Articulated Pose Estimation in a Learned Smooth Space of Feasible Solutions

A learning based framework is proposed for estimating human body pose from a single image. Given a differentiable function that maps from pose space to image feature space, the goal is to invert the process: estimate the pose given only image features. The inversion is an ill-posed problem as the inverse mapping is a one to many process. Hence multiple solutions exist, and it is desirable to restrict the solution space to a smaller subset of feasible solutions. For example, not all human body poses are feasible due to anthropometric constraints. Since the space of feasible solutions may not admit a closed form description, the proposed framework seeks to exploit machine learning techniques to learn an approximation that is smoothly parameterized over such a space. One such technique is Gaussian Process Latent Variable Modelling. Scaled conjugate gradient is then used find the best matching pose in the space of feasible solutions when given an input image. The formulation allows easy incorporation of various constraints, e.g. temporal consistency and anthropometric constraints. The performance of the proposed approach is evaluated in the task of upper-body pose estimation from silhouettes and compared with the Specialized Mapping Architecture. The estimation accuracy of the Specialized Mapping Architecture is at least one standard deviation worse than the proposed approach in the experiments with synthetic data. In experiments with real video of humans performing gestures, the proposed approach produces qualitatively better estimation results.

Finding optimal alternatives based on efficient comparative preference inference

Choosing the right or the best option is often a demanding and challenging task for the user (e.g., a customer in an online retailer) when there are many available alternatives. In fact, the user rarely knows which offering will provide the highest value. To reduce the complexity of the choice process, automated recommender systems generate personalized recommendations. These recommendations take into account the preferences collected from the user in an explicit (e.g., letting users express their opinion about items) or implicit (e.g., studying some behavioral features) way. Such systems are widespread; research indicates that they increase the customers' satisfaction and lead to higher sales. Preference handling is one of the core issues in the design of every recommender system. This kind of system often aims at guiding users in a personalized way to interesting or useful options in a large space of possible options. Therefore, it is important for them to catch and model the user's preferences as accurately as possible. In this thesis, we develop a comparative preference-based user model to represent the user's preferences in conversational recommender systems. This type of user model allows the recommender system to capture several preference nuances from the user's feedback. We show that, when applied to conversational recommender systems, the comparative preference-based model is able to guide the user towards the best option while the system is interacting with her. We empirically test and validate the suitability and the practical computational aspects of the comparative preference-based user model and the related preference relations by comparing them to a sum of weights-based user model and the related preference relations. Product configuration, scheduling a meeting and the construction of autonomous agents are among several artificial intelligence tasks that involve a process of constrained optimization, that is, optimization of behavior or options subject to given constraints with regards to a set of preferences. When solving a constrained optimization problem, pruning techniques, such as the branch and bound technique, point at directing the search towards the best assignments, thus allowing the bounding functions to prune more branches in the search tree. Several constrained optimization problems may exhibit dominance relations. These dominance relations can be particularly useful in constrained optimization problems as they can instigate new ways (rules) of pruning non optimal solutions. Such pruning methods can achieve dramatic reductions in the search space while looking for optimal solutions. A number of constrained optimization problems can model the user's preferences using the comparative preferences. In this thesis, we develop a set of pruning rules used in the branch and bound technique to efficiently solve this kind of optimization problem. More specifically, we show how to generate newly defined pruning rules from a dominance algorithm that refers to a set of comparative preferences. These rules include pruning approaches (and combinations of them) which can drastically prune the search space. They mainly reduce the number of (expensive) pairwise comparisons performed during the search while guiding constrained optimization algorithms to find optimal solutions. Our experimental results show that the pruning rules that we have developed and their different combinations have varying impact on the performance of the branch and bound technique.

Voltage properties of optically injected long wavelength VCSELs: Theory and experiment

Future high speed communications networks will transmit data predominantly over optical fibres. As consumer and enterprise computing will remain the domain of electronics, the electro-optical conversion will get pushed further downstream towards the end user. Consequently, efficient tools are needed for this conversion and due to many potential advantages, including low cost and high output powers, long wavelength Vertical Cavity Surface Emitting Lasers (VCSELs) are a viable option. Drawbacks, such as broader linewidths than competing options, can be mitigated through the use of additional techniques such as Optical Injection Locking (OIL) which can require significant expertise and expensive equipment. This thesis addresses these issues by removing some of the experimental barriers to achieving performance increases via remote OIL. Firstly, numerical simulations of the phase and the photon and carrier numbers of an OIL semiconductor laser allowed the classification of the stable locking phase limits into three distinct groups. The frequency detuning of constant phase values (ø) was considered, in particular ø = 0 where the modulation response parameters were shown to be independent of the linewidth enhancement factor, α. A new method to estimate α and the coupling rate in a single experiment was formulated. Secondly, a novel technique to remotely determine the locked state of a VCSEL based on voltage variations of 2mV−30mV during detuned injection has been developed which can identify oscillatory and locked states. 2D & 3D maps of voltage, optical and electrical spectra illustrate corresponding behaviours. Finally, the use of directly modulated VCSELs as light sources for passive optical networks was investigated by successful transmission of data at 10 Gbit/s over 40km of single mode fibre (SMF) using cost effective electronic dispersion compensation to mitigate errors due to wavelength chirp. A widely tuneable MEMS-VCSEL was established as a good candidate for an externally modulated colourless source after a record error free transmission at 10 Gbit/s over 50km of SMF across a 30nm single mode tuning range. The ability to remotely set the emission wavelength using the novel methods developed in this thesis was demonstrated.

