Chris Watkins
Royal Holloway, University of London, Computer Science, Faculty Member
Research Interests:
Research Interests:
Research Interests: Psychology, Consumer Health Information, Medicine, The Internet, Information seeking, and 12 moreHumans, Physicians, Health information, Information Seeking Behavior, Great Britain, Harm, Internet privacy, Information Seeking, Internet, Data Accuracy, Psychology and Cognitive Sciences, and Medical and Health Sciences
Research Interests:
Research Interests:
Research Interests:
When a disease breaks out in a human population, changes in behavior in response to the outbreak can alter the progression of the infectious agent. In particular, people aware of a disease in their proximity can take measures to reduce... more
When a disease breaks out in a human population, changes in behavior in response to the outbreak can alter the progression of the infectious agent. In particular, people aware of a disease in their proximity can take measures to reduce their susceptibility. Even if no centralized information is provided about the presence of a disease, such awareness can arise through first-hand observation and word of mouth. To understand the effects this can have on the spread of a disease, we formulate and analyze a mathematical model for the spread of awareness in a host population, and then link this to an epidemiological model by having more informed hosts reduce their susceptibility. We find that, in a well-mixed population, this can result in a lower size of the outbreak, but does not affect the epidemic threshold. If, however, the behavioral response is treated as a local effect arising in the proximity of an outbreak, it can completely stop a disease from spreading, although only if the in...
Research Interests:
Research Interests: Algorithms, Computer Graphics, Data Mining, Computational Biology, Principal Component Analysis, and 15 moreBiology, Medicine, Transcriptome, Biological Sciences, Data Visualization, Saccharomyces cerevisiae, Environmental Sciences, Humans, Mice, Animals, Cluster Analysis, Nucleic Acids, Neurons, Graphical User Interface, and Chick embryo
In this report we describe how the Support Vector (SV) technique of solving linear operator equations can be applied to the problem of density estimation 4]. We present a new optimization procedure and set of kernels closely related to... more
In this report we describe how the Support Vector (SV) technique of solving linear operator equations can be applied to the problem of density estimation 4]. We present a new optimization procedure and set of kernels closely related to current SV techniques that guarantee the monotonicity of the approximation. This technique estimates densities with a mixture of bumps (Gaussian-like shapes), with the usual SV property that only some coe cients are non-zero. Both the width and the height of each bump is chosen adaptively, by ...
Research Interests:
Research Interests:
Research Interests:
Research Interests: Cognitive Science, Mathematics, Computer Science, Artificial Intelligence, Natural Language Processing, and 15 moreMachine Learning, Data Mining, Principal Component Analysis, Closure, Inference, Grammar Induction, Language Use, Kernel Methods, Grammatical Inference, Multidimensional Analysis, Language Theory, High Dimensionality, Kernel Pca, Feature Space, and Hyperplane
We describe methods of representing strings as real valued vectors or matrices; we show how to integrate two separate lines of enquiry: string kernels, developed in machine learning, and Parikh matrices [8], which have been studied... more
We describe methods of representing strings as real valued vectors or matrices; we show how to integrate two separate lines of enquiry: string kernels, developed in machine learning, and Parikh matrices [8], which have been studied intensively over the last few years as a powerful tool in the study of combinatorics over words. In the field of machine learning, there is widespread use of string kernels, which use analogous mappings into high dimensional feature spaces based on the occurrences of subwords or factors. In this ...
Research Interests:
Research Interests:
Support Vector Machines using ANOVA Decomposition Kernels (SVAD)[Vapng] are a way of imposing a structure on multi-dimensional kernels which are generated as the tensor product of one-dimensional kernels. This gives more accurate control... more
Support Vector Machines using ANOVA Decomposition Kernels (SVAD)[Vapng] are a way of imposing a structure on multi-dimensional kernels which are generated as the tensor product of one-dimensional kernels. This gives more accurate control over the capacity of the learning machine (VC-dimension). SVAD uses ideas from ANOVA decomposition methods and extends them to generate kernels which directly implement these ideas. SVAD is used with spline kernels and results show that SVAD performs better ...
