipay) and comprehensive data on crime in China to investigate the relation between FinTech and the changing patterns of crime. In theory, the development of digital finance can bring significant changes to the pattern of crime because, first, the change in payment methods – from cash to e-payments and mobile payments – has significantly increased the costs and reduced the payoff of property crime. Second, digital finance makes credit more accessible and inclusive, providing support for people in urgent financial need and curbing the incentive to commit crime. By analyzing the data, we will test the following hypotheses. Hypothesis #1: The development of digital finance significantly reduces the prevalence of crime, especially crimes such as theft and robbery that involve cash as the target. Hypothesis #2: Digital finance changes the relative proportion of different types of crime. While the ratio of many types of traditional property crimes will decrease, cybercrime will become increasingly prevalent. In the meantime, the number of homicides and other personal crimes will not be influenced by the development of digital finance. Hypothesis #3: Digital finance has a stronger effect on crime in less developed areas, including rural areas and less developed cities. We will also study the implications of our findings and propose policy responses.

With the advent of Internet and social media, fake news has become a significant topic of interest to researchers and practitioners. Fake news can be broadly classified as false stories with no verifiability and have fabricated or misleading content. With a lot of automated decision making (such as algorithmic trading) using Fintech, there are concerns that such articles can impact companies and capital markets adversely. Issue of fake news in business is a severe problem as it can impact firm valuations and destroy shareholder wealth. While there is some work on fake news in general, there is limited methodology to systematically identify and label articles that can be potentially misleading in the world of finance because of difference in characteristics of business language from mainstream conversations. In this project, I aim to address two important issues related to fake news: 1) Based on the advancements in Natural Language Processing (NLP) and Artificial Intelligence (AI), and by combining domain knowledge in financial communications, develop and implement a methodology of detecting fake news articles. 2) The dataset of fake news articles generated as a result of the project would be used to study the prevalence of fake articles and their impact on firm valuations, stock prices, and disclosures.

I show that change in investor sentiment predict Bitcoin return on a daily basis. This predictor is the single most significant predictor of Bitcoin daily return, and generates a $R^2$ of 8\%. Despite the fact that there is large variation in Bitcoin price quotes around the world, this proxy for investor sentiment predicts daily return for the most liquid Bitcoin exchange. A simple long and short strategy based on this signal generates a daily alpha of 90 bps, although the transaction cost is also high enough to offset this alpha. These findings suggest that Bitcoin price react with a delay to information contained in the sentiment of institutional investors.

Blockchains are promising to embrace broad applications, including cryptocurrencies, inter- bank ledgers, supply chains, and medical record services. In a blockchain, participants form a network via computers (nodes) to confirm a strongly consistent, total order of appended blocks, each containing a number of transactions. Existing blockchain systems reside in two categories: public and permissioned. A public blockchain (typically, Bitcoin) allows any computer to join as a node, but it relies on cryptocurrencies to encourage nodes to follow its block appending protocol. This reliance makes public blockchains unsuitable for general distributed applications (e.g., medical blockchains). To decouple from cryptocurrencies, a permissioned blockchain (typically, Hyperledger Fabric, or HLF) selects a subset of joined nodes to run a distributed consensus protocol for appending blocks. However, these consensus nodes often become efficiency or reliability weak spots when more nodes join the blockchain (e.g., HLF evaluated merely 100 nodes). In short, no existing blockchain system meets all the three crucial requirements of efficiency, scalability, and generality. This proposal will pursue two objectives by creating two permissioned blockchain systems:Blockchain-powered In-Datacenter Ledger (BIDL) and Efficient, General, and Scalable Internet permissioned Blockchain (EGES). They both meet all these three requirements; they are respectively designed for mostly synchronous networks (datacenters) and asynchronous networks (Internet).

Over 50 million personal genomes were sequenced in the past decade through direct-to-customer DNA tests, such as 23andMe and AncestryDNA. However, under the current practices, the customers have no way of knowing where their genetic data goes or how it will be used. This lack of transparency has had consequences, from disincentivizing participation in programs that would benefit from sharing health or genetic data, to driving a profound lack of genetic diversity in clinical trials. A new model is needed for personal genetic data governance and exchange. We believe that a blockchain tool, leveraging non-fungible tokens and zero knowledge rollups to the Ethereum mainnet, would enable a degree of transparency, traceability, and profitability to allow individuals to actively control their own genetic data. In this proposal, we will design and implement such a blockchain tool, and test it with 2,800 personal genomes from our Chinese Genome Database. We will study what critical and optional genetic information should be stored and the tradeoff regarding whether it is either on-chain or off-chain. We will implement and benchmark smart contracts for actions such as insert, query, and modify. We will show how individuals could join pharmaceutical research projects spontaneously and anonymously and be rewarded.

Investment selection has been a challenging task for venture capitalists (VCs) due to information asymmetry and two-sided matching between VCs and startup companies. Guided with the proximity principle from social psychology and its applications in management and finance, we found that previous efforts in startup recommendation fall short because they did not take full advantage of relation information that can signal the level of trust, cost in private information exchange, and communication effectiveness between two parties. Equipped with this important preliminary finding, we set out to develop a novel framework called SocioLink for startup recommendation. First, a knowledge graph is constructed to describe social connections, geographic proximity, and industry relatedness between venture capitalists and startups. Then, a graph embedding approach and a meta- path-based approach are employed to model the multi-relational information in the graph. We plan to conduct computational experiments to evaluate the proposed recommendation framework. In addition, we also plan to develop a web-based prototype to provide startup recommendations to a given investor and offer explainable intelligence by illustrating the connectivity patterns between the investor and each startup.