Two closely coupled lasers, dynamics & applications

The dynamics of two mutually coupled identical single-mode semi-conductor lasers are theoretically investigated. For small separation and large coupling between the lasers, symmetry-broken one-colour states are shown to be stable. In this case the light output of the lasers have significantly different intensities whilst at the same time the lasers are locked to a single common frequency. For intermediate coupling we observe stable two-colour states, where both single-mode lasers lase simultaneously at two optical frequencies which are separated by up to 150 GHz. For low coupling but possibly large separation, the frequency of the relaxation oscillations of the freerunning lasers defines the dynamics. Chaotic and quasi-periodic states are identified and shown to be stable. For weak coupling undamped relaxation oscillations dominate where each laser is locked to three or more odd number of colours spaced by the relaxation oscillation frequency. It is shown that the instabilities that lead to these states are directly connected to the two colour mechanism where the change in the number of optical colours due to a change in the plane of oscillation. At initial coupling, in-phase and anti-phase one colour states are shown to emerge from “on” uncoupled lasers using a perturbation method. Similarly symmetry-broken one-colour states come from considering one free-running laser initially “on” and the other laser initially “off”. The mechanism that leads to a bi-stability between in-phase and anti-phase one-colour states is understood. Due to an equivariant phase space symmetry of being able to exchange the identical lasers, a symmetric and symmetry-broken variant of all states mentioned above exists and is shown to be stable. Using a five dimensional model we identify the bifurcation structure which is responsible for the appearance of symmetric and symmetry-broken one-colour, symmetric and symmetry-broken two-colour, symmetric and symmetry-broken undamped relaxation oscillations, symmetric and symmetry-broken quasi-periodic, and symmetric and symmetry-broken chaotic states. As symmetry-broken states always exist in pairs, they naturally give rise to bi-stability. Several of these states show multistabilities between symmetric and symmetry-broken variants and among states. Three memory elements on the basis of bi-stabilities in one and two colour states for two coupled single-mode lasers are proposed. The switching performance of selected designs of optical memory elements is studied numerically.

Practice-based personas: integrated user and practice models

Practices are routinised behaviours with social and material components and complex relationships over space and time. Practice-based design goes beyond interaction design to consider how these components and their relationships impact on the formation and enactment of a practice, where technology is just one part of the practice. Though situated user-centred design methods such as participatory design are employed for the design of practice, demand exists for additional methods and tools in this area. This paper introduces practice-based personas as an extension of the persona approach popular in interaction design, and demonstrates how a set of practice-based personas was developed for a given domain – academic practice. The three practice-based personas developed here are linked to a catalogue of forty practices, offering designers both a user perspective and a practice perspective when designing for the domain.

Contrasting the efficiency of radiation belt losses caused by ducted and nonducted whistler-mode waves from ground-based transmitters

It has long been recognized that whistler-mode waves can be trapped in plasmaspheric whistler ducts which guide the waves. For nonguided cases these waves are said to be "nonducted", which is dominant for L < 1.6. Wave-particle interactions are affected by the wave being ducted or nonducted. In the field-aligned ducted case, first-order cyclotron resonance is dominant, whereas nonducted interactions open up a much wider range of energies through equatorial and off-equatorial resonance. There is conflicting information as to whether the most significant particle loss processes are driven by ducted or nonducted waves. In this study we use loss cone observations from the DEMETER and POES low-altitude satellites to focus on electron losses driven by powerful VLF communications transmitters. Both satellites confirm that there are well-defined enhancements in the flux of electrons in the drift loss cone due to ducted transmissions from the powerful transmitter with call sign NWC. Typically, ∼80% of DEMETER nighttime orbits to the east of NWC show electron flux enhancements in the drift loss cone, spanning a L range consistent with first-order cyclotron theory, and inconsistent with nonducted resonances. In contrast, ∼1% or less of nonducted transmissions originate from NPM-generated electron flux enhancements. While the waves originating from these two transmitters have been predicted to lead to similar levels of pitch angle scattering, we find that the enhancements from NPM are at least 50 times smaller than those from NWC. This suggests that lower-latitude, nonducted VLF waves are much less effective in driving radiation belt pitch angle scattering. Copyright 2010 by the American Geophysical Union.

Facet-embedded thin-film III-V edge-emitting lasers integrated with SU-8 waveguides on silicon.

A thin-film InGaAs/GaAs edge-emitting single-quantum-well laser has been integrated with a tapered multimode SU-8 waveguide onto an Si substrate. The SU-8 waveguide is passively aligned to the laser using mask-based photolithography, mimicking electrical interconnection in Si complementary metal-oxide semiconductor, and overlaps one facet of the thin-film laser for coupling power from the laser to the waveguide. Injected threshold current densities of 260A/cm(2) are measured with the reduced reflectivity of the embedded laser facet while improving single mode coupling efficiency, which is theoretically simulated to be 77%.

