Huang, Y, Zhu, F, Porter, AL, Zhang, Y, Zhu, D & Guo, Y 2020, 'Exploring Technology Evolution Pathways to Facilitate Technology Management: From a Technology Life Cycle Perspective', IEEE Transactions on Engineering Management.View/Download from: Publisher's site
IEEE Technological innovation is a dynamic process that spans the life cycle of an idea, from scientific research to production. Within this process, there are often a few key innovations that significantly impact a technology's development, and the ability to identify and trace the development of these key innovations comes with a great payoff for researchers and technology managers. In this article, we present a framework for identifying the technology's main evolutionary pathway. What is unique about this framework is that we introduce new indicators that reflect the connectivity and the modularity in the interior citation network to distinguish between the stages of a technology's development. We also show how information about a family of patents can be used to build a comprehensive patent citation network. Finally, we apply integrated approaches of main path analysis (MPA)—namely global MPA and global key-route main analysis—for extracting technological trajectories at different technological stages. We illustrate this approach with dye-sensitized solar cells (DSSCs), a low-cost solar cell belonging to the group of thin-film solar cells, contributing to the remarkable growth in the renewable energy industry. The results show how this approach can trace the main development trajectory of a research field and distinguish key technologies to help decision makers manage the technological stages of their innovation processes more effectively.
Zhu, F, Lu, J, Lin, A & Zhang, G 2020, 'A Pareto-smoothing method for causal inference using generalized Pareto distribution', Neurocomputing, vol. 378, pp. 142-152.View/Download from: Publisher's site
© 2019 Elsevier B.V. Causal inference aims to estimate the treatment effect of an intervention on the target outcome variable and has received great attention across fields ranging from economics and statistics to machine learning. Observational causal inference is challenging because the pre-treatment variables may influence both the treatment and the outcome, resulting in confounding bias. The classic inverse propensity weighting (IPW) estimator is theoretically able to eliminate the confounding bias. However, in observational studies, the propensity scores used in the IPW estimator must be estimated from finite observational data and may be subject to extreme values, leading to the problem of highly variable importance weights, which consequently makes the estimated causal effect unstable or even misleading. In this paper, by reframing the IPW estimator in the importance sampling framework, we propose a Pareto-smoothing method to tackle this problem. The generalized Pareto distribution (GPD) from extreme value theory is used to fit the upper tail of the estimated importance weights and to replace them using the order statistics of the fitted GPD. To validate the performance of the new method, we conducted extensive experiments on simulated and semi-simulated datasets. Compared with two existing methods for importance weight stabilization, i.e., weight truncation and self-normalization, the proposed method generally achieves better performance in settings with a small sample size and high-dimensional covariates. Its application on a real-world heath dataset indicates its utility in estimating causal effects for program evaluation.
Lin, A, Lu, J, Xuan, J, Zhu, F & Zhang, G 2019, 'One-stage deep instrumental variable method for causal inference from observational data', Proceedings - IEEE International Conference on Data Mining, ICDM, pp. 419-428.View/Download from: Publisher's site
© 2019 IEEE. Causal inference from observational data aims to estimate causal effects when controlled experimentation is not feasible, but it faces challenges when unobserved confounders exist. The instrumental variable method resolves this problem by introducing a variable that is correlated with the treatment and affects the outcome only through the treatment. However, existing instrumental variable methods require two stages to separately estimate the conditional treatment distribution and the outcome generating function, which is not sufficiently effective. This paper presents a one-stage approach to jointly estimate the treatment distribution and the outcome generating function through a cleverly designed deep neural network structure. This study is the first to merge the two stages to leverage the outcome to the treatment distribution estimation. Further, the new deep neural network architecture is designed with two strategies (i.e., shared and separate) of learning a confounder representation account for different observational data. Such network architecture can unveil complex relationships between confounders, treatments, and outcomes. Experimental results show that our proposed method outperforms the state-of-the-art methods. It has a wide range of applications, from medical treatment design to policy making, population regulation and beyond.
Zhu, F, Lin, A, Zhang, G & Lu, J 2018, 'Counterfactual Inference with Hidden Confounders Using Implicit Generative Models', AI 2018: Advances in Artificial Intelligence, Australasian Joint Conference on Artificial Intelligence, Springer, Wellington, New Zealand, pp. 519-530.View/Download from: UTS OPUS or Publisher's site
In observational studies, a key problem is to estimate the causal effect of a treatment on some outcome. Counterfactual inference tries to handle it by directly learning the treatment exposure surfaces. One of the biggest challenges in counterfactual inference is the existence of unobserved confounders, which are latent variables that affect both the treatment and outcome variables. Building on recent advances in latent variable modelling and efficient Bayesian inference techniques, deep latent variable models, such as variational auto-encoders (VAEs), have been used to ease the challenge by learning the latent confounders from the observations. However, for the sake of tractability, the posterior of latent variables used in existing methods is assumed to be Gaussian with diagonal covariance matrix. This specification is quite restrictive and even contradictory with the underlying truth, limiting the quality of the resulting generative models and the causal effect estimation. In this paper, we propose to take advantage of implicit generative models to detour this limitation by using black-box inference models. To make inference for the implicit generative model with intractable likelihood, we adopt recent implicit variational inference based on adversary training to obtain a close approximation to the true posterior. Experiments on simulated and real data show the proposed method matches the state-of-art.
