Emma Hart (Edinburgh Napier): Lifelong Learning in Optimisation (School Seminar)

Abstract:

The previous two decades have seen significant advances in optimisation techniques that are able to quickly find optimal or near-optimal solutions to problem instances in many combinatorial optimisation domains. Despite many successful applications of both these approaches, some common weaknesses exist in that if the nature of the problems to be solved changes over time, then algorithms needs to be at best periodically re-tuned. In the worst case, new algorithms may need to be periodically redeveloped. Furthermore, many approaches are inefficient, starting from a clean slate every time a problem is solved, therefore failing to exploit previously learned knowledge.

In contrast, in the field of machine-learning, a number of recent proposals suggest that learning algorithms should exhibit life-long learning, retaining knowledge and using it to improve learning in the future. I propose that optimisation algorithms should follow the same approach – looking to nature, we observe that the natural immune system exhibits many properties of a life-long learning system that could be exploited computationally in an optimisation framework. I will give a brief overview of the immune system, focusing on highlighting its relevant computational properties and then show how it can be used to construct a lifelong learning optimisation system. The system exploits genetic programming to continually evolve new optimisation algorithms, which form a continually adapting ensemble of optimisers. The system is shown to adapt to new problems, exhibit memory, and produce efficient and effective solutions when tested in both the bin-packing and scheduling domains.

Speaker Bio:

Emma Hart is a Professor in Natural Computation at Edinburgh Napier University in Scotland, where she also directs the Centre for Algorithms, Visualisation and Evolving Systems. Prior to that, she received a degree in Chemistry from the University of Oxford and a PhD in Artificial Immune Systems for Optimisation and Learning from the University of Edinburgh.

Her research focuses on developing novel bio-inspired techniques for solving a range of real-world optimisation and classification problems, particularly through the application of hyper-heuristic approaches and genetic programming. Her recent research explores optimisation techniques which are capable of continuously improving through experience, as well as ensemble approaches to optimisation for solving large classes of problems.

She is Editor-in-Chief of the journal Evolutionary Computation (MIT Press), ) and an elected member of the ACM SIGEVO Executive Committee. She also edits SIGEVOlution, the magazine of SIGEVO. She was General Chair of PPSN 2016, and regularly acts as Track Chair at GECCO . She has recently given keynotes at EURO 2016, Poland, and IJCCI (Maderia, 2017) on Lifelong Optimisation.

Her work is funded by both national funding agencies (EPSRC) and the European, where has recently led projects in Fundamentals of Collective Adaptive System (FOCAS) and Self-Aware systems (AWARE). She has worked with a range of real-world clients including from the Forestry Industry, Logistics and Personnel Scheduling.

Event details

  • When: 14th November 2017 14:00 - 15:00
  • Where: Cole 1.33a
  • Series: School Seminar Series
  • Format: Seminar

Jessie Kennedy (Edinburgh Napier): Visualization and Taxonomy (School Seminar)

Abstract:

This talk will consider the relationship between visualization and taxonomy from two perspectives. Firstly, how visualization can aid understanding the process of taxonomy, specifically biological taxonomy and the visualization challenges this poses. Secondly, the role of taxonomy in understanding and making sense of the growing field of visualization will be discussed and the challenges facing the visualization community in making this process more rigorous will be considered.

Speaker Bio:

Jessie joined Edinburgh Napier University in 1986 as a lecturer, was promoted to Senior Lecturer, Reader, and then Professor in 2000 Thereafter she held the post of Director of the Institute for Informatics and Digital Innovation from 2010-14 and is currently Dean of Research and Innovation for the University.

Jessie has published widely, with over 100 peer-reviewed publications and over £2 million in research funding from a range of bodies, including EPSRC, BBSRC, National Science Foundation, and KTP, and has had 13 PhD students complete. She has been programme chair, programme committee member and organiser of many international conferences, a reviewer and panel member for many national and international computer science funding bodies, and became a Member of EPSRC Peer Review College in 1996 and a Fellow of the British Computer Society.

Jessie has a long-standing record of contribution to inter-disciplinary research, working to further biological research through the application of novel computing technology.

