By Tin Kam Ho, Head of the Statistics and Learning Research Department, Bell Labs.

The world remembers Dennis for his monumental contributions.  Apart from all these though, I remember Dennis also as someone who had profound impact on my own life and career, for he hired me into his department in Bell Labs Research in 1992, thereby starting my two decades of career here.

While I admire C and UNIX and use them on a daily basis, I came to Bell Labs to join something very different — an effort led by Henry Baird, a well respected scientist in our field, on building multi-lingual reading machines.  It was a natural extension of my dissertation research.  At that time this work was hosted in the department led by Dennis.  This brought about the surprising connection that I found myself in with Dennis.

By then Dennis was already world-famous for his great works, and the fundamental importance of programming languages and operating systems was universally appreciated.  On the contrary, the challenges of our pattern recognition research remained known to only those who had chosen to dive in — massive amount of noisy data, representational spaces with hundreds or thousands of dimensions, and many diverse sources of knowledge in need of integration.  But this did not stop Dennis from giving us the support we needed.

Dennis Ritchie (1941-2011)

For the next five years, Dennis encouraged and maintained an active interest in our pursuits even though they went far beyond his own territory. This was no easy task: both Henry and I were going after something rather unconventional.  Henry wanted an automated way to train computers to read as many languages as he could — books and documents printed not only in Latin and Cyrillic alphabets but also in chess symbols, Tibetan, and oriental languages.  Doing this in many other places would require an army of helpers to scan and annotate images and tune symbol classifiers.  But instead Henry sought to understand and model the imaging noise, so as to solve the problem once and for all.  With the image defect model we were able to generate millions and millions of noisy images of any symbol, which allowed us to fully exploit any classifier’s ability and see the limits of recognition, be it by machine or by human.  I was also looking to extend my research on combining classifier decisions,  except that instead of decisions by a few experts, I would be dealing with a massive number of essentially random guesses.  This brought me into creating a counter-intuitive technique that I called random decision forests.  It turned out to be a robust method for statistical learning that has since found usage in all sorts of disciplines.

Dennis never questioned our going after such extremes.  Rather, I often found him amused by our ambitions, and remained curious of how far we could go.  He liked seeing that we shared some interest in multi-lingual support with the Plan 9 research team, but he never made us feel that our work was in any way less important.  I would never know if Dennis had foreseen today’s viral spread of interests in massive-scale data analytics and statistical learning, or the role he had played in nurturing some of the early developments. I am forever grateful that his enthusiasm and generous support allowed me the opportunity to participate in the progress, and go to distant corners where no one had reached.

Watch the livestream of the Bell Labs event to honor the life and work of Dennis Ritchie (September 7, 9.30 AM Eastern time)