This project combines two fields of research that have gained significant traction over the last decade: textual analysis – using machine learning to process unstructured text from the news and other types of documents – and digital footprints – digital traces of how firms interact with digital services. This project combines these elements , studying firm-specific digital footprints in the form of how firms read internet content such as news. This category of data – intent data – is a new phenomenon in the data analytics space and represents a large untapped resource for financial applications. I have an ongoing collaboration with an industry partner who is the category leader in intent data. Through this collaboration, I have a dataset that measures specific firms reading specific pieces of internet content. On a daily basis, this involves over 1 billion content interactions. I show that these digital footprints are extremely powerful predictors of firm outcomes such as company earnings and ESG performance. The goal of this project is to develop further applications of this data, such as measuring firms that want to raise capital, the composition of firms’ corporate culture, or understanding how investors reading the news affects the formation of asset prices.

This research proposal seeks to apply a human rights-based approach to initiatives at the intersection of financial technology (FinTech), financial inclusion, and sustainable development. Studies at the intersection of these three subjects are nascent, particularly from a legal perspective, and have the potential to advance attainment of the UN Sustainable Development Goals (SDGs) through the design and development of appropriate digital financial ecosystems. However, missing from this design is a consideration of how individuals’ human rights may be impacted by this kind of systemic intervention. As the world grapples with the Covid-19 pandemic and the revelation of stark socioeconomic inequities within and across societies, there are greater calls and moves towards the introduction of human rights responsibilities and obligations for private actors. Achieving public social welfare through the protection of human rights is increasingly seen as a shared responsibility between the public and private sectors. Consequently, we propose the development of a human rights rubric through which FinTech developmental interventions can be planned to ensure responsible innovation.

In a recent paper, we argue financial technology (FinTech) is the key driver for financial inclusion, which in turn underlies sustainable balanced development, as embodied in the UN Sustainable Development Goals (SDGs).  The full potential of FinTech to support the SDGs may be realized with a progressive approach to the development of underlying infrastructure to support digital financial transformation.  Our research suggests that the best way to think about such a strategy is to focus on four primary pillars.  The first pillar requires the building of digital identity, simplified account opening and e-KYC systems, supported by the second pillar of open interoperable electronic payments systems.  The third pillar involves using the infrastructure of the first and second pillars to underpin electronic provision of government services and payments.  The fourth pillar – design of digital financial markets and systems – supports broader access to finance and investment. Implementing the four pillars is a major journey for any economy, but one which has tremendous potential to transform not only finance but economies and societies, through FinTech, financial inclusion and sustainable balanced development.  This project will monitor the implementation and impact of this strategy.

Blockchain-based cryptocurrencies such as Bitcoin provide a way to construct decentralized payment systems without the need of any trusted parties. However, these cryptocurrencies inherent the transparent feature of blockchain and thus lack privacy.  To tackle this problem, considerable efforts have been devoted to the development of privacy-preserving cryptocurrencies.  Examples, including Monero and Zcash, employ advanced cryptographic primitives to provide user privacy. However, strong privacy is a double-edge sword.  These cryptocurrencies could be abused and are often associated with illegal activities such as blackmailing or money laundering. Indeed, the strong privacy guarantee makes it challenging for auditing, as the accountability and anonymity are often viewed as contradicting.  Considering accountability is essential in many real-world, a decentralized cryptocurrency supporting privacy and accountability is desired.  Existing effort provides a balance between privacy and accountability, at the cost of additionally trusted third party, which violates the intrinsic property of decentralization, or only support fairly limited auditing operations. In this project, we plan to develop new mechanisms to support various compliance measures while maintaining privacy.  We aim to construct the first decentralized cryptocurrency capable of protecting the privacy of all participants and simultaneously offering full-fledged accountability.

Financial volatility not only creates uncertainty in the financial market but also impacts all aspects of society.  The research will study how financial market fluctuation affect corporates and households, and how digital finance can help mitigate it.  Utilizing the recent financial market fluctuations due to the outbreak of Covid-19, the study will first study the impact of the financial shock on corporate financing and how it will affect the household consumption decision, especially health care decision.  Previous literature find that uncertainty like financial volatility is related to financial constraints for firms and can affect consumer credit decision (e.g. Carvalho, 2018; Di Maggio et al., 2019).  Therefore, we will explore how financial technology can help firms (especially small business) and household rebuilt their credit lines to mitigates the adverse impact of financial volatility. 

Up to now, there are thousands of cryptocurrencies exist in the market, including bitcoin, Ether. But cryptocurrencies impose a critical threat to existing monetary systems due to its anonymity in nature, for examples, money laundry and making profits by disseminating rumors. In this project, we aim at designing and implementing a system to assist the analysis of detecting abnormal transactions in crypto-currencies. This system has two objectives. The first one is to allow investigators to visualize the problematic transaction clusters in a user-friendly manner. There is no existing free tools that allow a user to identify these problematic transaction clusters easily. The second objective is to allow investigators (e.g. investigators from law-enforcement units and/or commercial financial institutes) to receive warnings from our systems. That is, we hope to automatically identify some abnormal transaction clusters and provide warnings to the investigators. Technically, our system integrates data collection, data analysis, and data visualization to provide stakeholders (including exchanges and investors) with effective information about crypto-currencies so as to help them make rational decisions (in addition to the crime case investigation of law enforcement units).