OceanEx and Researchers Release New AI Framework SHORELINE to Mitigate Losses of Cryptocurrency Exchanges Hacks

Posted in: Press Releases
May 18, 2020 8:15 PM UTC

With the use of AI technology a risk control mathematical model that is more suitable for exchanges can be established. OceanEx’s research team found that this model greatly increased the operating efficiency of the system and minimized the risk of theft.

Michigan – Researchers from Michigan State University and OceanEx, the AI-powered digital asset trading platform, have released a paper detailing SHORELINE, a deep-learning framework that estimates the optimal threshold of hot wallets from historical wallet activities and dynamic trading networks. Its major use case details a process that would mitigate risk from cyber criminals and hackers on cryptocurrency exchanges.

OceanEx, founded by BitOcean Global in 2018, is an artificial intelligence digital asset trading platform built in the VeChainThor ecosystem. Professor Jiayu Zhou, Co-founder and CTO of OceanEx along with Xitong Zhang of Michigan State University and He Zhu of OceanEx Labs have documented his study on the possible efficacy of SHORELINE in terms of cryptocurrency trading platforms in his 2020 paper, “Shoreline: Data-Driven Threshold Estimation of Online Reserves of Cryptocurrency Trading Platforms”. The paper has been officially recognized and published by the Association for the Advancement of Artificial Intelligence (AAAI).

An in-depth look at the issues plaguing exchanges

Currently, cryptocurrency exchanges face large issues with asset safety. Over 4 billion USD have been compromised since 2019 due to hackers penetrating crypto currency exchanges and carefully maintained trading systems. Crypto exchanges manage accounts to accept deposits from and issue withdrawals to its customers. They also build and maintain order books for the customers to trade at specific prices.

However, what differentiates this from traditional financial systems, where fiat transactions are being settled between real persons, is that each blockchain transaction is between two addresses and is considered finalized once the sender signs the transaction by the corresponding private key. Such anonymity has brought significant security risks to crypto exchanges.

If a hacker manages to control a compromised server of the crypto exchange and steals the private key, the hacker can immediately withdraw all the assets in the exchange by signing a transaction to one of his/her addresses using the stolen private key. The private key that is stored in the exchange server and programmatically signs transactions when a customer requests a withdrawal. This is referred to as a hot wallet, and has been the most critically vulnerable part of crypto exchanges to date.

Cryptocurrency exchanges maintain a certain amount of assets in these hot wallets for several reasons, such as operational cost reduction and minimizing the exposure of cold storage private keys. Hot wallets are crucial for high-frequency trading for their operational efficiency. Due to the vulnerability of potential hacks, only a portion of total assets are stored on hot wallets while the rest of the assets are stored into “cold wallets”, digital wallets that do not sign private keys online.

Therefore, being able to accurately control the available assets in the hot wallets becomes the key to minimize and balance theft risk and efficiency of running an exchange. The ideal situation is that risk stays under control, even when the private keys of the hot wallet are leaked. However, determining such an optimal threshold remains a challenging task because of the complicated dynamics inside exchanges. The solution to the issue is SHORELINE, a deep learning-based threshold estimation framework that estimates the optimal threshold of hot wallets from historical wallet activities and dynamic trading networks.

Overview of the SHORELINE framework

For the first time, researchers provide a data-driven framework called SHORELINE to estimate the optimal threshold for a specific exchange, leveraging its historical data using machine learning techniques. There are two major components included as shown in Figure 1 on network embedding and optimal threshold estimation, as shown below each cryptocurrency is embedded into a low-dimensional vector space temporally, based on the sampled sequences from temporal random walks on dynamic trading networks. To estimate the threshold, researchers combine multiple data modalities from exchanges, such as the historical trading observations, withdraw and deposit history and currency embedding features.

