Georgios Alexandros Dimakis

I am a graduate student in the EECS Department
University of California at Berkeley
My office is in Cory Hall 264 in Wireless Foundations
Berkeley, CA 94720-1772
Phone no:510-643-9263

Interests | Publications | Links | Resume
Selected Writings on:
Coding for Distributed Systems
Gossip and Message Passing for Sensor Networks
LDPC codes and Pseudocodewords
Game theory
Statistical Signal Processing



Research Interests

I am interested in communications, signal processing and networking with applications in large-scale distributed systems and sensor networks.
I work with Kannan Ramchandran and Martin Wainwright
I am a member of Wireless Foundations and the BASiCS research group

I completed my undergraduate studies in Electrical and Computer Engineering department
National Technical Univerisity of Athens in 2003. My advisor was Petros Maragos

I received my Masters from UC Berkeley in 2005 and currently working towards my PhD.
During the summer of 2006 I visited LCAV in EPFL. I had the pleasure of working with Martin Vetterli and Patrick Thiran.
During the summer of 2007 I was with the Communication and Collaboration Systems group in Microsoft Research. With Yunnan Wu we worked on coding for Distributed Storage in Data Centers.


Hobbies

I also like to draw and paint occasionally. Have a look.



Research Publications


What's new:

  • Received the Eliahu Jury award for Communications, Control, or Signal Processing
  • Uploaded journal version of the network coding for distributed storage in data centers.
  • Uploaded our book chapter on distributed storage
  • Uploaded our ITW paper on LDGM codes for lossy compression (and slides)
  • Uploaded journal version of Geographic gossip.
  • See some blog entries on ITW, Allerton and MMSP in the Ergodic Walk.
  • Uploaded our paper on the existence of Pure Nash Equilibria in random games
  • Robust Message passing paper (IPSN'07) uploaded
  • I received the Microsoft Research Fellowship. This support is gratefully acknowledged.
  • I finally made pdf slides for Geographic Gossip ,Linear Programming decoding , Facet Guessing decoding and Decentralized erasure codes (email me for Powerpoint version).
  • Infocom 2007 paper on network coding for peer-to-peer storage uploaded,
  • IT paper on LP Decoding uploaded (extended version of SODA'07).


    Selected publications by subject

    Coding for Distributed systems

  • "Network Coding for Distributed Storage Systems" (new)
    A. G. Dimakis, P. B. Godfrey, Y. Wu, M. Wainwright and K. Ramchandran, submitted for publication.
  • (Preliminary versions appeared in Infocom 2007 and Allerton 2007)
    Data centers and sensor networks require reliable information storage over individually unreliable nodes. Storing a file using an erasure code, in fragments spread across nodes, requires less redundancy than simple replication for the same level of reliability. However, since fragments must be periodically replaced as nodes fail, a key question is how to generate a new encoded fragment in a distributed way while transferring as little data as possible across the network. For an erasure coded system, a common practice to repair from a node failure is for a new node to download subsets of data stored at a number of surviving nodes, reconstruct a lost coded block using the downloaded data, and store it at the new node. We show that this procedure is sub-optimal. We introduce the notion of regenerating codes, which allow a new node to download functions of the stored data from the surviving nodes. We show that regenerating codes can significantly reduce the repair bandwidth. Further, we show that there is a fundamental tradeoff between storage and repair bandwidth which we theoretically characterize using flow arguments on an appropriately constructed graph. By invoking constructive results in network coding, we introduce regenerating codes that can achieve any point in this optimal tradeoff.

  • "Network Coding for Distributed Storage in Wireless Networks"
    A. G. Dimakis and K. Ramchandran, Chapter in edited volume, Networked Sensing Information and Control, Signals and Communication Series, V. Saligrama (Ed.) Springer Verlag, 2007. (to appear).

  • "Unequal Growth Codes: Intermediate Performance and Unequal Error Protection for Video Streaming"
    A. G. Dimakis, J. Wang, K. Ramchandran, International Workshop on Multimedia Signal Processing (MMSP), October 2007.
    (Best Student Paper Runner-up)

  • "Decentralized Erasure Codes for Distributed Networked Storage"
    A. G. Dimakis, V. Prabhakaran, K. Ramchandran, IEEE Transactions on Information Theory, June 2006.
  • "Ubiquitous Access to Distributed Data in Large-Scale Sensor Networks through Decentralized Erasure Codes"
    A. G. Dimakis, V. Prabhakaran and K. Ramchandran, Symposium on Information Processing in Sensor Networks (IPSN '05)
    (Recipient of the Best Paper Award)
    Slides for Decentralized erasure codes (for Powerpoint slides, email me)
  • Investigates the problem of minimizing the communication required to construct an erasure code in a network. Decentralized erasure codes can be seen as linear randomized network coding on an optimally sparse random bipartie graph. The code construction is happening by randomly and independently moving the minimal number of messages in the network.

