Haitao Liu

Dalian University of Technology | DUT · School of Energy and Power Engineering

The demand of probabilistic time series forecasting has been recently raised in various dynamic system scenarios, for example, system identification and prognostic and health management of machines. To this end, we combine the advances in both deep generative models and state space model (SSM) to come up with a novel, data-driven deep probabilistic...

Bayesian optimization (BO) is well known to be sample efficient for solving black-box problems. However, BO algorithms may get stuck in suboptimal solutions even with plenty of samples. Intrinsically, such a suboptimal problem of BO can attribute to the poor surrogate accuracy of the trained Gaussian process (GP), particularly that in the regions w...

The flow field and the efficiency of the compressor can be improved and increased by the guide vane in the radial inlet chamber. However, the guide vane generates the wake and results in the rotor–stator interaction, which threatens the safety of the impeller. This paper investigated the guide vane with a self-induced slot (SIS) in a radial inlet,...

Many real-world problems, such as airfoil design, involve optimizing a black-box expensive objective function over complex-structured input space (e.g., discrete space or non-Euclidean space). By mapping the complex-structured input space into a latent space of dozens of variables, a two-stage procedure labeled as generative model-based optimizatio...

Multi-task regression attempts to exploit the task similarity in order to achieve knowledge transfer across related tasks for improving the quality of prediction and alleviating the demand of big data. The application of Gaussian process (GP) in this scenario yields the non-parametric yet informative Bayesian multi-task regression paradigm. Multi-t...

Multi-task Gaussian process (MTGP) is a well-known non-parametric Bayesian model for learning correlated tasks effectively by transferring knowledge across tasks. But current MTGP models are usually limited to the multi-task scenario defined in the same input domain, leaving no space for tackling the practical heterogeneous case, i.e., the features...

The annular seal between stator and rotor substantively acts as a bearing that affects the rotordynamic characteristic of the turbomachinery rotor system. The rotor wake turbulence in a canned motor Reactor Coolant Pump (RCP) will lead to inflow pressure distortion at the annular seal entrance, thus further affecting the seal rotordynamic character...

Expected improvement (EI), a function of prediction uncertainty \(\sigma (\mathbf{x})\)and improvement quantity \( {(\xi - {{\hat{y}}}({\mathbf{x}}))}\), has been widely used to guide the Bayesian optimization (BO). However, the EI-based BO can get stuck in sub-optimal solutions even with a large number of samples. The previous studies attribute su...

In dealing with the expensive multiobjective optimization problem, some algorithms convert it into a number of single-objective subproblems for optimization. At each iteration, these algorithms conduct surrogate-assisted optimization on one or multiple subproblems. However, these subproblems may be unnecessary or resolved. Operating on such subprob...

The radial-flow turbine, a key component of the supercritical CO2 ( S-CO2) Brayton cycle, has a significant impact on the cycle efficiency. The inlet volute is an important flow component that introduces working fluid into the centripetal turbine. In-depth research on it will help improve the performance of the turbine and the entire cycle. This ar...

Multi-task regression attempts to exploit the task similarity in order to achieve knowledge transfer across related tasks for performance improvement. The application of Gaussian process (GP) in this scenario yields the non-parametric yet informative Bayesian multi-task regression paradigm. Multi-task GP (MTGP) provides not only the prediction mean...

For a learning task, Gaussian process (GP) is interested in learning the statistical relationship between inputs and outputs, since it offers not only the prediction mean but also the associated variability. The vanilla GP however is hard to learn complicated distribution with the property of, e.g., heteroscedastic noise, multi-modality and non-sta...

The demand of probabilistic time series forecasting has been recently raised in various dynamic system scenarios, for example, system identification and prog-nostic and health management of machines. To this end, we combine the advances in both deep generative models and state space model (SSM) to come up with a novel, data-driven deep probabilisti...

Deep kernel learning (DKL) leverages the connection between the Gaussian process (GP) and neural networks (NNs) to build an end-to-end hybrid model. It combines the capability of NN to learn rich representations under massive data and the nonparametric property of GP to achieve automatic regularization that incorporates a tradeoff between model fit...

The radial inlet is a typical upstream component which is widely used in multiple-stage centrifugal compressors or blowers. Adding the guide vane in the radial inlet could improve the flow field along circumference distribution and increase the efficiency of the compressor as well. However, the guide vane in the radial inlet generates the wake whic...

Gaussian process classification (GPC) provides a flexible and powerful statistical framework describing joint distributions over function space. Conventional GPCs, however, suffer from: 1) poor scalability for big data due to the full kernel matrix and 2) intractable inference due to the non-Gaussian likelihoods. Hence, various scalable GPCs have b...

