Hamilton Institute Seminars (HD / large)

Speaker:

Dr. M. Flanagan

Abstract:

Doubly-generalized low-density parity-check (D-GLDPC) codes offer an attractive compromise between algebraic and random code design philosophies. In this talk we introduce the concept of D-GLDPC codes,and then provide a solution for the asymptotic growth rate of the weight distribution of any D-GLDPC ensemble. This tool is then used for detailed analysis of a case study, namely, a rate-1/2 D-GLDPC ensemble where all the check nodes are (7,4) Hamming codes and all the variable nodes are length-7 single parity-check codes. It is illustrated how the variable node representations can heavily affect the code properties and how different variable node representations can be combined within the same graph to enhance some of the code parameters. The analysis is conducted over the binary erasure channel. Interesting features of the new codes include the capability of achieving a good compromise between waterfall and error floor performance while preserving graphical regularity, and values of threshold outperforming LDPC counterparts.

Speaker:

Prof. E.C. Velasco

Abstract:

Volterra integral and difference equations may be used to model the dynamics of physical systems (viscoelasticity, motion of bodies with reference to hereditary) and biological systems (populations dynamics, biomechanics). In this talk we discuss about asymptotic properties of solutions of both, Volterra integral and Volterra difference equations. For the Volterra difference equations, we derive stability conditions based on the direct Lyapunov method and present some examples to illustrate them.

Speaker:

Prof. M. Alanyali

Abstract:

This talk will touch on wireless coexistence issues that arise due to higher spatial density of spectrum usage. We consider a fairness perspective for autonomous scheduling of transmissions by distinct sessions, subject to constraints that are represented by a conflict graph. The emphasis is on randomized backoff-based CSMA algorithms. The resulting transmission dynamics is represented by a Markovian model whose analysis suggests practical challenges in fair sharing of spectrum by distinct sessions that subscribe to a common standard, as well as by those that do not possess a common signaling protocol.

Speaker:

Dr. F. Markowetz

Abstract:

Cellular mechanisms are driven by interactions between proteins, DNA and RNA, working together in cellular pathways. Current knowledge of information flow in the cell is still very incomplete and dissection of cellular pathways is one of the major challenges of systems biology. Computational approaches integrating heterogeneous genomic data sources into one joint model promise a comprehensive view on cellular processes. However, to be successful, computational methods need to account for the specific features of each data source. In this talk I will focus on data from gene perturbation experiments, where individual pathway members are experimentally silenced and effects of these perturbations are measured in genomic assays. I will describe Nested Effects Models, a probabilistic graphical model especially designed to reconstruct signaling pathways from gene perturbation data.

Speaker:

Dr. Björn Schelter

Abstract:

Nowadays, data are recorded with increasing spatio as well as temporal resolution. This calls for new methods to analyze these data sets. Caused by the high spatio as well as temporal resolution of the recorded signals, inference of the causal network structure underlying them becomes feasible. In many applications a detailed analysis of these networks allows deeper insights into the normal functioning or malfunctioning of the system. In Neurology this helps to understand certain diseases like epilepsy or Parkinsons disease.

Novel concepts to analyze multivariate data consisting of both time series as well as point processes will be presented. By means of an application to tremor in Parkinsons disease, the abilities and limitations of these techniques are discussed.

Speaker:

Dr. Rudolf Kulhavý

Abstract:

There is a common perception in todays business that the world around us becomes less hierarchical and more networked and flat. While the shift towards a networked and decentralised business environment generally creates more freedom to act, it does not increase automatically the chances of success. Understanding the dynamics of networked systems — in particular the interplay between the performance of an individual node and of the entire network, and the importance of effective bonding for the well-being of an organisation — becomes a critical skill. Replacing mental models with a formal, quantitative model can improve such understanding and ultimately allow for systematic network optimisation. To this end, we propose to combine stochastic system dynamics modelling of individual nodes with probabilistic graphical modelling of a network configuration. The latter is closely related to theoretical constructs such as the Ising model in statistical mechanics or Markov random fields in image analysis. Modelling of value networks in business turns out to be even more complex because of the random structure of a network. In this talk, we discuss the economic substance and mathematical representation of node-to-node bonds, formulate a general Bayesian solution to the problem of estimating unknown state and parameter values in the resulting model, and discuss its Markov chain Monte Carlo implementation. To illustrate the concepts introduced, we revisit Clayton Christensens qualitative model of the dynamic behaviour of new entrants versus incumbents when dealing with sustaining and disruptive innovation — and consider its reformulation as a probabilistic interaction network. We conclude by looking outside business for other instances of value networks.

