I work as a research assistant in the Adaptive Integrated Microsystems Laboratory at Michigan State University under the direction of Dr. Shantanu Chakrabartty, a group focused on bringing adaptability and learning techniques to Microsystems. My research is primarily concerned with the development of biologically-inspired vision systems to produce adaptive imaging techniques with high dynamic range, ultra-low power, and increased sensitivity. I have also started working on support vector machine-based classifiers for applications in speaker (voice) verification, but I hope to apply this work to image classification and hardware-based face recognition also.
Adaptive Imager Chip:
More to come!
Some preliminary results from this work have been published at the 2006 IEEE International Symposium on Circuits and Systems.
Support Vector Machines:
A support vector machine is a type of classifier that has been used for applications in pattern recognition. A great introduction to SVMs is available on Wikipedia, which shows the basic formulation as well as an introduction to the highly non-linear kernel "trick."
I am currently working with a two class problem that involves identifying whether the subject being tested is the true speaker or an imposter. This involves applying the incoming voice signal (with features extracted) to the kernel function and training vectors. As you might expect, the N-dimensional feature space is generated by splitting the audio band into N-distinct dimensions and computing the power at each frequency band.
Publications:
- P. Kucher and S. Chakrabartty. "An Adaptive CMOS Imager with Time-Based Compressive Active-Pixel Response." ISCAS 2006, Kos, Greece.