9:57 AM: My paper is being presented right now in Rateless Codes II. I spent the morning practicing with the speaker.
10:06 AM: The talk is over. Good job Bertrand on your first ISIT talk.
11:07 AM: A very interesting paper by Venkiah et al. on the design of rateless codes over ranges of channel parameters. Their design technique uses EXIT charts and explicitly optimizes the degree sequence over mulitple capacities.
Friday, July 11, 2008
Thursday, July 10, 2008
ISIT Liveblog: July 10
Having some network troubles this morning ...
10:50 AM: Attended "Sensor Networks". The first talk was about optimization of sensor networks on 2-d correlated random fields. The work looks interesting but unfortunately I didn't get much out of the talk.
12:09 PM: Just for fun, I'm sitting in a session on error exponents. There was a very interesting talk on a new class of exponents that beat the Gallager bounds, as well as an easily computable approximation to this bound. The following talk was done on handwritten overhead slides, which I haven't seen in use in about five years. It was about the calculation of error exponents via point processes ... with the overhead slides, it could have been deadly, but the speaker was entertaining and interesting, in spite of the fact that I understood very limited amounts of the talk.
2:56 PM: In "Detection and Estimation": Fascinating paper on inversion of matrices by message-passing over "random fields" using the Gaussian distribution. The method seems quite clever but I didn't exactly follow how the imputed random model is produced ... I will be sure to look this paper up in the proceedings.
4:31 PM: In "Rateless Codes 1": a raptor code-based distributed storage algorithm. Sounds an awful lot like both network coding and peer-to-peer applications that I've heard of before. Also a talk about complexity-performance tradeoffs in rateless codes -- the main idea is to discard any output symbol that is sufficiently unreliable. A paper about "graph-based" LT codes, containing only output degrees 1 and 2.
10:50 AM: Attended "Sensor Networks". The first talk was about optimization of sensor networks on 2-d correlated random fields. The work looks interesting but unfortunately I didn't get much out of the talk.
12:09 PM: Just for fun, I'm sitting in a session on error exponents. There was a very interesting talk on a new class of exponents that beat the Gallager bounds, as well as an easily computable approximation to this bound. The following talk was done on handwritten overhead slides, which I haven't seen in use in about five years. It was about the calculation of error exponents via point processes ... with the overhead slides, it could have been deadly, but the speaker was entertaining and interesting, in spite of the fact that I understood very limited amounts of the talk.
2:56 PM: In "Detection and Estimation": Fascinating paper on inversion of matrices by message-passing over "random fields" using the Gaussian distribution. The method seems quite clever but I didn't exactly follow how the imputed random model is produced ... I will be sure to look this paper up in the proceedings.
4:31 PM: In "Rateless Codes 1": a raptor code-based distributed storage algorithm. Sounds an awful lot like both network coding and peer-to-peer applications that I've heard of before. Also a talk about complexity-performance tradeoffs in rateless codes -- the main idea is to discard any output symbol that is sufficiently unreliable. A paper about "graph-based" LT codes, containing only output degrees 1 and 2.
Wednesday, July 9, 2008
ISIT Liveblog: July 9
11:02 AM: Interesting plenary by Robert Gray on source coding -- it's not really my field, so I was lost on some of the more technical discussion, but it was a nice overview of his work.
12:06 PM: A couple of interesting talks by Urbanke's students on the scaling law for finite-length LDPC code analysis. I was unaware of this work, which has been around since 2003, but it looks like an efficient and accurate way to estimate the performance of short LDPC codes.
It's a half day today, so that's it for the conference until tomorrow ...
12:06 PM: A couple of interesting talks by Urbanke's students on the scaling law for finite-length LDPC code analysis. I was unaware of this work, which has been around since 2003, but it looks like an efficient and accurate way to estimate the performance of short LDPC codes.
It's a half day today, so that's it for the conference until tomorrow ...
Tuesday, July 8, 2008
ISIT Liveblog: July 8
10:07 AM: Sadly missed the plenary. However, I did find the following in this morning's Globe and Mail: Chess boxing! Yes, it's exactly what it sounds like.
10:28 AM: Very interesting talk by Christine Kelley on LDPC code constructions from "voltage graphs", a special algebraic way of representing graph-based codes, which allow easier identification of cycles.
11:03 AM: Neat paper by Ardakani et al on universal LDPC codes -- i.e., designing an LDPC code to converge in two different channels (e.g., BSC and BEC) with the same capacity. An intriguing conjecture: "An LDPC code which converges over two channels with equal capacity will also converge over any channel in the convex hull of those channels" -- they claim this conjecture can be proven for special cases. They also give a set of "basis" subchannels for each capacity, for which each channel can be expressed as a convex combination of those channels.
3:41 PM: I sat in on a session of "Estimation" ... there was a nice paper on entropy estimation, and one on change point estimation, as well as an interesting paper by Loeliger et al. on estimation in 2-d Markov models.
5:51 PM: In "New code constructions", a paper by Grover and Sahai takes a holistic view of power consumed in communication over short range links, combining power used in computation and communication. A neat application for the traditional complexity-performance problem.