Zhu, F, Lin, A, Zhang, G, Lu, J & Zhu, D 2018, 'Pareto-smoothed inverse propensity weighing for causal inference', Data Science and Knowledge Engineering for Sensing Decision Support, World Scientific, Belfast, Northern Ireland, UK, pp. 413-420.View/Download from: UTS OPUS or Publisher's site
Causal inference has received great attention across different fields ranging from economics, statistics, biology, medicine, to machine learning. Observational causal inference is challenging because confounding variables may influence both the treatment and outcome. Propensity score based methods are theoretically able to handle this confounding bias problem. However, in practice, propensity score estimation is subject to extreme values, leading to small effective sample size and making the estimators unstable or even misleading. Two strategies — truncation and normalization — are usually adopted to address this problem. In this paper, we propose a new Pareto-smoothing strategy to tackle this problem. Simulations and a real-world example validate the effectiveness.
Huang, Y, Zhu, F, Guo, Y, Porter, AL, Zhang, Y & Zhu, D 2016, 'Exploring Technology evolution pathways to facilitate Technology management: A study of Dye-sensitized solar cells (DSSCs)', PICMET 2016 - Portland International Conference on Management of Engineering and Technology: Technology Management For Social Innovation, Proceedings, 2016 Portland International Conference on Management of Engineering and Technolog, IEEE, Honolulu, HI, USA, pp. 764-776.View/Download from: UTS OPUS or Publisher's site
© 2016 Portland International Conference on Management of Engineering and Technology, Inc. Market competition drives attention to the prospects of New and Emerging Science & Technologies (NESTs), which are fast changing and, so far, have relatively limited applications. Technology evolution pathways, as a powerful representation of the development of technology, have caught researchers' interest as a tool to trace historical progression, explore knowledge diffusion, and forecast future NESTs trends. Citation analysis approaches are actively applied to structure a large number of patents, map patent distribution, and capture knowledge transfer and change in technologies or industries. This paper (1) introduces the indicator of connectivity and modularity in the interior citation network to identify the technology development stage; (2) takes family patent information into the process of building a comprehensive patent citation network; (3) extracts technological trajectories by applying integrated approaches of main path analyses, namely global main path analysis and global key-route main analysis, among different technological stages. We illustrate this approach with Dye-sensitized solar cells (DSSCs), as an example of a promising NEST, contributing to the remarkable growth in the renewable energy industry. The results show how our method can trace the main development trajectory of a research field and discern the technology focus to help decision-makers facilitate technology management.
Zhu, F, Zhang, G, Lu, J & Zhu, D 2017, 'First-order causal process for causal modelling with instantaneous and cross-temporal relations', Proceedings of the International Joint Conference on Neural Networks, International Joint Conference on Neural Networks, IEEE, Anchorage, USA, pp. 380-387.View/Download from: UTS OPUS or Publisher's site
© 2017 IEEE. Motivated by the real damped simple harmonic oscillator (SHO) system, in this paper, we propose a process interpretation of causality and the first-order causal process (FoCP) model for temporal causal modelling. Compared with existing causal models that are able to model feedbacks, such as the structural equation model (SEM) and the structure vector autoregressive (SVAR) model, the FoCP model entails a novel 2-stage evolution semantic for instantaneous and cross-temporal causal relations existing in many real world dynamic systems. Graphical representations are developed to illustrate the causal structure compactly. Useful properties of the new model are identified and used to develop a conditional independence based algorithm for learning the causal structure from a multivariate time series dataset. Experiments on both simulated and real data validate the feasibility of the method to discover simple while meaningful causal structures of dynamic systems.
Zhu, F, Yang, C & Zhang, G 2016, 'SAO-BASED TOPIC MODELING FOR COMPETITIVE TECHNICAL INTELLIGENCE: A CASE STUDY IN GRAPHENE', Proceedings of the 12th International FLINS Conference (FLINS 2016).View/Download from: Publisher's site
Competitive technical intelligence (CTI) tries to identify key technology, current R&D emphases, and key players for intellectual and policy reasons in academia and industry. Many researches apply Latent Dirichlet Allocation (LDA) to CTI mining based on the assumptions of “bag-of-words”, “bag-of-n-grams” or “bag-of-phrases”, which produce the topics at words/phrases level. However, technological words/phrases are not enough to explore problem & solution patterns hidden in technological documents, which are the most important technology intelligence for solution-oriented CTI mining. In this paper, we propose a Subject-Action-Object (SAO)-based LDA model to identify underlying topics represented by related SAOs and explore the problem & solution patterns embodied in SAO structures. SAO-Based LDA model is built based on the “bag-of-SAO” assumption and perform technology analysis at concept level. The validity and feasibility of the proposed method are tested by a case study in the Graphene technology.
Zhu, F, Wang, X, Zhu, D & Liu, Y 2014, 'USER DEMAND-DRIVEN PATENT TOPIC CLASSIFICATION USING MACHINE LEARNING TECHNIQUES', DECISION MAKING AND SOFT COMPUTING, 11th International Conference on Fuzzy Logic and Intelligent Technologies in Nuclear Science (FLINS), WORLD SCIENTIFIC PUBL CO PTE LTD, Joao Pessoa, BRAZIL, pp. 657-663.