A Cost-Sensitive Deep Belief Network for Imbalanced Classification.

Imbalanced data with a skewed class distribution are common in many real-world applications. Deep Belief Network (DBN) is a machine learning technique that is effective in classification tasks. However, conventional DBN does not work well for imbalanced data classification because it assumes equal costs for each class. To deal with this problem, cost-sensitive approaches assign different misclassification costs for different classes without disrupting the true data sample distributions. However, due to lack of prior knowledge, the misclassification costs are usually unknown and hard to choose in practice. Moreover, it has not been well studied as to how cost-sensitive learning could improve DBN performance on imbalanced data problems. This paper proposes an evolutionary cost-sensitive deep belief network (ECS-DBN) for imbalanced classification. ECS-DBN uses adaptive differential evolution to optimize the misclassification costs based on the training data that presents an effective approach to incorporating the evaluation measure (i.e., G-mean) into the objective function. We first optimize the misclassification costs, and then apply them to DBN. Adaptive differential evolution optimization is implemented as the optimization algorithm that automatically updates its corresponding parameters without the need of prior domain knowledge. The experiments have shown that the proposed approach consistently outperforms the state of the art on both benchmark data sets and real-world data set for fault diagnosis in tool condition monitoring.

In situ monitoring of selective laser melting using plume and spatter signatures by deep belief networks.

Critical quality issues such as high porosity, cracks, and delamination are common in current selective laser melting (SLM) manufactured components. This study provides a flexible and integrated method for in situ process monitoring and melted state recognition during the SLM process, and it is useful for process optimization to decrease part quality issues. The part qualities are captured by images obtained from an off-axis setup with a near-infrared (NIR) camera. Plume and spatter signatures are closely related to the melted states and laser energy density, and they are employed for the SLM process monitoring in an adapted deep belief network (DBN) framework. The melted state recognition with the improved DBN and original NIR images requires little signal preprocessing, less parameter selection and feature extraction, obtaining the classification rate 83.40% for five melted states. Compared to the other methods of neural network (NN) and convolutional neural networks (CNN), the proposed DBN approach is identified to be accurate, convenient, and suitable for the SLM process monitoring and part quality recognition.

Fabrication of three-dimensional porous scaffolds with controlled filament orientation and large pore size via an improved E-jetting technique.

Biodegradable polymeric scaffolds have been widely used in tissue engineering as a platform for cell proliferation and subsequent tissue regeneration. Conventional microextrusion methods for three-dimensional (3D) scaffold fabrication were limited by their low resolution. Electrospinning, a form of electrohydrodynamic (EHD) printing, is an attractive method due to its capability of fabricating high-resolution scaffolds at the nanometer/micrometer scale level. However, the scaffold was composed of randomly orientated filaments which could not guide the cells in a specific direction. Furthermore, the pores of the electrospun scaffold were small, thus preventing cell infiltration. In this study, an alternative EHD jet printing (E-jetting) technique has been developed and employed to fabricate 3D polycaprolactone (PCL) scaffolds with desired filament orientation and pore size. The effect of PCL solution concentration was evaluated. Results showed that solidified filaments were achieved at concentration >70% (w/v). Uniform filaments of diameter 20 µm were produced via the E-jetting technique, and X-ray diffraction and attenuated total reflectance Fourier transform infrared spectroscopic analyses revealed that there was no physicochemical changes toward PCL. Scaffold with a pore size of 450 µm and porosity level of 92%, was achieved. A preliminary in vitro study illustrated that live chondrocytes were attaching on the outer and inner surfaces of collagen-coated E-jetted PCL scaffolds. E-jetted scaffolds increased chondrocytes extracellular matrix secretion, and newly formed matrices from chondrocytes contributed significantly to the mechanical strength of the scaffolds. All these results suggested that E-jetting is an alternative scaffold fabrication technique, which has the capability to construct 3D scaffolds with aligned filaments and large pore sizes for tissue engineering applications.

Glucose-insulin regulation model with subcutaneous insulin injection and evaluation using diabetic inpatients data.

Closed-loop insulin delivery systems often implement glucose measurement and insulin administration in the subcutis. However some existing models for glucose-insulin system ignored the dynamics of subcutaneous glucose and subcutaneously-injected insulin. This paper reports a two-compartment model that includes glucose and insulin dynamics in subcutis, and its evaluation using patient data. Clinical information such as glucose level, insulin dosage, insulin injection time and meals of anonymous diabetes inpatients was collected. Measured glucose level of the diabetic inpatients agrees with that of computer simulation. Due to the lack of glucose-insulin model with subcutaneously-injected insulin for type 2 diabetic patients, our model was compared with existing model for type 1 subjects. The new glucose-insulin model can mimic dynamics of glucose and insulin under the disturbance of insulin injections and meals. Model parameters were estimated using nonlinear least square method and their effect on pathology and physiology of diabetes were analyzed.