I teach Pattern Recognition "18-794" in the Fall Semester. I urge interested students to take my class, I spend a good portion of class material explaining how and go to the board and derive from scratch different pattern recognition methods and algorithms so that students can start to think for themselves and understand the fundamentals. All the text book algorithms like historically well known PCA, LDA are derived very simply through setting up specific optimization function goals and solving them. Linear Algebra and Probability and Statistics are a prerequisite for understanding these derivations. Likewise, the birth of many other pattern recognition, and dimensionality reduction/discrimination algorithms are born through setting up new optimization objectives!  The difference from my course to other machine learning is that I also go through some of the Advanced Correlation Filter algorithms, derive their optimization the Fourier Frequency domain where we end up with computationally simple closed form solutions! There are many advantages of working in the Frequency domain, many optimization forms can be simplified. Theory is tested through homeworks and mid-term exam, and final project demonstrations allow students to apply what they learned in class and build real-time systems which they demonstrate to the Department! I will put up videos and photos of great past projects - stay tuned!

I also co-teach 18-551 Digital Communication and Signal Processing Design Capstone with Prof. Tom Sullivan. My contribution to 18-551 was the introduction of Android platform and programming modules in 2012 allowing students to develop their 18-551 Capstone Digital Signal Processing projects on the Android Platform.  Students have developed some really cool projects, below is just one group I suggested they do something different and nudged them to build a Mobot Robot and use the Android Camera and processor for computer vision and navigation to control the robot. Using some very simple computer vision methods, running on the Android Phone they won #1 Place in 2012 MOBOT Races!!! My students are the 2nd team in 17 years to win 1st place and complete the whole Mobot track. Please watch the videos, its really awesome! Particularly how they programmed the robot to do a 'double' loop at the end to choose the right gate! 

18-551 MOBOT "Gaussian Blur" 1st Run              18-551 MOBOT "Gaussian Blur" Final Run through 14 gates

© Carnegie Mellon University Biometrics Center 2015

  • s-twitter
  • s-facebook