Pasquale Malacaria, Reader in Computer Science,
School of Electronic Engineering
and Computer Science
Queen Mary University of London, Mile End Road, E1 4NS
+44 020 7 8825231
pm@dcs.qmul.ac.uk
Main interest:Information Theory and its uses in computer security.
For example, suppose someone stole your cash card (but doesn't know your pin number). How much information about your pin number can he get by inserting your card in a cash machine and trying his luck?
Here is an old poster describing
in simple terms the basis of this work. The motivation of this research is simple really: systems are never 100% secure, so can we measure how secure they are?
(This may be a good paper to read if you like the idea.)
Current work: quantitative analysis of leakage of real code, e.g. quantifying leakage of confidential information in the Linux Kernel. The paper (to be presented at ACSAC 2010 ) is here .
Recent work: ( PLAS 2008 ,
ASIACCS 2009) addresses the problem of maximal leakage of secure information, translating it into the information theoretical problem
of channel capacity. The problem is then approached using Lagrange multipliers.
More recent work use also Lagrange Multipliers to compute loss of anonymity in anonymity protocols
( ASIACCS 2009 ).
principal investigator in the EPSRC project "Quantitative Information Flow",
( Click here for details.)
principal investigator in the EPSRC project "Model Checking and Program Analysis for Quantifying Interference".
co-investigator in the EPSRC platform grant "Extreme Reasoning".
Interested in a PhD? Funding available, further details
click here.
Applicants should be able to show some knowledge and interest in my areas of research.
Also interested in applications of information theory to
machine learning in particular
Adaboost. Adaboost significantly boosts the performance of
classifiers, which are algorithms that allow to classify new cases based on previous cases. An example could be a
program which classify if a patient has/has not cancer given the symptoms based on a database of previous cases (this database is called the training set).
In this paper we related Adaboost with Kelly's theory of optimal betting and by doing so we significantly simplify the Adaboost algorithm.
Past works: game semantics and its algorithmic applications.
Game semantics provides a very fine graded analysis of computation. It has been used to successfully answer some open questions in the semantic of programming languages (Full abstraction for PCF, see publication sections).
Here are some motivations for that work.
Corrado Bohm supervised my master thesis on second order lambda calculus. Another remarkable scientist,
Jean-Yves Girard supervised my PhD on Stone duality.