10:28 AM: Very interesting talk by Christine Kelley on LDPC code constructions from "voltage graphs", a special algebraic way of representing graph-based codes, which allow easier identification of cycles.
11:03 AM: Neat paper by Ardakani et al on universal LDPC codes -- i.e., designing an LDPC code to converge in two different channels (e.g., BSC and BEC) with the same capacity. An intriguing conjecture: "An LDPC code which converges over two channels with equal capacity will also converge over any channel in the convex hull of those channels" -- they claim this conjecture can be proven for special cases. They also give a set of "basis" subchannels for each capacity, for which each channel can be expressed as a convex combination of those channels.
3:41 PM: I sat in on a session of "Estimation" ... there was a nice paper on entropy estimation, and one on change point estimation, as well as an interesting paper by Loeliger et al. on estimation in 2-d Markov models.
5:51 PM: In "New code constructions", a paper by Grover and Sahai takes a holistic view of power consumed in communication over short range links, combining power used in computation and communication. A neat application for the traditional complexity-performance problem.
Monday, July 7, 2008
ISIT Liveblog: July 7
8:48 AM: I'm in Calderbank's very interesting plenary on sensing and detection in the context of algebraic codes. Carefully designed waveforms with good autocorrelation properties can be used in radar sensing. I saw an earlier version of this talk in Edmonton at CWIT 2007. Calderbank likes Blahut's coding book, and commented: "Classical coding theory and Fourier analysis are two sides of the same coin". But, as a "modern" coding theorist, I'll have to think a bit about what that means.
10:21 AM: Skipping between Iterative Decoding and the Wiretap Channel. Nice talk about bilayer LDPC codes, designed with respect to a rate constraint, in Iterative Decoding. In Wiretap Channels, I saw "Physical Layer Encryption with Stream Ciphers". I'm a little confused about this paper: using ciphers, there seems to be no reason to enhance traditional cryptography with physical layer security; and they claim the eavesdropper can obtain the key via plaintext attacks (which is true traditionally, and what if the transmitted data is compressed or otherwise equiprobable 1/2-1/2?). I think the contribution here is more along the lines of protecting encrypted text with error-correcting codes.
10:35 AM: Neat paper by Gunduz, Erkip and Poor on multiterminal compression with security constraints: eavesdropper can observe a subset of the terminals; only the "decipherer" (i.e., destination) can see all of them.
I think I get the wiretap channel now: given a side channel that is secure (e.g., secured with traditional cryptography, or otherwise private), it is possible to use information theory to make a much larger system private.
12:13 PM: In "Wireless Networks", a very interesting generalization to scaling laws by Ozgur et al., revealing a new operating regime for a wireless network that ignores the area density of nodes.
5:35 PM: Here's something I didn't know (if I remember it right): if z = x+y, then 2^(2H(Z)) >= 2^(2H(X)) + 2^(2H(Y)), with equality if and only if x, y, and z are Gaussian.
10:21 AM: Skipping between Iterative Decoding and the Wiretap Channel. Nice talk about bilayer LDPC codes, designed with respect to a rate constraint, in Iterative Decoding. In Wiretap Channels, I saw "Physical Layer Encryption with Stream Ciphers". I'm a little confused about this paper: using ciphers, there seems to be no reason to enhance traditional cryptography with physical layer security; and they claim the eavesdropper can obtain the key via plaintext attacks (which is true traditionally, and what if the transmitted data is compressed or otherwise equiprobable 1/2-1/2?). I think the contribution here is more along the lines of protecting encrypted text with error-correcting codes.
10:35 AM: Neat paper by Gunduz, Erkip and Poor on multiterminal compression with security constraints: eavesdropper can observe a subset of the terminals; only the "decipherer" (i.e., destination) can see all of them.
I think I get the wiretap channel now: given a side channel that is secure (e.g., secured with traditional cryptography, or otherwise private), it is possible to use information theory to make a much larger system private.
12:13 PM: In "Wireless Networks", a very interesting generalization to scaling laws by Ozgur et al., revealing a new operating regime for a wireless network that ignores the area density of nodes.
5:35 PM: Here's something I didn't know (if I remember it right): if z = x+y, then 2^(2H(Z)) >= 2^(2H(X)) + 2^(2H(Y)), with equality if and only if x, y, and z are Gaussian.
Thursday, July 3, 2008
See my paper at ISIT
My paper at ISIT:
Authors: B. Ndzana Ndzana, A. W. Eckford, M. A. Shokrollahi, and G. I. Shamir
Title: Fountain codes for piecewise stationary channels
Session: Fr-AM-6 (Friday, July 11, morning)
Location: Dominion North
See you in Toronto.
Authors: B. Ndzana Ndzana, A. W. Eckford, M. A. Shokrollahi, and G. I. Shamir
Title: Fountain codes for piecewise stationary channels
Session: Fr-AM-6 (Friday, July 11, morning)
Location: Dominion North
See you in Toronto.
Subscribe to:
Posts (Atom)