For a learning task, Gaussian process (GP) is interested in learning the statistical relationship between inputs and outputs, since it offers not only the prediction mean but also the associated variability. The vanilla GP however struggles to learn complicated distribution with the property of, e.g., heteroscedastic noise, multi-modality and non-s...

Deep kernel learning (DKL) leverages the connection between Gaussian process (GP) and neural networks (NN) to build an end-to-end, hybrid model. It combines the capability of NN to learn rich representations under massive data and the non-parametric property of GP to achieve automatic regularization that incorporates a trade-off between model fit a...

Heteroscedastic regression considering the varying noises among observations has many applications in the fields like machine learning and statistics. Here we focus on the heteroscedastic Gaussian process (HGP) regression which integrates the latent function and the noise function together in a unified non-parametric Bayesian framework. Though show...

Gaussian process classification (GPC) provides a flexible and powerful statistical framework describing joint distributions over function space. Conventional GPCs however suffer from (i) poor scalability for big data due to the full kernel matrix, and (ii) intractable inference due to the non-Gaussian likelihoods. Hence, various scalable GPCs have...

Gaussian process classification (GPC) provides a flexible and powerful statistical framework describing joint distributions over function space. Conventional GPCs however suffer from (i) poor scalability for big data due to the full kernel matrix, and (ii) intractable inference due to the non-Gaussian likelihoods. Hence, various scalable GPCs have...

In transfer learning, we aim to improve the predictive modeling of a target output by using the knowledge from some related source outputs. In real-world applications, the data from the target domain is often precious and hard to obtain, while the data from source domains is plentiful. Thus, since the complexity of Gaussian process based multi-task...

In transfer learning, we aim to improve the predictive modeling of a target output by using the knowledge from some related source outputs. In real-world applications, the data from the target domain is often precious and hard to obtain, while the data from source domains is plentiful. Thus, since the complexity of Gaussian process based multi-task...

[Full PDF version at https://arxiv.org/abs/1811.01179] Heteroscedastic regression which considers varying noises across input domain has many applications in fields like machine learning and statistics. Here we focus on the heteroscedastic Gaussian process (HGP) regression which integrates the latent function and the noise together in a unified non...

As a non-parametric Bayesian model which produces informative predictive distribution, Gaussian process (GP) has been widely used in various fields, like regression, classification and optimization. The cubic complexity of standard GP however leads to poor scalability, which poses challenges in the era of big data. Hence, various scalable GPs have...

As a non-parametric Bayesian model which produces informative predictive distribution, Gaussian process (GP) has been widely used in various fields, like regression, classification and optimization. The cubic complexity of standard GP however leads to poor scalability, which poses challenges in the era of big data. Hence, various scalable GPs have...

In order to scale standard Gaussian process (GP) regression to large-scale datasets, aggregation models employ factorized training process and then combine predictions from distributed experts. The state-of-the-art aggregation models, however, either provide inconsistent predictions or require time-consuming aggregation process. We first prove the...

[See the PDF version at https://arxiv.org/abs/1807.01065] The vast quantity of information brought by big data as well as the evolving computer hardware encourages success stories in the machine learning community. In the meanwhile, it poses challenges for the Gaussian process (GP), a well-known non-parametric and interpretable Bayesian model, whi...

In order to scale standard Gaussian process (GP) regression to large-scale datasets, aggregation models employ factorized training process and then combine predictions from distributed experts. The state-of-the-art aggregation models, however, either provide inconsistent predictions or require time-consuming aggregation process. We first prove the...

The metamodel-based high-dimensional model representation (e.g., RBF-HDMR) has recently been proven to be very promising for modeling high dimensional functions. A frequently encountered scenario in practical engineering problems is the need of building accurate models under limited computational budget. In this context, the original RBF-HDMR appro...

Metamodeling is becoming a rather popular means to approximate the expensive simulations in today’s complex engineering design problems since accurate metamodels can bring in a lot of benefits. The metamodel accuracy, however, heavily depends on the locations of the observed points. Adaptive sampling, as its name suggests, places more points in reg...

Multi-output regression problems have extensively arisen in modern engineering community. This article investigates the state-of-the-art multi-output Gaussian processes (MOGPs) that can transfer the knowledge across related outputs in order to improve prediction quality. We classify existing MOGPs into two main categories as (1) symmetric MOGPs tha...

Multi-fidelity modeling (MFM) frameworks, especially the Bayesian MFM, have gained popularity in simulation based modeling, uncertainty quantification and optimization, due to the potential for reducing computational budget. In the view of multi-output modeling, the MFM approximates the high-/low-fidelity outputs simultaneously by considering the o...

