|Published (Last):||9 April 2005|
|PDF File Size:||17.40 Mb|
|ePub File Size:||16.98 Mb|
|Price:||Free* [*Free Regsitration Required]|
Are you aware of how your work in quantization found applications? Widrow: I published these three papers, and then I did something that would be academically not so smart.
I became interested in, even fascinated by, learning systems, and changed the direction of my academic career. I was pretty young and pretty naive but I knew the significance of what I did. I knew it that day when I came into my office and I drew that Gaussian-like density on the blackboard. I was making more and more discoveries. It was sort of like being an explorer. I knew that no one had ever thought of this before. I just knew it. Goldstein: Well did that make you want to rush out and tell people?
Widrow: Of course. But then something happened. In the summer of I completed my thesis, received the doctoral degree, and joined the faculty at MIT. That was the first time I gave a talk to a major audience. I was scared to death. The biggest group I have talked to was on the subject of neural nets, and that was at this conference in San Diego June 21 to 24, There were two thousand people.
So that was the biggest group I ever talked to. That talk went very well. I was just in a good mood that night, everybody was in a good mood and all enthusiastic about neural networks. When I got done, I knew I had given a good talk, but I was not prepared to see the whole two thousand people get up and give me a standing ovation. Widrow: Some of my colleagues were there. One, a good friend, said, "Never in my life have I ever seen anything like that at a scientific meeting, where people get up and give a standing ovation to a talk.
To answer your question about what happened when I did this. First of all, there was no field of digital signal processing at the time.
There was no name for this stuff. We called it sampled data systems. The number of computers that existed in the whole world at that time you can practically count on the fingers of one hand. Today you would call it sort of a mainframe. I think it was an IBM , and it had a magnetic core memory in it, and it was just like a jewel. It was identical to the core memories that we built, except that all the memory planes were chrome-plated.
We used gray wrinkle finish. They made it really pretty. Artificial intelligence Widrow: I had finished the work on quantization, then joined the faculty during the summer vacation. We were in a U. Navy-sponsored lab developing all kinds of new circuits for computing.
A colleague named Ken Shoulders was working on field effect devices, and his dream was to make integrated circuits. There was no such thing at the time. The first person I ever heard of talking about making active electronic components and the wiring all at the same time was Shoulders. We had been working on this since , years before integrated circuited. He was thinking of a new device. He was a brilliant guy and "crazy Texan.
He was trying to make a microscopic field effect device. He was kind of a visionary, and he told me that some people were doing this stuff called artificial intelligence. He wanted to find out what it was about. The two of us got into his car and went up Hanover, New Hampshire to attend the meeting. We stayed there for a week.
Minsky was there, McCarthy was there. Maybe if we had just taken photographs of the people who were there. It was an incredible collection of people. They were talking about making a brain-like machine. This thing hit me so hard that I never got over it. This has been a monkey on my back. I wanted to do something to build a brain. I wanted to build a machine that thinks.
I stopped working on quantization. The sun rose and set in Cambridge, Massachusetts. It was amazing. There were all kinds of smart people all over the place, thinking of all kinds of things.
So during the summer of , I began work on artificial intelligence. If I had been a real smart guy and had known how to play the academic game, I would have milked that thing for the rest of my life. You know, when you discover the whole theory of something reasonably major, you spend your whole life on it. But I was young and foolish, and I was absolutely captivated with the concept of artificial intelligence.
Publications; adaptive filtering So what approach did you try? Widrow: The new discoveries in quantization theory are amazing. We have discovered new things based on the original bedrock, but now we are applying this to floating point quantization.
One book was on adaptive signal processing, written with Sam Stearns, and the second book with Eugene Walach is called Adaptive Inverse Control, which uses adaptive filtering and signal processing methods to do adaptive control.
We developed an entire theory of adaptive control based on adaptive filtering. Goldstein: Can we talk about how you got involved with adaptive filtering? Widrow: Certainly. He was working for ten years on the theory of quantization before I met him.
I had always intended to go back and write a book on quantization. They do it with great fear and trepidation, and no one, no book has discussed the theory, yet the theory has been out there for years.
Anyone who understands digital signal processing with an hour of instruction from Professor Widrow would understand the theory and would relate to things that they already know. How do you take what you already know to solve a new problem? The problem of quantization noise is one that you deal with but you kind of ignore. Anybody could have written this book for me, but so far nobody has.
Over the years to come many people will rewrite the book for us and add their own tweaks to it. The books that I write always work. I never write things that summarize the field. Kollar has extended the subject, to develop a theory for floating point quantization. We found a way to use the ancient work to apply to the new problems and got very similar and amazing results with floating point. We have developed a nice theory of floating point quantization that anybody can understand.
It just takes a little more effort to do it. So, how did I get into adaptive filtering? Well, I came back to MIT after a conference at Dartmouth College, and I spent about six months thinking about what thinking is about; trying to understand thinking.
I still hold the same theories today as to how thinking works. Did you have your degree? Widrow: No, I already had the degree. I made a transformation from research assistant to assistant professor. The year that I graduated was an interesting year. I can remember a couple of guys that got their doctorates at the same time. One was Amar Bose from Bose Loudspeakers.
I think that ten of us out of the twenty joined the faculty and stayed there for a few years. You become an assistant professor, but it was sort of a professor training program. Everybody else went somewhere else, and I came here. I taught at MIT for three years and then came here. When I came back to MIT after the conference at Dartmouth I began thinking about thinking, and I began thinking about how do you make a machine that can do it.
I came to the conclusion that it would be about twenty years before anything of any consequence in engineering would come from an attempt to build a machine that thinks.
Follow the Author
A treatment of adaptive signal processing featuring frequent use of examples. Widrow writes in a clear and easy-to-follow style which delivers all of the mathematical theory and detail of the process of adaptation without drowning the reader in formalism. Statistical signal processing, adaptation dynamics, steady-state behavior, performance - this book explains all of these fundamentals. If you want to become an expert in adaptive signal processing, start with this book. Get This Book if you want to get This Subject. By Mr.
ADAPTIVE SIGNAL PROCESSING BY BERNARD WIDROW PDF
Adaptive signal processing by bernard widrow pdf
Adaptive signal processing