Speaker:

Dr. M. Cryan

Abstract:

Suppose we are given two lists r and c of positive integers, where r=(r[1],...., r[m]) represents a list of prescribed row sums and c=(c[1], ..., c[n]) is a list of prescribed column sums. We require that (r[1] + ... + r[m]) =(c[1] + ... + c[n]). In this setting, we say that a m-by-n matrix X of non-negative integers is a Contingency Table (for these given row/column values) if X simultaneously satisfies all of the given row and column sums. The problem of determining whether at least one contingency table exists can be solved in polynomial-time (in fact, this question is fairly trivial).

In my talk, we are interested in the more-difficult problem of randomly sampling a table uniformly at random, from the entire set of contingency tables. This problem has some applications in practical statistics which I will mention. We study a very natural Markov chain on the set of contingency tables called the 2-by-2 heat bath: one step of this chain operates by selecting 2 rows and 2 columns uniformly at random, computing the induced row sums and column sums on this 2-by-2 window, then replacing the window with a table chosen randomly from all 2-by-2 tables with the induced row and column sums. This Markov chain converges to the uniform distribution on contingency tables - our goal is to show that it approaches uniformity within polynomial-time. We are able to achieve this result for the case when the number of rows m is some fixed constant. Our proof is by application of the canonical paths method of Jerrum and Sinclair.

(Joint work with Martin Dyer, Leslie Goldberg, Mark Jerrum and Russell Martin)

Speaker:

Dr. D. Giustiniano

Abstract:

ADSL is becoming the standard form of residential and small-business broadband access to the Internet due, primarily, to its low deployment cost. These ADSL residential lines are often deployed with Access Points (AP) that provide wireless connectivity. While the ADSL technology has showed evident limits in terms of capacity, the short-range wireless communication can guarantee a similar or higher capacity. Even more important, it is often possible for a residential wireless client to be in range of several other APs belonging to nearby neighbors with ADSL connections. Therefore, it is possible for a wireless client to simultaneously connect to several APs in range and effectively aggregate their available ADSL bandwidth. Recent works have shown promising results within this area, but main important questions are still unresolved:i) how can we guarantee a fair distributed bandwidth allocation among clients? ii) how the latency of TCP connection can be affected by AP connections over multiple frequencies? iii) how can we minimize the MAC cost of managing these multiple APs? In order to answer to these questions, we introduce ClubADSL, a prototype wireless client that can aggregate the capacity of multi-frequency APs. ClubADSL achieves fairness through distributed pressure schemes and minimizes the impact of end-to-end latency on the system performance with a resource allocation scheme based on Access-Point slot assignment. We show the feasibility of such a system in seamlessly transmitting TCP traffic, and validate its experimental implementation over commodity hardware in controlled scenarios. [Joint Work with Alberto Lopez, Eduard Goma, Julian Morillo,Pablo Rodriguez].

Speaker:

Dr. W. D. Smith

Abstract:

We describe a new simple but more powerful form of linear cryptanalysis. It appears to break AES (and undoubtedly other cryptosystems too, e.g. SKIPJACK).

*But the break is "nonconstructive".

*Even if this break is broken (due to the underlying models inadequately approximating the real world) we explain how AES still could contain "trapdoors" which would make cryptanalysis unexpectedly easy for anybody who knew the trapdoor.

We then discuss how to use the theory of BLECCs to build cryptosystems provably

*not containing trapdoors of this sort,

*secure against our strengthened form of linear cryptanalysis,

*secure against "differential" cryptanalysis,

*secure against D.J. Bernstein's timing attack.

Using this technique we prove a fundamental theorem: it is possible to thus encrypt N bits with security 2^(cN), via an circuit Q_N containing <= cN two-input logic gates and operating in <= c log(N) gate-delays, where Q_N is constructible in polynomial (in N) time.

Speaker:

Prof. C. Dovrolis

Abstract:

The issue of router buffer sizing is still open and significant. Previous work either considers open-loop traffic or only analyzes persistent TCP flows. Our work differs in two ways. First, it considers the more realistic case of non-persistent TCP flows with heavy-tailed size distribution. Second, instead of only looking at link metrics, we focus on the impact of buffer sizing on TCP performance. Through a combination of test bed experiments, simulation, and analysis, we reach the following conclusions: The output/input capacity ratio at a network link largely determines the drops exponentially with the buffer size and the optimal buffer size is close to zero. Otherwise, if the output/input capacity ratio is lower than one, the loss rate follows a power-law reduction with the buffer size and significant buffering is needed, especially with flows that are mostly in congestion-avoidance. Smaller transfers, which are mostly in slow-start, require significantly smaller buffers. We conclude by revisiting the ongoing debate on "small versus large" buffers from a new perspective.