As a well-known approximation method, Kriging is widely used in process engineering design and optimization for saving computational budget. The Kriging model for a target function is fitted to a set of sample points, the responses of which are expensive to obtain in practice and the sample distribution of which has a great impact on the model pred...

Recent metamodel-based global optimization algorithms are very promising for box-constrained expensive optimization problems. However, few of them can tackle constrained optimization problems. This article presents an improved constrained optimization algorithm, called eDIRECT-C, for expensive constrained optimization problems. In the eDIRECT-C alg...

Some adaptive sampling approaches have been developed to efficiently and accurately build global metamodels for the deterministic single-response problems. Most complex engineering problems, however, yield multiple responses during one simulation. This article adjusts the framework of the CV-Voronoi adaptive sampling approach for a multi-response s...

The radial basis function-based high dimensional model representation (RBF-HDMR) is very promising as a metamodel for high dimensional costly simulation-based functions. But in the modeling procedure, it requires well-structured regular points sampled on cut lines and planes. In practice, we usually have some existing random points that do not lie...

The radial basis functions (RBF) interpolation model has been extensively used in various engineering fields. All these applications call for accurate RBF models. The RBF predictions are affected by the choice of basis functions, whereas the proper basis function is problem dependent. To avoid the choice of basis functions and improve the predictio...

To help engineers perform engineering design and optimization efficiently, this article systematically compares the performance of typical sampling strategies and metamodeling techniques based on nine engineering responses with different levels of dimensionality and nonlinearity. In the comparative study, four metamodels, three characteristics of s...

Based on Computational fluid dynamics (CFD) simulation, this study investigates the interior flow in a radial inlet, and helps to figure out the flow phenomenon in radial inlet and the aerodynamic load on the downstream impellers. Three different radial inlets, i.e., original model without guide vanes (OGV), with evenly distributed guide vanes (EGV...

With the development of aero-engine technology, the demand for fan blade with light weight and better structural performances is urgent. This paper focuses on the weight optimization of a hollow fan blade through changing the thickness and distribution of the inner triangular reinforcing ribs under the constraints of equivalent stress, displacement...

In engineering practice, most centrifugal compressors use variable inlet guide vanes which can provide pre-whirl and control volume flow rates. As the impeller of a centrifugal compressor passes through the wakes created from the guide vanes, the aerodynamic parameters change significantly. The concept of adding dual slots at the trailing-edge of t...

Computer simulations have been increasingly used to study physical problems in various fields. To relieve computational budgets, the cheap-to-run metamodels, constructed from finite experiment points in the design space using the design of computer experiments (DOE), are employed to replace the costly simulation models. A key issue related to DOE i...

This paper aimed to improve the volute’s efficiency at both design point and off-design conditions by a robust shape optimization of the volute cross section. The objective function was to maximize the minimal value among volute static pressure coefficient at 90 percent, 100 percent and 110 percent of design flow rate respectively. The design param...

The sensitivity information of objective and constraint functions is required in gradient-based optimization algorithms. Usually the more accurate the gradients are estimated, the faster the convergence speed will be. Commonly used gradient estimation methods, e.g., the finite differences, either have low estimate accuracy or require much computing...

This article develops a novel global optimization algorithm using potential Lipschitz constants and response surfaces (PLRS) for computationally expensive black box functions. With the usage of the metamodeling techniques, PLRS proposes a new approximate function \({\hat{F}}\) to describe the lower bounds of the real function \(f\) in a compact way...

In the field of engineering design and optimization, metamodels are widely used to replace expensive simulation models in order to reduce computing costs. To improve the accuracy of metamodels effectively and efficiently, sequential sampling designs have been developed. In this article, a sequential sampling design using the Monte Carlo method and...

This article presents a global optimization algorithm via the extension of the DIviding RECTangles (DIRECT) scheme to handle problems with computationally expensive simulations efficiently. The new optimization strategy improves the regular partition scheme of DIRECT to a flexible irregular partition scheme in order to utilize information from irre...

A sequential sampling algorithm based on Monte Carlo-based space reduction and local boundary search was introduced. This algorithm utilized the information of the current samples to reduce the design space in order to generate new samples with better space-filling and projective properties. The comparative results with existing sequential sampling...

Surrogate models are widely used in simulation-based engineering design and optimization to save the computing cost. The choice of sampling approach has a great impact on the metamodel accuracy. This article presents a robust error-pursuing sequential sampling approach called cross-validation (CV)-Voronoi for global metamodeling. During the samplin...