Interactive Visualization of Correlations in High-Dimensional Streams: Unterschied zwischen den Versionen

Aus IPD-Institutsseminar
Zur Navigation springen Zur Suche springen
Zeile 5: Zeile 5:
 
|betreuer=Edouard Fouché
 
|betreuer=Edouard Fouché
 
|termin=Institutsseminar/2018-10-26
 
|termin=Institutsseminar/2018-10-26
|kurzfassung=One of the main challenges about data mining is to analyse the high-dimensional data streams. In this thesis, we would get a deep understanding of correlation analysis and become familiar with high-dimensional data streams. What´s more, we try to visualize these highly valuable data sets through a webservice and evaluate it.
+
|kurzfassung=Correlation analysis aims at discovering and summarizing the relationship between the attributes of a data set. For example, in financial markets, the price of stocks evolves over time. Via a careful estimation of the relationship between stocks, one can try to predict which stock to buy or sell to maximize the wealth of a portfolio.
 +
The standard tool of correlation analysis is the computation of a correlation matrix. However, in the case of streams with many dimensions, it is difficult to extract actionable insights from the correlation matrix, as the number of pairs of attributes increases quadratically and the coefficients evolve over time in unforeseen ways. Thus, novel visualization methods are required.  
 +
In this thesis, we will investigate how to visualize the evolution of correlation in high-dimensional data streams in an intuitive way. We will, for example, discuss visualization methods based on force-directed graphs. Also, we will develop a web interface to visualize the correlation structure of data streams and evaluate it systematically via user studies.
 
}}
 
}}

Version vom 18. Oktober 2018, 20:05 Uhr

Vortragende(r) Yimin Zhang
Vortragstyp Proposal
Betreuer(in) Edouard Fouché
Termin Fr 26. Oktober 2018
Kurzfassung Correlation analysis aims at discovering and summarizing the relationship between the attributes of a data set. For example, in financial markets, the price of stocks evolves over time. Via a careful estimation of the relationship between stocks, one can try to predict which stock to buy or sell to maximize the wealth of a portfolio.

The standard tool of correlation analysis is the computation of a correlation matrix. However, in the case of streams with many dimensions, it is difficult to extract actionable insights from the correlation matrix, as the number of pairs of attributes increases quadratically and the coefficients evolve over time in unforeseen ways. Thus, novel visualization methods are required. In this thesis, we will investigate how to visualize the evolution of correlation in high-dimensional data streams in an intuitive way. We will, for example, discuss visualization methods based on force-directed graphs. Also, we will develop a web interface to visualize the correlation structure of data streams and evaluate it systematically via user studies.