2021 IEEE International Conference on Services Computing (SCC), 2021
Innovation in Artificial Intelligence (AI) continues to produce a wealth of techniques, mostly co... more Innovation in Artificial Intelligence (AI) continues to produce a wealth of techniques, mostly coming from the inductive form of AI also known as Machine Learning (ML). The majority of ML algorithms is industry-neutral and business process agnostic. ML innovation is propelled by publicly available research, which gets harvested into Open Source for wide distribution through software and Cloud vendors.Ongoing AI technology work creates an immense source of assets for data-driven modeling, delivered as software libraries. However, the application of these assets for data monetization in finance does not happen with nearly comparable success or speed. The latter challenge is commonly known as the "scalability problem of AI". As new techniques continue to grow vigorously, the investment from large finance institutions to cost-effectively produce applications for a variety of lines-of-business (LoBs) and business processes will increase. The availability of ML capabilities on Public Cloud is a way for enterprises to increase productivity by benefiting from the best AI assets available from providers and startups. But data is constrained in terms of location, access and use in most finance competences by either laws or internal Governance, Risk and Compliance (GRC) rules. Legal limitations include, and go beyond, Privacy Acts, impacting non-retail processes where AI techniques must be explained in layperson language to decision-makers and regulators before field deployment. The latter is not yet achieved satisfactorily. Lastly, a large percentage of AI projects fail, in part due to unsuitable ML modeling for analytics and forecasting problems in finance. The variety and complexity of human behavior present in most finance processes calls for understanding AI at a level of cognitive depth that has no precedent in other industries. It is imperative that AI be approached so that finance competence and functional specificity are embedded a-priori into ML techniques and not as use-case afterthoughts. For acceleration of AI assessments, it is critical that ML techniques available in software implement models readily aligned to finance problems.This paper presents an approach to building an Architecture for Artificial Intelligence (AI) in Finance by focusing on analytics and forecasting for business-to-business operations. This AI Architecture hinges on three axes and their interplay: Design Dimensions, Modeling Building-Blocks and Work-Practice. The goal is to support finance practitioners navigate the plethora of AI options more effectively and accelerate data monetization. While ML techniques in data analytics and forecasting apply to many scenarios, this paper focuses on selected competences in Banking, Financial Markets and Chief Finance Officer (CFO) operations.The architecture and method introduced in this paper is a first step toward a service practice. We harvest from our work carried out in banks, asset management firms and CFO lines-of-business as well as R&D experiences in new finance technologies for over one decade. As with any other architecture and deployment methodology, this work requires further harvesting, more information technology tools and sharing experiences across practitioners. It is hoped that finance organizations could adopt these new capabilities in their own Centers of Excellence or other internal organizations leading data-driven transformation and monetization across the firm.
Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, 2018
With the rapid growth of e-tail, the cost to handle returned online orders also increases signifi... more With the rapid growth of e-tail, the cost to handle returned online orders also increases significantly and has become a major challenge in the e-commerce industry. Accurate prediction of product returns allows e-tailers to prevent problematic transactions in advance. However, the limited existing work for modeling customer online shopping behaviors and predicting their return actions fail to integrate the rich information in the product purchase and return history (e.g., return history, purchase-no-return behavior, and customer/product similarity). Furthermore, the large-scale data sets involved in this problem, typically consisting of millions of customers and tens of thousands of products, also render existing methods inefficient and ineffective at predicting the product returns. To address these problems, in this paper, we propose to use a weighted hybrid graph to represent the rich information in the product purchase and return history, in order to predict product returns. The ...
Recent Advances in Stochastic Operations Research II, 2009
ABSTRACT Accelerated life testing (ALT) is a method for obtaining failure time data of test units... more ABSTRACT Accelerated life testing (ALT) is a method for obtaining failure time data of test units quickly under more severe conditions than the normal operating conditions. Typical ALT plans require the determination of stress types, stress levels, allocation of test units to those stress levels, duration of the test and other test parameters. Traditionally, ALT is conducted under constant stresses during the entire test duration. In practice, the constant-stress tests need more test unites and a long time at low stress levels to yield sufficient failure data. However, due to budget and time constraints, there are increasing necessities to design testing plans that can shorten the test duration and reduce the total cost while achieving the equivalent accuracy of reliability estimate. In this chapter, we develop an equivalent step-stress testing plan such that the reliability predictions at normal conditions using the results of this plan will be approximately “equivalent” to the corresponding constant-stress test plan but the test duration is significantly shortened. We determine the optimum parameters of the test plan through a numerical example and evaluate the equivalence of the test plans using simulation. We also investigate the sensitivity of the ALT model parameters.