Her research in the areas of user interfaces to databases and data visualisation in biology contributed to the establishment of the field of biological visualisation. She hosted the first biological visualisation workshop at the Royal Society of Edinburgh in 2008, was an invited speaker at a BBSRC workshop on Challenges in Biological Visualisation in 2010, was a founding member of the International Symposium in Biological Visualisation – being Programme Chair in 2011, General Chair in 2012 and 2013 – and steering committee member since 2014.

She has been keynote speaker at related international conferences and workshops, such as VIZBI, the International Visualisation conference and BioIT World, and is currently leading a BBSRC network on biological visualisation.

Her research in collaboration with taxonomists at the Royal Botanic Gardens, Edinburgh, produced a data model for representing differing taxonomic opinions in Linnaean classification. This work led to collaboration on a large USA-funded project with ecologists from six US universities and resulted in a data standard for the exchange biodiversity data that has been adopted by major global taxonomic and biodiversity organisations.

Event details

  • When: 7th November 2017 14:00 - 15:00
  • Where: Cole 1.33a
  • Series: School Seminar Series
  • Format: Seminar

Barnaby Martin (Durham): The Complexity of Quantified Constraints (School Seminar)

Abstract:

We elaborate the complexity of the Quantified Constraint Satisfaction Problem, QCSP(A), where A is a finite idempotent algebra. Such a problem is either in NP or is co-NP-hard, and the borderline is given precisely according to whether A enjoys the polynomially-generated powers (PGP) property. This reduces the complexity classification problem for QCSPs to that of CSPs, modulo that co-NP-hard cases might have complexity rising up to Pspace-complete. Our result requires infinite languages, but in this realm represents the proof of a slightly weaker form of a conjecture for QCSP complexitymade by Hubie Chen in 2012. The result relies heavily on the algebraic dichotomy between PGP and exponentially-generated powers (EGP), proved by Dmitriy Zhuk in 2015, married carefully to previous work of Chen.

Event details

  • When: 24th October 2017 14:00 - 15:00
  • Where: Cole 1.33a
  • Series: School Seminar Series
  • Format: Seminar

Maja Popović (Humboldt-Universität zu Berlin): (Dis)similarity Metrics for Texts (School Seminar)

Abstract:
Natural language processing (NLP) is a multidisciplinary field closely related to linguistics, machine learning and artificial intelligence. It comprises a number of different subfields dealing with different kinds of analysis and/or generation of natural language texts. All these methods and approaches need some kind of evaluation, i.e. comparison between the obtained result with a given gold standard. For tasks dealing with text generation (such as speech recognition or machine translation), a comparison between two texts has to be carried out. This is usually done either by counting matched words or word sequences (which produces a similarity score) or by calculating edit distance, i.e. a number of operations needed to transform the generated word sequence into a desired word sequence (which produces a “dissimilarity” score called “error rate”). The talk will give an overview of advantages, disadvantages and challenges related to this type of metrics mainly concentrating on machine translation (MT) but also relating to some other NLP tasks.

Speaker bio:
Maja Popović graduated at the Faculty of Electrical Engineering, University of Belgrade and continued her studies at the RWTH Aachen, Germany, where she obtained her PhD with the thesis “Machine Translation: Statistical Approach with Additional Linguistic Knowledge”. After that, she continued her research at the DFKI Institute and thereafter at the Humboldt University of Berlin, mainly related to various approaches for evaluation of machine translation. She has developed two open source evaluation tools, (i) Hjerson, a tool for automatic translation error classification, and (ii) chrF, an automatic metric for machine translation evaluation based on character sequence matching.

Event details

  • When: 29th September 2017 13:00 - 14:00
  • Where: Cole 1.33a
  • Series: School Seminar Series
  • Format: Seminar

Semantics for probabilistic programming, Dr Chris Heunen

Semantics for probabilistic programming, Dr Chris Heunen

03.10.17, 1pm, Room JCB 1.33B

Abstract: Statistical models in e.g. machine learning are traditionally
expressed in some sort of flow charts. Writing sophisticated models
succintly is much easier in a fully fledged programming language. The
programmer can then rely on generic inference algorithms instead of
having to craft one for each model. Several such higher-order functional
probabilistic programming languages exist, but their semantics, and
hence correctness, are not clear. The problem is that the standard
semantics of probability theory, given by measurable spaces, does not
support function types. I will describe how to get around this.