  • "Codes on Graphs for Distributed Storage in Wireless Networks"
    Alexandros G. Dimakis, Master thesis report, UC Berkeley, Fall 2005.


  • Gossip and Message Passing Algorithms

  • Geographic Gossip : Efficient Averaging for Sensor Networks
    A. G. Dimakis, A.D. Sarwate, and M.J. Wainwright. To appear, IEEE Transactions on Signal Processing.
  • Conference version:
  • Geographic Gossip : Efficient Aggregation for Sensor Networks
    A. G. Dimakis, A.D. Sarwate, and M.J. Wainwright.in Proceedings of the 5th International ACM/IEEE Symposium on Information Processing in Sensor Networks (IPSN '06) , Nashville, TN, April 2006.
    Slides for Geographic Gossip

    Gossip (or average Consensus) algorithms, are simple randomized message-passing algorithms that can be used to compute global functions of data distributed over a network with minimal coordination. The proposed gossip algorithm (Geographic Gossip) uses geographic informationto dramatically improve the convergence time, hence minimizing the number of communicated messages.

  • Order-Optimal Consensus through Randomized Path Averaging (new)
    F. Benezit, A. G. Dimakis, P. Thiran and M. Vetterli. Submitted for publication. Preliminary version appeared in Allerton 2007.
  • We extend Geographic gossip by allowing all the nodes on a random path to be averaged and show how this path averaging can be implemented with minimal complexity in a sensor network if sensors have location information. The main result is that Geographic Gossip with path averaging converges using a number of messages linear in the number of nodes, which improves by a sqrt(n) factor over geographic gossip and is order-optimal.

  • Robust Message Passing for Statistical Inference in Sensor Networks
    J. Schiff, D. Antonelli, A.G. Dimakis, D. Chu, and M.J. Wainwright in Proceedings of the 6th International ACM/IEEE Symposium on Information Processing in Sensor Networks (IPSN '07), Cambridge, MA, April 2007.

    Large-scale sensor network applications require in-network processing and data fusion to compute statistically relevant summaries of the sensed measurements. We study distributed message-passing algorithms, in which neighboring nodes in the network pass local information relevant to a global computation, for performing statistical inference. We focus on the class of reweighted belief propagation (RBP) algorithms, which includes as special cases the standard sum-product and max-product algorithms for general networks with cycles, but in contrast to standard algorithms has attractive theoretical properties (uniqueness of fixed points, convergence, and robustness). Our main contribution is to design and implement a practical and modular architecture for implementing RBP algorithms in real networks. In addition, we show how intelligent scheduling of RBP messages can be used to minimize communication between motes and prolong the lifetime of the network. Our simulation and Mica2 mote deployment indicate that the proposed algorithms achieve accurate results despite realworld problems such as dying motes, dead and asymmetric links, and dropped messages. Overall, the class of RBP provides provides an ideal fit for sensor networks due to their distributed nature, requiring only local knowledge and coordination, and little requirements on other services such as reliable transmission.


    LDPC Codes and Linear Programming Decoding

  • Lower bounds on the Rate-Distortion function of LDGM codes
    A. G. Dimakis, M.J. Wainwright, and K. Ramchandran. In Information Theory Workshop (ITW) 2007.
  • We show that LDGM codes with constant degrees will have w.h.p. a constant gap from optimal rate distortion for binary symmetric sources. Our result can be seen as the dual of Gallager's classical result for LDPC codes with constant degrees having a gap from capacity. (journal version in preparation) Slides from ITW

  • "Probabilistic Analysis of Linear Programming decoding" (Updated)
    C. Daskalakis, A. G. Dimakis, R. M. Karp and M. J. Wainwright,
    to appear, IEEE Transactions on Information Theory, shorter version appeared in ACM-SIAM Symposium on Discrete Algorithms (SODA), 2007.
    Slides for probabilistic analysis of LP decoding
  • Abstract: We initiate the probabilistic analysis of linear programming (LP) decoding of low-density parity-check (LDPC) codes. Specifically, we show that for a random LDPC code ensemble, the linear programming decoder of Feldman et al. succeeds in correcting a constant fraction of errors with high probability. The fraction of correctable errors guaranteed by our analysis surpasses all prior non-asymptotic results for LDPC codes, and in particular exceeds the best previous finite-length result on LP decoding by a factor greater than ten. This improvement stems in part from our analysis of probabilistic bit-flipping channels, as opposed to adversarial channels. At the core of our analysis is a novel combinatorial characterization of LP decoding success, based on the notion of a generalized matching. An interesting by-product of our analysis is to establish the existence of "probabilistic expanders" i.e. bipartite graphs where almost every (as opposed to every) set (and its subsets) expands, with expansion coefficients much larger than the classical case.

