Hanqing Zhao

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

In this paper, we propose a Bayesian probabilistic model to describe collective behavior generated by a finite number of agents competing for limited resources. In this model, the strategy for each agent is a binary choice in the Minority Game and it can be modeled by a Binomial distribution with a Beta prior. The strategy of an agent can be learned given a sequence of historical choices by using Bayesian inference. Aggregated micro-level choices constitute the observable time series data in macro-level, therefore, this can be regarded as a machine learning model for time series prediction. To verify the effectiveness of the new model, we conduct a series of experiments on artificial data and real-world stock price data. Experimental results demonstrate the new proposed model has a better performance comparing to a genetic algorithm based decomposition model.

Published in: 30th International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems (IEA/AIE 2017), Lecture Notes in Computer Science (LNCS), Volume 10350, pp 82-91, Springer, 2017.

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

Data clustering is the process of dividing data elements into clusters so that items in the same cluster are as similar as possible, and items in different clusters are as dissimilar as possible. One of the key features for clustering is how to define a sensible similarity measure. Such measures usually handle data in one modality, but unable to cluster data from different modalities. Based on fuzzy set and prototype theory interpretations of label semantics, two (dis) similarity measures are proposed by which we can automatically cluster data and vague concepts represented by logical expressions of linguistic labels. Experimental results on a toy problem and one in image classification demonstrate the effectiveness of new clustering algorithms. Since our new proposed measures can be extended to measuring distance between any two granularities, the new clustering algorithms can also be extended to cluster data instance and imprecise concepts represented by other granularities.

Published in: Annals of Operations Research, Volume 256, Issue 2, pp 393–416, Springer, 2017.

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

A new theory of modeling the uncertainty associated with vague concepts is introduced. We consider the problem of quantifying an agents uncertainty concerning which labels are appropriate to describe a given observation. This can be regarded as a simplified model of natural language communication. Semantic meaning conveyed by high-level knowledge representation is often inherently uncertain. Such uncertainty is referred to semantic uncertainty and dominated by fuzzy modeling. In this framework, from an epistemic point of view, labels are precise and uncertainty comes from the undecidable boundary between labels in agents conceptual space. In this framework the boundary is regarded as a random variable and it can be modeled by a probability distribution. We also propose a functional calculus to measure how appropriate of using a certain label to describe an observation. In this way, a vague concept can be represented by a distribution on the labels. The new theory is verified by applying it to the vague category game.

Published in: 5th International Symposium on Integrated Uncertainty in Knowledge Modeling and Decision Making, Lecture Notes in Computer Science (LNCS), vol. 9978, pp. 64-75, Springer, 2016.

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

Clustering analysis is well-used in data mining to group a set of observations into clusters according to their similarity, thus, the (dis)similarity measure between observations becomes a key feature for clustering analysis. However, classical clustering analysis algorithms cannot deal with observation contains both data and vague concepts by using traditional distance measures. In this paper, we proposed a novel (dis)similarity measure based on a prototype theory interpreted knowledge representation framework named label semantics. The new proposed measure is used to extend classical K-means algorithm for clustering data instances and the vague concepts represented by logical expressions of linguistic labels. The effectiveness of proposed measure is verified by experimental results on an image clustering problem, this measure can also be extended to cluster data and vague concepts represented by other granularities.

Published in: 4th International Symposium on Integrated Uncertainty in Knowledge Modeling and Decision Making,Lecture Notes in Computer Science (LNCS), vol. 9376, pp. 236-246, Springer, 2015.

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

Mandarin, also known as Standard Chinese is the official language of China and Singapore, there are certain differences when mandarin is spoken by people from different homeplaces. The homeplace classification is important in speech recognition and machine translation. In this paper, we proposed a novel model named Bag-of-phonemes (BOP) for homeplace classification of mandarin speakers, which follows the conceptually similar idea of the Bag-of-words (BOW) model in text processing. The low-level Mel-frequency cepstral coefficients (MFCC) speach features of each homeplace are clustered into a set of codewords referred to as phonemes. With this codebook, each speech signal can be represented by a feature vector of distribution on phonemes. Classical classifiers such as support vector machine (SVM) can be applied for classification. This model is tested by RASC863 database, empirical studies show that the new model has a better performance on the RASC863 database comparing to previous works

Published in: 7th Iberian Conference Pattern Recognition and Image Analysis, Lecture Notes in Computer Science (LNCS), vol. 9117, pp. 683-690, Springer, 2015.

  Intelligent Computing and Machine Learning Lab, ASEE, Beihang University

An algorithm for extracting the main melody from multi-part musical scores. This model is referred to as the TuneRank model that has the conceptually similar idea of the PageRank model. If each musical note can be considered like a web page in the Internet, and the dissonance value between two notes is like the quantity of links between two web pages. The TuneRank (rank of becoming main melody) of each note is calculated using Markov transition probability. This model is tested on the ECPK4 database. This note-based model is more effective for processing scores containing main melody in multiple parts. Also, the accuracy does not change with the increase of the number of parts. In general, this model can be used for extracting the single-part main melody of digital musical scores.

Published in: Proceeding of Sixth International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC), vol. 2, pp.176-180, IEEE, 2014.