Research Interests:
We show that evolutionary computation can be implemented as standard Markov-chain Monte-Carlo (MCMC) sampling. With some care, `genetic algorithms' can be constructed that are reversible Markov chains that satisfy detailed balance; it... more
We show that evolutionary computation can be implemented as standard Markov-chain Monte-Carlo (MCMC) sampling. With some care, `genetic algorithms' can be constructed that are reversible Markov chains that satisfy detailed balance; it follows that the stationary distribution of populations is a Gibbs distribution in a simple factorised form. For some standard and popular nonparametric probability models, we exhibit Gibbs-sampling procedures that are plausible genetic algorithms. At mutation-selection equilibrium, a population of genomes is analogous to a sample from a Bayesian posterior, and the genomes are analogous to latent variables. We suggest this is a general, tractable, and insightful formulation of evolutionary computation in terms of standard machine learning concepts and techniques. In addition, we show that evolutionary processes in which selection acts by differences in fecundity are not reversible, and also that it is not possible to construct reversible evolutiona...
We show that evolutionary computation can be implemented as standard Markov-chain Monte-Carlo (MCMC) sampling. With some care, `genetic algorithms' can be constructed that are reversible Markov chains that satisfy detailed balance; it... more
We show that evolutionary computation can be implemented as standard Markov-chain Monte-Carlo (MCMC) sampling. With some care, `genetic algorithms' can be constructed that are reversible Markov chains that satisfy detailed balance; it follows that the stationary distribution of populations is a Gibbs distribution in a simple factorised form. For some standard and popular nonparametric probability models, we exhibit Gibbs-sampling procedures that are plausible genetic algorithms. At mutation-selection equilibrium, a population of genomes is analogous to a sample from a Bayesian posterior, and the genomes are analogous to latent variables. We suggest this is a general, tractable, and insightful formulation of evolutionary computation in terms of standard machine learning concepts and techniques. In addition, we show that evolutionary processes in which selection acts by differences in fecundity are not reversible, and also that it is not possible to construct reversible evolutiona...
In addition, we show that evolutionary processes in which selection acts by differences in fecundity are not reversible, and also that it is not possible to construct reversible evolutionary models in which each child is produced by only... more
In addition, we show that evolutionary processes in which selection acts by differences in fecundity are not reversible, and also that it is not possible to construct reversible evolutionary models in which each child is produced by only two parents.
Research Interests:
Research Interests:
Research Interests:
In this report we show how the class of adaptive prediction methods that Sutton called" temporal difference," or TD, methods are related to the theory of squential... more
In this report we show how the class of adaptive prediction methods that Sutton called" temporal difference," or TD, methods are related to the theory of squential decision making. TDmethods have been used as" adaptive critics" in connectionist learning systems, and have beenproposed as models of animal learning in classical conditioning experiments. Here we relate TDmethods to decision tasks formulated in terms of a stochastic dynamical system whose behaviorunfolds over time under the influence of a decision maker's actions ...
Research Interests:
Abstract 1. Used a computer-based task to investigate the problem that young children have in constructing diagonals. The computer made it feasible to change how lines of different orientations had to be formed. It was predicted that if... more
Abstract 1. Used a computer-based task to investigate the problem that young children have in constructing diagonals. The computer made it feasible to change how lines of different orientations had to be formed. It was predicted that if diagonals are difficult to construct because of the operations required to conceptualize them, then changing how they had to be formed might make it possible for children to construct diagonals better than horizontals and verticals.
Animals and plants are intricately adapted to their environments, and much genomic information is needed to construct them. In each generation, genomic information is degraded by mutation, and it is also in some sense restored by... more
Animals and plants are intricately adapted to their environments, and much genomic information is needed to construct them. In each generation, genomic information is degraded by mutation, and it is also in some sense restored by selection. It is reasonable to ask how “information from selection” may be defined. Given a suitable definition, we may then ask whether there are limits on how much information can result from selection, and if so what determines the limits?
Abstract-Selective breeding is considered as a communication channel, in a novel way. The Shannon informational capacity of this channel is an upper limit on the amount of information that can be put into the genome by selection: this is... more
Abstract-Selective breeding is considered as a communication channel, in a novel way. The Shannon informational capacity of this channel is an upper limit on the amount of information that can be put into the genome by selection: this is a meaningful upper limit to the adaptive complexity of evolved organisms. We calculate the maximum adaptive complexity achievable for a given mutation rate for simple models of sexual and asexual reproduction.
