After a postdoctoral position at the Massachusetts Institute of Technology, Cambridge, working in theoretical particle physics, he joined AT&T Bell Labs in 1986, working on transmission performance and routing algorithms for the CCS7 signaling network. He then worked on applying neural networks to handwriting recognition, for zip codes and bank checks. While at AT&T Bell Labs, and later at Lucent Technologies, he worked on support vector machines, concentrating on training methods, pattern recognition, computational efficiency, theory。