Gaowei Yan

Multi-modal Deep Extreme Learning Machine for Robotic Grasping Recognition

Learning rich representations efficiently plays an important role in multi-modal recognition task, which is crucial to achieve high generalization performance. To address this problem, in this paper, we propose an effective Multi-Modal Deep Extreme Learning Machine (MM-DELM) structure, while maintaining ELM’s advantages of training efficiency. In this structure, unsupervised hierarchical ELM is conducted for feature extraction for all modalities separately. Then, the shared layer is developed by combining these features from all of modalities. Finally, the Extreme Learning Machine (ELM) is used as supervised feature classifier for final decision. Experimental validation on Cornell grasping dataset illustrates that the proposed multiple modality fusion method achieves better grasp recognition performance.

Weakly paired multimodal fusion using multilayer extreme learning machine

Multimodal data have recently become nearly ubiquitous in the real world. Exploring the multimodal fusion is beneficial to improve the performance of the system. However, it is difficult to ensure that data collected from different sources are full pairing. In this paper, we will focus on the weakly paired case of multimodal data, i.e., each modality is partitioned into multiple groups, only paired information on groups is known instead of full pairing between data samples. A new framework of weakly paired multimodal fusion based on multilayer extreme learning machine (ML-ELM) is proposed in this paper, which will find complex nonlinear transformations of each modality of data such that the resulting representations are highly correlated. In this framework, unsupervised hierarchical ELM performs feature extraction for all modalities separately. Then, the higher-level representations from all modalities perform joint dimension reduction by weakly paired maximum covariance analysis. We evaluate our framework on three challenging cross-modal datasets, and the results have proved the effectiveness of proposed method.

Time series forecasting based on deep extreme learning machine

Multi-layer Artificial Neural Networks (ANN) has caught widespread attention as a new method for time series forecasting due to the ability of approximating any nonlinear function. In this paper, a new local time series prediction model is established with the nearest neighbor domain theory, in which the hybrid Euclidean distance is used as the similarity measurement between two sets of time series. In order to improve the efficiency, prediction performance, as well as the ability of real-time updating of the model, in this paper, the recombination samples of the model is derived by Deep Extreme Learning Machine (DELM). The experiments show that local prediction model gets accurate results in one-step and multi-step forecasting, and the model has good generalization performance through the test on the five data sets selected from Time Series Database Library (TSDL).

Similarity measurement based on cloud models for time series prediction

Time series prediction has been extensively used for decision-making in many areas such as economics, engineering and medicine. And the useful data can be excavated by similarity measure of time series from a mass of historical data for predicting. The collected data from the real world is often uncertain, and the cloud model can be a good solution for the problem of uncertainty. This paper proposes a method based on the similarity degree of cloud model and combines it with back propagation network for prediction. In addition to the sequence itself, the trend of the sequence is used as another index of similarity. The neighbour set of query sequence from the training set is selected by similarity measure. Based on the neighbour set, a back propagation network is trained and used for prediction. Experimental results from the six time series show that the proposed method obtains better prediction accuracies than the comparative methods, which reveal its effectiveness.

Soft sensor modeling of mill level based on convolutional neural network

A soft sensor model based on Convolutional Neural Network (CNN) is proposed for the measurement of fill level in highly complex environment inside ball mill. CNN has achieved success in the field of image and speech recognition due to the use of local filtering and max-pooling, which is applied to frequency domain in our method to acquire high invariance to signal translation, scaling and distortion. A pair of convolution layer and max-pooling layer is added at the lowest end of neural network as a method to extract the high level abstraction from the vibration spectral features of the mill bearing. Then, the learned features are transferred to the Extreme Learning Machine (ELM) to model the mapping between extracted features and mill level. Experimental results show that the proposed CNN-ELM method can get more accurate and efficient measurement.

Soft sensor modeling of mill level based on Deep Belief Network

Accurate measurement of the mill level is a key factor to improve the ball mills productive efficiency, safety and economy. Aiming at solving the critical problem of the mill level soft sensor, feature extraction of the processing parameters, a novel method based on Deep Belief Network (DBN) is proposed. DBN is one of the deep learning methods, which focuses on learning deep hierarchical models of data. In this paper, basic features, namely power spectrum density are obtained from the vibration signal of ball mill by Welchs method firstly. Then DBN is built on the basic features to learn high level deep features. Finally a supervised learning algorithm named back propagation neural network is used to model the relationships between extracted features and mill level. Experimental results indicate that the DBN based method outperforms traditional feature extraction methods.

A novel swarm optimization algorithm based on Social Force model for multimodal functions

Social Force model is a dynamic model used to the simulation of crowd behaviors. The Social Force model explains the formation of self-organization from the dynamic. In this paper, a new Swarm Optimization algorithm based on Social Force model (SFSO) is proposed. The SFSO algorithm is a population based optimization technique which is inspired from the behaviors of pedestrian. In SFSO algorithm, the searching characteristics of the pedestrians, such as target selecting, information exchange, overtaking search and scene understanding, are the special abstraction to the pedestrians movement and psychology. The results on benchmark problems indicated that SFSO is a promising optimization method and an effective approach to solve multimodal numerical optimization problems.