News for hackers. Stuff that matters.
Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by userbinator on Sunday January 17, 2021 @08:20PM (#25817165)
...and LZ is such a simple algorithm - replacing repeated occurrences of a sequence with a shorter backreference - that I'm surprised it wasn't discovered sooner. Huffman wrote about his algorithm in the early 50s, but LZ only showed up in the late 70s.
Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by beagle3 on Monday January 18, 2021 @03:48AM (#25819646)
The important contribution is not in the simple algorithm (which I suspect others have come up with before, and I know others have independently come up with after):

Rather, it was what Lempel and Ziv proved about both of their simple algorithms -- The one published in 1977, usually referred to as LZ77 which is the basis of all <length,distance> variants, and the one published in 1978, usually referred to as LZ78 which is the basis of LZW, and other <word,char> variants.

They proved that for a very large class of sources (those described by markov models, context trees, among many others), these algorithms asymptotically give the best possible compression. In fact, some applications (also proved correct by Ziv) use the compressed datastream statistics to estimate source entropy, because it's much easier to implement than any direct measurement.

Both Ziv and Lempel did a lot of other things; However, Lempel did more in the area of switching systems and his other contributions are numerous and interesting, but were less profound in any specific field -- Whereas Ziv's contribution to Information Theory through the last 60 years or so has been much more profound.

Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by sn41 on Monday January 18, 2021 @04:26AM (#25819899)
The pointwise otimality of the LZ algorithm - the fact that it beats every other information-lossless finite-state compressor like Huffman encoding asymptotically on /every/ string, is an ingenious theorem. For versions of the algorithm (LZ77 in particular), it was open till the late 1990s when Paul Shields settled it.

The algorithm is easy, but the analysis of its effectiveness is not at all easy.

Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by tester34 on Monday January 18, 2021 @06:54AM (#25820825)
>...and LZ is such a simple algorithm -

then you have monsters like this

https://en.wikipedia.org/wiki/Asymmetric_numeral_systems

Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by andi999 on Monday January 18, 2021 @02:01AM (#25819042)
I think repeated sequences in data only became relevant when there was large image data. I mean, like Shannon was concerned about language transmitted.
Re:IEEE Medal of Honor Goes to Data Compression Pioneer Jaco... by beagle3 on Monday January 18, 2021 @04:02AM (#25819731)
Not at all. Shannon's work is incredibly profound, and deals with distortion of analog signals, channel capacity under different constraints, and many other things. He was very much aware and discussing repeated sequences in the 1952 class in which Huffman constructed his code and proved it is optimal. (According to lore, Shannon said "hey, Fano and I have this Shannon-Fano code, which we know is close but not optimal. Improve on it and you can skip the final")

Shannon's original paper, "A mathematical theory of communication" is very readable (especially discrete sources and channels; you need some calculus for the continuous sources and channel stuff), and yet touches on many things and has many insights. At least as late as 2000, one very-well-published prof at the uni I was attending told me that whenever he wants to break out to a new direction, he rereads it 5-10 times and realizes another thing that Shannon considered trivial and that was yet unexplored.

Also, there's basically no repeated sequences in large image data. JPEG, H264 and all their friends are all about compressible approximations that are still close enough, and which rarely if ever have repeated data.