Date | Venue | Speaker | Title and Abstract |
---|---|---|---|
16 Oct | RCH/204 |
Susan Stepney
CS, University of York |
Sub-symbolic artificial chemistries
Artificial Chemistries (AChems) are often suggested as a
suitable basis for the rich behaviour and dynamics needed to underpin
Artificial Life systems. However, traditional AChems have their
limitations, often requiring ad hoc programming extensions to
encompass new behaviours. Here I describe our work on sub-symbolic
AChems, which overcome some of these limitations, potentially
providing a uniform approach to developing systems exhibiting
unbounded novelty in silico.
|
30 Oct |
Biology BB002
2:15-3:15pm |
Chris McEwan
Microsoft Research |
Expressing and understanding cellular heterogeneity
It has been said that biology is the science of heterogeneous systems. The apparent obviousness of this statement belies the fact that we do not really have satisfactory methods for formulating and answering questions regarding the functional role, or lack thereof, of experimentally observable heterogeneity. The lab biologists' methods of analysis are often crude and largely descriptive. The mathematical biologists' methods are sophisticated but homogenising, making formulating questions of heterogeneity difficult. The computational biologists' methods can express heterogeneity, but the resulting "answers" tend to be scientifically weaker. In this talk I will present research aimed at finessing these issues. I will first show how mathematical and computational techniques can be applied in tandem such that a spectrum of heterogeneous conditions can be traversed and the relative strengths of different techniques exploited. A simple, method-agnostic graphical language for describing heterogeneous systems provides a high-level facade to the underlying techniques and relieves the burden of developing and maintaining concurrent encodings of "the same" biological phenomenon. I will then present preliminary research using the same approach to directly organise experimental data and hypotheses, with the intention of moving towards more objective model-driven data-analysis and data-driven synthesis of dynamical models. |
13 Nov |
RCH/204
2:15-3:15pm |
James Mistry
BT |
Heuristic Zero-Day Shellcode Detection
The first stage of many malware attacks involves the transmission of shellcode, fragments of machine code lacking any metadata. The ability to accurately detect shellcode on the network would provide a way to identify and block zero-day malware before it ever reaches its target. Unfortunately, differentiating shellcode from benign network traffic is not trivial, particularly given the complex obfuscation techniques employed by malware authors. In this talk, I will discuss a research project underway at BT aimed at developing a system that can heuristically identify shellcode at line rate using various machine learning approaches. |
11 Dec |
RCH/204
2:15-3:15pm |
Simon Harding
University of Bristol |
CGP-IP: Cartesian Genetic Programming for Image Processing
Cartesian Genetic Programming (CGP) is a powerful, multipurpose machine learning technique that has been used in many different domains. In this talk, I will discuss the use of CGP for object detection, segmentation and image processing. In CGP-IP, computer programs are evolved, that take images and apply a sequence of operations. Leveraging domain knowledge from the computer vision community, these programs are built from well known and well understood operations. We will see examples in medical imaging, terrain classification, robotics and industrial applications. I will show that the approach is fast, scalable and robust. In addition, we will see that it can generate human readable programs that can be further customised and tuned. |
| |||
15 Jan |
Physics PT006
2:15-3:15pm |
Hywel Williams
University of Exeter |
Diversity, evolvability, and the 'adjacent possible' in host-parasite coevolution
Coevolution between hosts and their parasites can often lead to diversification; hosts diversify to escape their parasites, while parasites diversify to track their evolving hosts. Since the host traits that coevolve are also subject to other (non-parasite) selection pressures, such diversification will affect host evolution beyond its role in reducing infection. Here I explore the relationship between coevolved diversity and evolvability, using a mathematical model of bacteria coevolving with bacteriophage. Phage-driven diversification allows bacteria to evolve across valleys in the fitness landscape and reach peaks that would otherwise be unachievable - thus the phage actually benefit their host population by increasing evolvability. There are interesting parallels with ideas of robustness and evolvability in molecular evolution, which are discussed together with the related concept of the `adjacent possible'. This work has potential implications for phage therapy, microbial ecology, and evolutionary computation. |
29 Jan |
CSE/083
2:15-3:15pm |
Hector Zenil
University of Sheffield |
Measures of Algorithmic Information Theory for Systems and Synthetic Biology
I will show that tools based on algorithmic information theory can help characterise the behaviour of a system by means of its information content and algorithmic complexity. I will survey some proposed metrics that I will claim are suitable to typify qualitative properties of the behaviour of biological entities, allowing a numerical mapping between external stimuli and output response. My approach intends to help shorten tasks such as the iterative process between simulation and testing in the lab in the practice of systems or synthetic biology. A case study with abstract symbolic systems such as cellular automata and with Porphyrin molecules will be briefly discussed. |
12 Feb |
Physics PT006
2:15-3:15pm |
||
26 Feb |
RCH/204
2:15-3:15pm |
Tim Taylor |
Exploring the Concept of Open-Ended Evolution