SHORELINE at work: a case study on OceanEx

SHORELINE provides a deep-learning data-driven framework which provides an optimal threshold estimate for exchanges to determine how funds should be allocated in hot wallets to balance theft risk and operational efficiency. The framework does this by leveraging historical data using machine learning techniques such as Dynamic Networks Embedding and Optimal Threshold Estimation, as demonstrated above. The historical data includes data modalities such as historical trading observations, withdraw and deposit history and currency embedding features. With SHORELINE, cryptocurrency exchanges can run optimal efficient operations while still mitigating risk from hackers and reducing the exposure of funds for cybercrimes.

It is reported that OceanEx mainly uses three wallets– cold wallet, hot wallet and warm wallet which is used to accept user deposits. Cold wallet withdrawal requires multiple people in the risk control team to operate at the same time, so it is very safe but inefficient, generally only suitable for N + 1 operations. The hot wallet withdrawal is completely automated by the exchange system, so while the risk is higher than the cold wallet, it is very efficient and suitable for a large number of users to withdraw their needs. When a deposit enters into the user’s warm wallet, the system will immediately transfer the deposit to a hot wallet or a cold wallet according to a certain strategy.

With the use of AI technology a risk control mathematical model that is more suitable for exchanges can be established. In the case of OceanEx, their exchange is an organic operation where market activity, the linkage with multiple tokens, macro user activities, and more fine-grained user behavior habits, work together. After evaluation, OceanEx’s research team found that this model compared with the previous simple model, greatly increased the operating efficiency of the system and minimized the risk of theft.

The future for exchanges

Security is not static, but a very dynamic concept. As early as the 19th century, cryptography was theoretically based on encryption and decryption methods, such as Charles Babbage’s multi-character encryption analysis. After World War I and World War II, the encryption algorithm has been continuously innovated by mathematicians and researchers, and has been continuously cracked by researchers.

Nowadays, encryption algorithms have been widely used in various places in the computer and communication fields, including not only cryptocurrency, from encryption protocols at the bottom link layer of the network, to SSL certificates at the secure socket layer, and hash algorithms to store user passwords. Every day, there is new research to enhance these algorithms being tracked by researchers worldwide, trying to crack these algorithms. Therefore, today’s secure encryption algorithm will be less secure tomorrow. Likewise, it is routine work for the OceanEx technical team to discover new threats and use cutting-edge technology to minimize the risks that unknown threats may bring in this dynamic security field.

For any study on cryptocurrencies trading, researchers should consider all other relevant information of the exchange to build a loss function, and use historical data to perform machine learning to get suitable. The combination of the above information and data would yield a clearer view of the problems and opportunities, so as to build an overall more robust infrastructure, and SHORELINE is one significant step towards this goal.

For more information regarding SHORELINE, please write an email to  jiayuz(at)cse.msu.edu

About Professor Jiayu Zhou

Professor Jiayu Zhou, is Co-founder and CTO of OceanEx. He is also an assistant professor at the Department of Computer Science and Engineering, Michigan State University. Before joining MSU, he led the large-scale recommendation system of Samsung Research Corporation (USA), provided targeted advertisements to millions of Samsung Smart TV users, and has extensive experience in developing AI systems and machine learning algorithms. Professor Jiayu Zhou is currently a member of the VeResearch project and leads the research and development of the distributed data automatic release (DDV) framework on the blockchain. He has a broad research interest in large-scale machine learning, artificial intelligence, data mining, and biomedical informatics.

About OceanEx

Launched by BitOcean Global in 2018, OceanEx Digital Asset Trading Platform uses AI technologies to provide a secure and ultra-fluid digital currency trading market. Capable of comprehensive quantitative trading, it offers a rich set of investment tools and products to meet the needs of all types of investors and investment strategies. Members of its team come from Morgan Stanley, BNP Paribas, Deloitte, Samsung Electronics Research Institute, Cisco and other internationally renowned companies. OceanEx is committed to creating more professional digital asset trading services for investors.

This is a paid-for submitted press release. CCN does not endorse, nor is responsible for any material included below and isn’t responsible for any damages or losses connected with any products or services mentioned in the press release. CCN urges readers to conduct their own research with due diligence into the company, product or service mentioned in the press release.

Last modified: May 19, 2020 8:09 PM UTC

Show comments