United States Patent [w] Denker et al. US006064878A [ii] Patent Number: 6,064,878 [45] Date of Patent: May 16,2000 [54] METHOD FOR SEPARATELY PERMISSIONED COMMUNICATION [75] Inventors: John Stewart Denker, Leonardo, NJ; Christopher JCH... more
United States Patent [w] Denker et al. US006064878A [ii] Patent Number: 6,064,878 [45] Date of Patent: May 16,2000 [54] METHOD FOR SEPARATELY PERMISSIONED COMMUNICATION [75] Inventors: John Stewart Denker, Leonardo, NJ; Christopher JCH Watkins, London, United Kingdom [73] Assignee: AT&T Corp., New York, NY [21] Appl. No.: 08/735,849 [22] Filed: Oct. 23, 1996 [51] Int.
In this report we show how the class of adaptive prediction methods that Sutton called" temporal difference," or TD, methods are related to the theory of squential decision making. TDmethods have been used as" adaptive critics" in... more
In this report we show how the class of adaptive prediction methods that Sutton called" temporal difference," or TD, methods are related to the theory of squential decision making. TDmethods have been used as" adaptive critics" in connectionist learning systems, and have beenproposed as models of animal learning in classical conditioning experiments.
Abstract Many of the artificial neural network models so far proposedlearn'nonlinear functional mappings from training examples. For example, the multilayer perceptron of DE Rumelhart and JL McClelland (1984) and the CMAC of JA Albus... more
Abstract Many of the artificial neural network models so far proposedlearn'nonlinear functional mappings from training examples. For example, the multilayer perceptron of DE Rumelhart and JL McClelland (1984) and the CMAC of JA Albus (1981) are both devices of this type. Neural networks are not the only function approximation methods available, and there is interest in other methods; a number of types of function learning module have been reviewed by S. Omohundro (1987).
We describe methods of representing strings as real valued vectors or matrices; we show how to integrate two separate lines of enquiry: string kernels, developed in machine learning, and Parikh matrices [8], which have been studied... more
We describe methods of representing strings as real valued vectors or matrices; we show how to integrate two separate lines of enquiry: string kernels, developed in machine learning, and Parikh matrices [8], which have been studied intensively over the last few years as a powerful tool in the study of combinatorics over words. In the field of machine learning, there is widespread use of string kernels, which use analogous mappings into high dimensional feature spaces based on the occurrences of subwords or factors.
[57] ABSTRACT A method of automatic verification of personal identity is provided. The method includes the automatic generation of cue-response pairs and the allocation of such cue-response pairs to authorized people. Verification of the... more
[57] ABSTRACT A method of automatic verification of personal identity is provided. The method includes the automatic generation of cue-response pairs and the allocation of such cue-response pairs to authorized people. Verification of the identity of an applicant as an authorized person is achieved by presenting the cues from one or more cue-response pairs previously allocated to the authorized person to an applicant and verifying replies entered by the applicant by comparing the replies with the responses in the cue-response pairs.
Reinforcement learning is one of the means by which animals and artificial systems can learn to optimize their behaviour in the face of rewards and punishments. Reinforcement learning algorithms have been developed that are closely... more
Reinforcement learning is one of the means by which animals and artificial systems can learn to optimize their behaviour in the face of rewards and punishments. Reinforcement learning algorithms have been developed that are closely related to methods of dynamic programming, which is a general approach to optimal control. Reinforcement learning phenomena have been observed in psychological studies of animal behaviour, and in neurobiological investigations of neuromodulation and addiction.
Abstract When a disease breaks out in a human population, changes in behavior in response to the outbreak can alter the progression of the infectious agent. In particular, people aware of a disease in their proximity can take measures to... more
Abstract When a disease breaks out in a human population, changes in behavior in response to the outbreak can alter the progression of the infectious agent. In particular, people aware of a disease in their proximity can take measures to reduce their susceptibility. Even if no centralized information is provided about the presence of a disease, such awareness can arise through first-hand observation and word of mouth.
