240 likes | 344 Views
Using Interactive Evolution for Exploratory Data Analysis. Tomáš Řehořek Czech Technical University in Prague. CIG Research Group. Czech Technical University in Prague Faculty of Electrical Engineering (FEL) Faculty of Information Technology (FIT). CIG Research Group. Data Mining
E N D
Using Interactive Evolution for Exploratory Data Analysis Tomáš Řehořek Czech Technical University in Prague
CIG Research Group • Czech Technical University in Prague • Faculty of Electrical Engineering (FEL) • Faculty of Information Technology (FIT)
CIG Research Group • Data Mining • Algorithms, Visualization, Automation • Biologically inspired algorithms • Evolutionary computation • Artificial neural networks • Artificial Intelligence • Machine learning, Optimization
Optimization in Data Mining • Main objective of the CIG research group ArtificialNeural Networks Evolutionarycomputation Data Mining Machinelearning ArtificialIntelligence Optimization
Dimensionality Reduction and Visualization in Data Mining • Linear projections • Principal Component Analysis (PCA) • Linear Discriminant Analysis (LDA) • Non-linear projections • Multidimensional Scaling (MDS) • Sammon Projection • Kernel PCA
Interactive Evolutionary Computation (IEC) • Evolutionary Computation using human evaluation as the fitness function • Currently used almost exclusivelyfor artistic purposes • Images, Sounds, Animations… • Inspiration: http://picbreeder.org
Interactive Evolution PicBreeder by Jimmy Secretan Kenneth Stanley
Next generation … and so on …
And after 75 generations ... ... you eventually get something interesting
The technology hidden behind x grayscale z z x Neural net draws the image
Neuroevolution x grayscale z By clicking, you increase fitness of nets Next generations inherit fit building patterns
Using Interactive Evolutionin Exploratory Data Analysis • Experiment with evolvingprojections Examples inn-dimensionalspace 2D
Interactive Evolution of Projections Candidateprojections Machine Feedback Feedback Human
Interactive Evolution of Projections Candidateprojections Machine Feedback Feedback Human
Data Projection Experiments • Linear transformation • Evolve coefficient matrix • Do the transformation using formula: … resulting a point in 2D-space
Data Projection Experiments • Sigmoidal transformation • Evolve coefficient matrix • Do the transformation using formula: b a c
Experiments with Wine Dataset PCA SOM
There are many possible goals! „Blue points down“ – 5 generations, sigmoid projection Outlier Detection – 8 generations, linear projection
Conclusion • Interactive Evolution can be used in Exploratory Data Analysis • Our experiments show that complex projections can be easily evolved • In future, we plan to investigate such evolution in fields of Data Mining other than EDA
Thank you for your attention! Tomáš Řehořek tomas.rehorek@fit.cvut.cz Computational Intelligence Group (CIG) Faculty of Information Technology (FIT) Czech Technical University (CTU) in Prague