  • "Guessing Facets: Polytope Structure and Improved LP Decoding"
    A. G. Dimakis, A. A. Gohari and M. Wainwright, submitted for publication, IEEE Transactions on Information Theory. (email me for the paper)
  • "Guessing Facets: Polytope Structure and Improved LP Decoder"
    A. G. Dimakis and M. Wainwright, IEEE International Symposium on Information Theory (ISIT), 2006.
    Slides for Facet Guessing decoding
  • Abstract: A new approach for decoding binary linear codes by solving a linear program (LP) over a relaxed codeword polytope was recently proposed by Feldman et al. In this paper we investigate the structure of the polytope used in the LP relaxation decoding. This relaxed polytope has both integral vertices which correspond to codewords and non-integral vertices called fractional pseudocodewords. We show that for expander codes, every fractional pseudocodeword always has at least a constant fraction of non-integral bits. We also show that codewords have much more facets touching them relative to fractional pseudocodewords. Based on these results on the polytope structure, we propose an decoding algorithm which provably outperforms the LP decoder for finite blocklengths. The proposed algorithm proceeds by guessing facets of the polytope and resolving the linear program on these facets. Guessing facets is a direct generalization of bit guessing decoders which have been explored in the recent literature. While the LP decoder succeeds only if the ML codeword has the highest likelihood over all pseudocodewords, we prove that for expander codes the proposed algorithm succeeds even with a constant number of pseudocodewords of higher likelihood. The complexity of the proposed algorithm is only a constant factor larger than that of the LP decoder.


    Game Theory and Applied Probabiilty

  • "Connectivity and Equilibrium in Random Games"
    C. Daskalakis, A. G. Dimakis and E. Mossel
    submitted for publication.

  • We study how the structure of the interaction graph affects the Nash equilibria of the resulting game. In particular, for a fixed interaction graph, we are interested if there exist Nash equilibria which arise when random utility tables are assigned to the players. We provide conditions for the structure of the graph under which equilibria are likely to exist and complementary conditions which make the existence of equilibria highly unlikely. Our results have immediate implications for many deterministic graphs and generalize known results for games on the complete graph. In particular, our results imply that the probability that bounded degree graphs have Nash equilibria is exponentially small in the size of the graph and yield a simple algorithm that finds small non-existence certificates for a large family of graphs. In order to obtain a refined characterization of the degree of connectivity associated with the existence of equilibria, we study the model in the random graph setting. In particular, we look at the case where the interaction graph is drawn from the Erdos-Renyi, G(n,p), model where each edge is present independently with probability p. For this model we establish a double phase transition for the existence of pure Nash equilibria as a function of the average degree p n consistent with the non-monotone behavior of the model. We show that when the average degree satisfies n p > (2 + Omega(1)) log n, the number of pure Nash equilibria follows a Poisson distribution with parameter 1. When 1/n << n p < (0.5 -Omega(1)) log n pure Nash equilibria fail to exist with high probability. Finally, when n p << 1/n a pure Nash equilibrium exists with high probability.


    Statistical Signal Processing

    My undergraduate thesis was supervised by Prof. Maragos. We proposed a nonlinear model for time-varying resonances where the instantaneous phase of a sinusoidal oscillation is a self-similar stochastic process. We have theoretically derived the second order statistics of the modulated process for a very general class of Self-similar processes that include Fractional Brownian Motion and Fractional Stable Levy Motion as special cases. We have also applied these ideas to data from fricative phonemes and demonstrated their applicability for speech modelling.

    (Copyright belongs to the publisher in most cases. The author/authors assert copyright in all other cases.)


    Links

    I was the Webmaster of the IEEE Student branch in National Technical University of Athens (NTUA). Visit the site .

    Fun upper bounds: My Erdos Number is less than or equal to 3: (Paul Erdos-> Noga Alon-> Richard Karp-> myself)
    No known finite upper bounds on my Bacon number exist.

    Friends

    Tasos Argyros
    Charis Ermopoulos
    Alexandra Meliou
    Themos Kallos
    Apostolos Fertis
    Emmanouil Kioupakis
    Victor Panaretos
    Konstantinos Daskalakis
    Nikolas Geroliminis
    Krish Eswaran
    Bobak Nazer
    Vinod Prabhakaran
    Ben Rubinstein
    Anand Sarwate