Abstract The complexity of gene expression data generated from microarrays and high-throughput sequencing make their analysis challenging. One goal of these analyses is to define sets of co-regulated genes and identify patterns of gene... more
Abstract The complexity of gene expression data generated from microarrays and high-throughput sequencing make their analysis challenging. One goal of these analyses is to define sets of co-regulated genes and identify patterns of gene expression. To date, however, there is a lack of easily implemented methods that allow an investigator to visualize and interact with the data in an intuitive and flexible manner.
Strings can be mapped into Hilbert spaces using feature maps such as the Parikh map. Languages can then be defined as the pre-image of hyperplanes in the feature space, rather than using grammars or automata. These are the planar... more
Strings can be mapped into Hilbert spaces using feature maps such as the Parikh map. Languages can then be defined as the pre-image of hyperplanes in the feature space, rather than using grammars or automata. These are the planar languages. In this paper we show that using techniques from kernel-based learning, we can represent and efficiently learn, from positive data alone, various linguistically interesting context-sensitive languages.
Resume Motivation: Genomes contain the information for constructing organisms. In some sense, this information is put in by selection, and it is degraded by mutation and genetic drift. It is natural to pose some basic questions of... more
Resume Motivation: Genomes contain the information for constructing organisms. In some sense, this information is put in by selection, and it is degraded by mutation and genetic drift. It is natural to pose some basic questions of principle. What is the maximum amount of information that can be maintained at mutation-selection equilibrium?
In this report we describe how the Support Vector (SV) technique of solving linear operator equations can be applied to the problem of density estimation 4]. We present a new optimization procedure and set of kernels closely related to... more
In this report we describe how the Support Vector (SV) technique of solving linear operator equations can be applied to the problem of density estimation 4]. We present a new optimization procedure and set of kernels closely related to current SV techniques that guarantee the monotonicity of the approximation. This technique estimates densities with a mixture of bumps (Gaussian-like shapes), with the usual SV property that only some coe cients are non-zero.
Abstract A typical problem in portfolio selection in stock markets is that it is not clear which of the many available strategies should be used. We apply a general algorithm of prediction with expert advice (the Aggregating Algorithm) to... more
Abstract A typical problem in portfolio selection in stock markets is that it is not clear which of the many available strategies should be used. We apply a general algorithm of prediction with expert advice (the Aggregating Algorithm) to two different idealizations of the stock market.
Abstract. Using string kernels, languages can be represented as hyperplanes in a high dimensional feature space. We present a new family of grammatical inference algorithms based on this idea. We demonstrate that some mildly context... more
Abstract. Using string kernels, languages can be represented as hyperplanes in a high dimensional feature space. We present a new family of grammatical inference algorithms based on this idea. We demonstrate that some mildly context sensitive languages can be represented in this way and it is possible to efficiently learn these using kernel PCA. We present some experiments demonstrating the effectiveness of this approach on some standard examples of context sensitive languages using small synthetic data sets.
Abstract Support Vector Machines using ANOVA Decomposition Kernels (SVAD)[Vapng] are a way of imposing a structure on multi-dimensional kernels which are generated as the tensor product of one-dimensional kernels. This gives more accurate... more
Abstract Support Vector Machines using ANOVA Decomposition Kernels (SVAD)[Vapng] are a way of imposing a structure on multi-dimensional kernels which are generated as the tensor product of one-dimensional kernels. This gives more accurate control over the capacity of the learning machine (VC-dimension). SVAD uses ideas from ANOVA decomposition methods and extends them to generate kernels which directly implement these ideas.
Q-learning (Watkins, 1989) is a simple way for agents to learn how to act optimally in controlled Markovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by... more
Q-learning (Watkins, 1989) is a simple way for agents to learn how to act optimally in controlled Markovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by successively improving its evaluations of the quality of particular actions at particular states.
Research Interests:
The solution of binary classi cation problems using the Support Vector (SV) method is well developed. Multi-class pattern recognition problems (where one has k> 2... more
The solution of binary classi cation problems using the Support Vector (SV) method is well developed. Multi-class pattern recognition problems (where one has k> 2 classes) are typically solved using voting scheme methods based on combining many binary classi cation decision functions 5, 2]. We propose two extensions to the SV method of pattern recognition to solve k-class problems in one (formal) step. We describe two di erent methods of SV k-class classi cation which do not use a combination of binary classi cation rules. ...