The term "open-ended evolution" has been widely used in the Artificial Life community to refer to the kind of long-term evolutionary dynamics observed in the biosphere. The goal of many Artificial Life researchers is to achieve similar dynamics within a virtual world. The term implies an on-going production of new forms, and an increase in the complexity and diversity of species; a system capable of open-ended evolution could spontaneously generate rich ecosystems of complex organisms. The term is often used to distinguish such work (a) from the field of evolutionary computation, where evolution is employed as an optimisation technique to solve a specific problem defined by an explicit fitness function, and (b) from systems which lack an explicit fitness function, and yet still quickly reach a quasi-stable state beyond which no further adaptive changes are observed. However, despite the centrality of the concept to such work, the term "open-ended evolution" remains rather vague and lacking a concise, useful, and widely agreed definition. In this talk I will try to disentangle the concept and identify the various issues that it entails. I hope that improved clarity on these issues will suggest directions for progress in future research. |
12 Mar |
Physics PT006
2:15-3:15pm |
Jamie Luo
University of Exeter |
Studying Evolution in Boolean Networks and the Dynamics of Bacterial Persistence
This talk is comprised of three parts. The first two utilise Boolean Networks as an abstract model to investigate questions of evolution and structure in genetic networks.Part one will study the limits of neutral evolutionary processes without the need for evolutionary experiments, looking in detail on how different definitions of biological function influence the resulting neutral network/landscape. The second part of the talk will then employ evolutinary algorithms geared to design sensitive network topologies. A definition of dynamical sensitivity is provided which implies that certain types of network topolgy are more geared to be sensitive to perturbations than others. This leads to hypotheses about a possible sensitivity slope relating to different levels of cell organisation. The final part will look into the intriguing phenomenon of bacterial persistence and phenotypic models of the phenomenon. |
| |||
30 Apr |
RCH/204
2:15-3:15pm |
Gaetana Spedalieri
CS, University of York |
Quantum hypothesis testing
In this short talk (PhD literature review seminar) we give a rapid overview of the current status in the field of quantum hypothesis testing, discussing the basic problems of quantum state discrimination and quantum channel discrimination. The main mathematical tools will be reviewed, including the Helstrom bound and the recently-derived quantum Chernoff bound. |
14 May |
Physics PT005
2:15-3:15pm |
POSTPONED | |
28 May |
RCH/204
2:15-3:15pm |
John Woodward
Stirling University |
10,000 Algorithms: The (Semi-)automatic Design of (Heuristic) Algorithms.
Heuristics aim to provide practical solutions to computationally intractable problems. A researcher traditionally proposes a novel algorithm and demonstrates its comparatively better performance on a set of benchmark problem instances. It will be argued that this is a theoretically flawed approach, and instead of manually designing single algorithms we can automatically generate vast numbers of algorithms for the problem class at hand. Three examples of automated algorithm development will be given; these domains include bin packing, components of a genetic algorithm and probability distributions. In all cases, the automatically-designed algorithm statistically outperforms the human designed counterpart. |
11 Jun |
Biology BB002
2:15-3:15pm |
Andrew Phillips
Microsoft Research |
Software for Programming Cells
Cells are the building blocks of life. If we could program living cells as effectively as we program digital computers we could make breakthroughs in medical treatment, sustainable agriculture and clean energy, while also better understanding of how living systems compute. In spite of this potential there are still many challenges to overcome. Programming cells is highly complex and error-prone, and we are at a point where powerful computer software is needed to accelerate further progress. This talk presents ongoing work to develop computer software for programming cells at three levels: molecular circuits, genetic devices and cell colonies. We present software for programming molecular circuits made of DNA, and for characterising genetic parts that can be combined into devices for programming cell function. Finally, we present software for simulating cell biofilms using 3D biophysical methods, which can be used to predict the effect of cell shape on colony morphology. Just as software for programming digital computers heralded a new era of technology, software for programming cells could enable new industries in biotechnology. |
25 Jun |
RCH/204
2:15-3:15pm |
Ed Clark
University of York |
DNA based replication in artificial chemistries: Implementing von Neumann machines
In 1966 von Neumann described self replicating automata, detailing in principle how machines could replicate machines by having a coded tape or genome. Artificial chemistries including Avida, Tierra and Stringmol were originally written in such a way that machines could copy themseslves by direct inspection, often referred to as RNA world, where there is no genotype-phenotype divide. Recently researchers have been attempting to implement DNA based replication in these artificial chemistries. We compare our implementation of DNA based replication in Stringmol with recent implementations in Avida and Tierra. We examine the evolutionary behaviour and outline why our Stringmol implementation is able to evolve new general constructors and copiers, while the implementations in Tierra and Avida revert to pathogenic constructors and self copiers respectively when they are allowed to evolve. |