Dr Kizito Salako

Overview

Dr Kizito Salako is an applied mathematician and software developer at the Centre for Software Reliability (CSR), City University London. He holds a first-class double honours degree in Mathematics and Statistics from the University of Lagos; a Master of Advanced Study in Mathematics degree from the University of Cambridge (where he was both a Shell Centenary Scholar and a Commonwealth Scholar); and a PhD in Computer Science from City, University of London.

Kizito is passionate about applications of probability theory, Bayesian statistics, geometry and machine-learning techniques, when trying to simulate, assess and forecast the (failure) behaviour of software-based systems. His doctoral thesis clarifies and significantly extends the applicability of a family of probabilistic models used to describe the failure behaviour of multi-version software. He also developed a novel geometric approach to extremising the expected system reliability of a class of fault-tolerant software-based systems; so-called 1-out-of-N systems. Currently, he is particularly interested in the assessment/forecasting challenges that arise when these systems rely on evolving machine-learning solutions.

Since joining CSR, Kizito has contributed to several projects (e.g. DISPO, DIRC, IRRIIS, PIA-FARA, AFTER, DIDERO-PC, DISIEM). He is experienced in building simulations of complex systems using C++, and was instrumental in developing the PIA-FARA simulation engine -- creating modular software that implements and analyses the probabilistic, functional and process relationships that may exist between the components of large-scale interdependent complex critical infrastructure.

Research Interests

Quantitative assessment of the dependability of software-based systems (in particular, systems that are complex, safety/security critical, and that may rely on evolving machine learning solutions):

The development, validation and application of advanced statistical approaches for dependability assessment;
Conservative Bayesian assessment methods, that take into account various forms of dependability evidence when assessing system dependability;
Monte-Carlo methods for simulating complex systems (where these systems are modelled as Generalised Semi-Markov and Markov Regenerative Processes);
The combined use of probabilistic conditional independence relations and physics models in modelling interdependencies between complex systems
The efficacy (in terms of improved reliability) of using diverse-redundant software in fault-tolerant configurations


Statistical forecasting/analyses of system risk, dependability and cyber-security issues:

Investigating the impact of software development strategies on system reliability;
Studying the efficacy of mitigation strategies for a system under cyber-attack;
Studying the efficacy of diverse redundant system architectures in mitigating cyber-attacks, and the implications of such strategies for the likelihood of confidentiality and integrity breaches;

Research Projects

Kizito has collaborated on several national and international research projects. Details of these projects can be found here. The projects include:

DISIEM (funded by the EU horizon-2020 program) Diversity Enhancements for Security Information and Events Management Systems, 2016–2019
CEDRICS (funded by the UK’s EPSRC) Communicating and Evaluating Cyber–Risks and Dependencies, 2014–2017
AFTER (EU funded FP7 project) A Framework for sysTems vulnerability identification, dEfence and Restoration, 2011–2014
DIDERO-PC (funded by the UK’s EPSRC) DIverse DatabasE ReplicatiOn Performance Comparison, 2013–2014
PIA-FARA (funded by the UK’s Technology Strategy Board) Probabilistic Interdependency Analysis: Framework/data-Analysis/Risk–Assessment, 2009–2010
CETIFS (funded by the UK’s Technology Strategy Board) A Critical Infrastructure Interdependency Modelling Feasibility Study, 2008
IRRIIS (EU funded FP6 Project) Integrated Risk–Reduction of Information-based Infrastructure Systems, 2006–2009
DIRC (funded by the UK’s EPSRC) Dependability Integrated Research Collaboration, 2000–2006

Mathematical Interests

Probability Theory and Mathematical Statistics:

measure theoretic probability – probability spaces, integration, conditional expectation, characteristic functions, Radon-Nikodym derivatives and change-of-measure;
stochastic models and processes – filtered probability spaces; stationary/non-stationary; time-series analyses; point processes; order-statistics models; Gaussian processes; Markov processes (Markov chains, Markov renewal processes, Markov reward processes, Markov decision processes, hidden Markov models); hybrid stochastic/deterministic models;
stochastic analysis – convergence of random variables, stochastic calculus/stochastic differential equations (Ito-calculus, jump-diffusion processes);
statistical (Bayesian) inference and modelling – parametric and non-parameteric; estimation theory; posterior predictive distributions; prequential/statistical forecasting systems; asymptotic theory (Slutsky’s theorem, delta-method, portmanteau theorem, weak/strong laws); hypothesis testing; goodness-of-fit analyses; ROC techniques;

Machine Learning Approaches and Considerations: supervised learning (Regression, GLMs, SVMs, ANNs, CNNs, LSTMs); deep reinforcement learning (approximate Q-learning, temporal difference methods)
Mathematical Finance: mean-variance portfolio theory; CAPM, option-pricing, Wiener processes/Brownian motion, Martingales, risk-neutral measures, the Black-Scholes model;
Mathematical Analysis and Vector-Space Theory: real and complex analysis; metric spaces; infinite-dimensional vector-spaces – functional analysis (Banach/Hilbert spaces); finite-dimensional vector-spaces – linear/multilinear algebra, differential geometry/calculus on manifolds, tensor calculus; applications of fixed-point theorems; non-linear dynamical systems theory;
Mathematical/Statistical Optimisation: linear programming, convex optimisation, stochastic gradient methods, dynamic programming, simulated annealing, steepest descent;

Publications by category