Seminar: Propagation and Reification: SAT and SMT in Prolog (continued)

Jacob Howe, City University, London

Abstract: This talk will recap how a watched literal DPLL based SAT solver can be succinctly coded in 20 lines of Prolog. The focus of the talk will be the extension of this solver to an SMT solver which will be discussed with a particular focus on the case where the theory is that of rational-tree constraints, and its application in a reverse engineering problem.
[Note change of time from that previously advertised]

Event details

  • When: 23rd June 2017 13:00 - 14:00
  • Where: Cole 1.33a
  • Series: AI Seminar Series
  • Format: Seminar

SRG Seminar: Evaluation Techniques for Detection Model Performance in Anomaly Network Intrusion Detection System by Amjad Al Tobi

Everyday advancements in technology brings with it novel challenges and threats. Such advancement imposes greater risks than ever on systems and services, including individual privacy information. Relying on intrusion specialists to come up with new signatures to detect different types of new attacks, does not seem to scale with excessive traffic growth. Therefore, anomaly-based detection provides a promising solution for this problem area.

Anomaly-based IDS applies machine learning, data mining and/or artificial intelligence along with many other methods to solve this problem. Currently, these solutions seem not to be tractable for real production environments due to the high false alarms rate. This might be a result of such systems not being able to determine the point at which an update is required. It is not clear how detection models will behave over time, when traffic behaviour has changed since the last time the model was re-generated.
Continue reading

Event details

  • When: 1st June 2017 13:00 - 14:00
  • Where: Cole 1.33b
  • Series: Systems Seminars Series
  • Format: Seminar

SRG Seminar: New Network Functionality using ILNPv6 and DNS by Khawar Shehzad

This research deals with the introduction of a new network functionality based on Identifier-Locator Network Protocol version 6 (ILNPv6), and Domain Name System (DNS). The chosen area of concern is security and specifically mitigation of Distributed Denial of Service (DDoS). The functionality proposed and tested deals with the issues of vulnerability testing, probing, and scanning which directly lead to a successful DDoS attack. The solutions presented can be used as a reactive measure to these security issues. The DDoS is chosen because in recent years DDoS have become the most common and hard to defend attacks. These attacks are on the availability of system/site. There are multiple solutions in the literature but no one solution is based on ILNPv6, and are complex in nature. Similarly, the solutions in literature either require modification in the providers’ networks or they are complex if they are only site-based solutions. Most of these solutions are based on IPv6 protocol and they do not use the concept of naming, as proposed by ILNPv6.

The prime objectives of this research are:

  • to defend against DDoS attacks with the use of naming and DNS
  • to increase the attacker’s effort, reduce vulnerability testing, and random probing by attackers
  • to practically demonstrate the effectiveness of the ILNPv6-based solution for security

Event details

  • When: 18th May 2017 13:00 - 14:00
  • Where: Cole 1.33b
  • Series: Systems Seminars Series
  • Format: Seminar

SRG Seminar: Investigation of Virtual Network Isolation Security in Cloud Computing Using Penetration Testing by Haifa Al Nasseri

Software Defined Networking (SDN) or Virtual Networks (VNs) are required for cloud tenants to leverage demands. However, multi-tenancy can be compromised without proper isolation. Much research has been conducted into VN Isolation; many researchers are not tackling security aspects or checking if their isolation evaluation is complete. Therefore, data leakage is a major security concern in the cloud in general.

This research uses an OpenStack VN and OpenStack Tenant Network to test multi-tenancy features. We are evaluating the relationship between isolation methods used in cloud VN and the amount of data being leaked through using penetration tests. These tests will be used to identify the vulnerabilities causing cloud VN data leakage and to investigate how the vulnerabilities, and the leaked data, can compromise the tenant Virtual Networks.

Event details

  • When: 4th May 2017 13:00 - 14:00
  • Where: Cole 1.33b
  • Series: Systems Seminars Series
  • Format: Talk