2021 IEEE International Conference on Services Computing (SCC), 2021
Innovation in Artificial Intelligence (AI) continues to produce a wealth of techniques, mostly co... more Innovation in Artificial Intelligence (AI) continues to produce a wealth of techniques, mostly coming from the inductive form of AI also known as Machine Learning (ML). The majority of ML algorithms is industry-neutral and business process agnostic. ML innovation is propelled by publicly available research, which gets harvested into Open Source for wide distribution through software and Cloud vendors.Ongoing AI technology work creates an immense source of assets for data-driven modeling, delivered as software libraries. However, the application of these assets for data monetization in finance does not happen with nearly comparable success or speed. The latter challenge is commonly known as the "scalability problem of AI". As new techniques continue to grow vigorously, the investment from large finance institutions to cost-effectively produce applications for a variety of lines-of-business (LoBs) and business processes will increase. The availability of ML capabilities on Public Cloud is a way for enterprises to increase productivity by benefiting from the best AI assets available from providers and startups. But data is constrained in terms of location, access and use in most finance competences by either laws or internal Governance, Risk and Compliance (GRC) rules. Legal limitations include, and go beyond, Privacy Acts, impacting non-retail processes where AI techniques must be explained in layperson language to decision-makers and regulators before field deployment. The latter is not yet achieved satisfactorily. Lastly, a large percentage of AI projects fail, in part due to unsuitable ML modeling for analytics and forecasting problems in finance. The variety and complexity of human behavior present in most finance processes calls for understanding AI at a level of cognitive depth that has no precedent in other industries. It is imperative that AI be approached so that finance competence and functional specificity are embedded a-priori into ML techniques and not as use-case afterthoughts. For acceleration of AI assessments, it is critical that ML techniques available in software implement models readily aligned to finance problems.This paper presents an approach to building an Architecture for Artificial Intelligence (AI) in Finance by focusing on analytics and forecasting for business-to-business operations. This AI Architecture hinges on three axes and their interplay: Design Dimensions, Modeling Building-Blocks and Work-Practice. The goal is to support finance practitioners navigate the plethora of AI options more effectively and accelerate data monetization. While ML techniques in data analytics and forecasting apply to many scenarios, this paper focuses on selected competences in Banking, Financial Markets and Chief Finance Officer (CFO) operations.The architecture and method introduced in this paper is a first step toward a service practice. We harvest from our work carried out in banks, asset management firms and CFO lines-of-business as well as R&D experiences in new finance technologies for over one decade. As with any other architecture and deployment methodology, this work requires further harvesting, more information technology tools and sharing experiences across practitioners. It is hoped that finance organizations could adopt these new capabilities in their own Centers of Excellence or other internal organizations leading data-driven transformation and monetization across the firm.
Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, 2018
With the rapid growth of e-tail, the cost to handle returned online orders also increases signifi... more With the rapid growth of e-tail, the cost to handle returned online orders also increases significantly and has become a major challenge in the e-commerce industry. Accurate prediction of product returns allows e-tailers to prevent problematic transactions in advance. However, the limited existing work for modeling customer online shopping behaviors and predicting their return actions fail to integrate the rich information in the product purchase and return history (e.g., return history, purchase-no-return behavior, and customer/product similarity). Furthermore, the large-scale data sets involved in this problem, typically consisting of millions of customers and tens of thousands of products, also render existing methods inefficient and ineffective at predicting the product returns. To address these problems, in this paper, we propose to use a weighted hybrid graph to represent the rich information in the product purchase and return history, in order to predict product returns. The ...
Recent Advances in Stochastic Operations Research II, 2009
ABSTRACT Accelerated life testing (ALT) is a method for obtaining failure time data of test units... more ABSTRACT Accelerated life testing (ALT) is a method for obtaining failure time data of test units quickly under more severe conditions than the normal operating conditions. Typical ALT plans require the determination of stress types, stress levels, allocation of test units to those stress levels, duration of the test and other test parameters. Traditionally, ALT is conducted under constant stresses during the entire test duration. In practice, the constant-stress tests need more test unites and a long time at low stress levels to yield sufficient failure data. However, due to budget and time constraints, there are increasing necessities to design testing plans that can shorten the test duration and reduce the total cost while achieving the equivalent accuracy of reliability estimate. In this chapter, we develop an equivalent step-stress testing plan such that the reliability predictions at normal conditions using the results of this plan will be approximately “equivalent” to the corresponding constant-stress test plan but the test duration is significantly shortened. We determine the optimum parameters of the test plan through a numerical example and evaluate the equivalence of the test plans using simulation. We also investigate the sensitivity of the ALT model parameters.
Uploads
Papers by Yada Zhu