Interviews > dr.matthew vincent mahoney



A BIG NAME IN TODAY'S DATA COMPRESSION RESEARCH AND TECHNOLOGY.

Dr.Matthew Vincent Mahoney is currently senior scientist at Ocarina Networks, he specializes in data compression. Got his MS (1988) and Ph.D (2003) degree from Florida Tech. He is working on data compression technology since long time. He wrote his first data compression algorithm back in year 2000 which was a neural network compressor. Most remarkable thing you would love to know about Dr.Matt is, he has made core of all his research downloadable on his website in the form of source as well as almost fully explained each and every algorithm he developed. Dr.Matthew's background includes programming in Assembly Language for Intel x86 and MMX as well as network programming in Java programming language. Dr.Matt has an extensive research on Intrusion Detection, his research papers may be found here. In his interview he pointed a compression algorithm called SR2 which he states as fastest among all of his compression methods. Meg.s of C and C++ source code on his site makes it possible for every developer / researcher to get right compressor / decompresser, and yes, it sometimes confuses to choose among that many resources. While reviewing his page dedicated to data compression, we really got confused seeing all those algorithms, and then double checked to confirm we are on the right page. Apart from his usual work of writing data compression algorithms, he loves running and mountain climbing. As far we have seen on his website (that he keeps updating time to time), he has documented every vacation since 1998 upto current year. He has maintained the race calendar and results for the Space Coast Runners in east-central Florida. Dr.Matt has 7 cats in his home. His personal website MattMahoney.net contains a lot more you might be interested in.

Interview with Dr.Matt Mahoney


Q. As you have done your Ph.D in compression technology, when exactly you started to think the compression still requires more algorithms, and why?

A. Actually I started my dissertation in data compression but I had to switch to intrusion detection to get funding. Data compression was still my first interest, however.

Q. When and what was the very first compression algorithm you wrote? tell us some about it.

A. It was a neural network compressor which I described in my paper, M. Mahoney, Fast Text Compression with Neural Networks, Proc. AAAI FLAIRS, Orlando, 2000 (C) 2000, AAAI. http://mattmahoney.net/dc/mmahoney00.pdf

Q. Do you still have the source code of your first algorithm you wrote?

A. Yes, the programs P5, P6, and P12 at http://mattmahoney.net/dc/paq.html#neural

Q. We see there are a lot of algorithms with source codes, which one do you think is the best one so far amongst them (the ones on your site) ? Speed-wise and compression-wise.

A. My fastest are SR2 and FLZP. The best compression is the PAQ series. Most of my programs are experiments with new algorithms.

Q. UCL and LZO are claimed to be the fastest and less memory consuming compression libraries (as per author), do you think they are, if yes then what will be the best competitor of UCL among your libraries / algorithms (and libraries of your fellow at your site). Where PAQ series seems quite impressive.

A. Those are faster than mine, but don't compress as well. For high speed with compression a little better than zip I would recommend SR2. For better compression I recommend LPAQ or PAQ9A.

Q. How many exactly are your own algorithms among the ones posted at your site sir? Because we see there are many contributed by your fans / fellow developers.

A. PAQ8L is the latest version of PAQ written entirely by me about 2 years ago. A lot of other people have made improvements on it since then. The others are ZPAQ, PAQ9A, LPAQ1, BBB, SR2, FLZP, and some of the FPAQ0 series. ZPAQ, PAQ9A and LPAQ1 are context mixing compressors designed for good compression but fast enough to be useful. BBB is a memory efficient BWT compressor, the first to use only 1.25x block size for memory, rather than 5x or 6x. SR2 is designed for speed. It is an improvement over the very old and rarely used symbol ranking algorithm program SRANK. FLZP is a preprocessor designed to improve compression by other compressors by removing high order redundancy. FPAQ0 is a series of experiments on order-0 arithmetic coders. I also implemented the first asymmetric binary coders (an idea by Jarek Duda) in FPAQA, FPAQB, FPAQC and LPAQ1A. These are all described at http://mattmahoney.net/dc.

Q. You definitely know today when keeping the source code open attracts companies/individuals to form commercial libraries/products (and keeping them closed source), result in the origin fades out, you ever thought about it?

A. I think it was a good decision to release the source code. You can't make money selling compression software because there is a lot of free competition. But you can make money selling your skills. PAQ would not have gotten to the top of the benchmarks without others improving it, and then I would not have gotten a good job at Ocarina working in my field of interest.

Q. When exactly you formed this site www.mattmahoney.net ? and did you code the html yourself?

A. Yes, I wrote all the HTML myself.

Q. What exactly is your job description at Ocarina Networks, I mean do you program there? or simply do research work and then that is implemented by juniors?

A. I work at home designing new algorithms for specialized data types and write the code. Others at the company will integrate my code into their products.

Q. Do you still teach anywhere?

A. I am no longer teaching but I have taught programming classes part time at Florida Tech from 1999 until this year.

Q. How easily a student can reach you sir? I mean a researcher doing MS or Ph.D might want to meet you for guidance, how exactly they can approach you? Through email? IRC? Chat etc?

A. Email or on the data compression forums like http://www.encode.ru/forum/

Q. Any other hobbies? other than going out for vacations and scientific research. And by the way, do your cats annoy you sometimes while working?

A. I enjoy running. The cats will occasionally walk on my keyboard but I don't mind them.

Thanks for your time Dr.Matt Mahoney. It is really a big honor for our Research Center to publish your interview.