Adaptive Mixture Modelling Metropolis Methods for Bayesian Analysis of Non-linear State-Space Models.

We describe a strategy for Markov chain Monte Carlo analysis of non-linear, non-Gaussian state-space models involving batch analysis for inference on dynamic, latent state variables and fixed model parameters. The key innovation is a Metropolis-Hastings method for the time series of state variables based on sequential approximation of filtering and smoothing densities using normal mixtures. These mixtures are propagated through the non-linearities using an accurate, local mixture approximation method, and we use a regenerating procedure to deal with potential degeneracy of mixture components. This provides accurate, direct approximations to sequential filtering and retrospective smoothing distributions, and hence a useful construction of global Metropolis proposal distributions for simulation of posteriors for the set of states. This analysis is embedded within a Gibbs sampler to include uncertain fixed parameters. We give an example motivated by an application in systems biology. Supplemental materials provide an example based on a stochastic volatility model as well as MATLAB code.

High-fidelity, broadband stimulated-Brillouin-scattering-based slow light using fast noise modulation.

We demonstrate a 5-GHz-broadband tunable slow-light device based on stimulated Brillouin scattering in a standard highly-nonlinear optical fiber pumped by a noise-current-modulated laser beam. The noisemodulation waveform uses an optimized pseudo-random distribution of the laser drive voltage to obtain an optimal flat-topped gain profile, which minimizes the pulse distortion and maximizes pulse delay for a given pump power. In comparison with a previous slow-modulation method, eye-diagram and signal-to-noise ratio (SNR) analysis show that this broadband slow-light technique significantly increases the fidelity of a delayed data sequence, while maintaining the delay performance. A fractional delay of 0.81 with a SNR of 5.2 is achieved at the pump power of 350 mW using a 2-km-long highly nonlinear fiber with the fast noise-modulation method, demonstrating a 50% increase in eye-opening and a 36% increase in SNR in the comparison.

POCS-based reconstruction of multiplexed sensitivity encoded MRI (POCSMUSE): A general algorithm for reducing motion-related artifacts.

PURPOSE: A projection onto convex sets reconstruction of multiplexed sensitivity encoded MRI (POCSMUSE) is developed to reduce motion-related artifacts, including respiration artifacts in abdominal imaging and aliasing artifacts in interleaved diffusion-weighted imaging. THEORY: Images with reduced artifacts are reconstructed with an iterative projection onto convex sets (POCS) procedure that uses the coil sensitivity profile as a constraint. This method can be applied to data obtained with different pulse sequences and k-space trajectories. In addition, various constraints can be incorporated to stabilize the reconstruction of ill-conditioned matrices. METHODS: The POCSMUSE technique was applied to abdominal fast spin-echo imaging data, and its effectiveness in respiratory-triggered scans was evaluated. The POCSMUSE method was also applied to reduce aliasing artifacts due to shot-to-shot phase variations in interleaved diffusion-weighted imaging data corresponding to different k-space trajectories and matrix condition numbers. RESULTS: Experimental results show that the POCSMUSE technique can effectively reduce motion-related artifacts in data obtained with different pulse sequences, k-space trajectories and contrasts. CONCLUSION: POCSMUSE is a general post-processing algorithm for reduction of motion-related artifacts. It is compatible with different pulse sequences, and can also be used to further reduce residual artifacts in data produced by existing motion artifact reduction methods.

Role of DNA binding sites and slow unbinding kinetics in titration-based oscillators.

Genetic oscillators, such as circadian clocks, are constantly perturbed by molecular noise arising from the small number of molecules involved in gene regulation. One of the strongest sources of stochasticity is the binary noise that arises from the binding of a regulatory protein to a promoter in the chromosomal DNA. In this study, we focus on two minimal oscillators based on activator titration and repressor titration to understand the key parameters that are important for oscillations and for overcoming binary noise. We show that the rate of unbinding from the DNA, despite traditionally being considered a fast parameter, needs to be slow to broaden the space of oscillatory solutions. The addition of multiple, independent DNA binding sites further expands the oscillatory parameter space for the repressor-titration oscillator and lengthens the period of both oscillators. This effect is a combination of increased effective delay of the unbinding kinetics due to multiple binding sites and increased promoter ultrasensitivity that is specific for repression. We then use stochastic simulation to show that multiple binding sites increase the coherence of oscillations by mitigating the binary noise. Slow values of DNA unbinding rate are also effective in alleviating molecular noise due to the increased distance from the bifurcation point. Our work demonstrates how the number of DNA binding sites and slow unbinding kinetics, which are often omitted in biophysical models of gene circuits, can have a significant impact on the temporal and stochastic dynamics of genetic oscillators.

Case base adaptation using solution-space metrics

In this paper we propose a generalisation of the k-nearest neighbour (k-NN) retrieval method based on an error function using distance metrics in the solution and problem space. It is an interpolative method which is proposed to be effective for sparse case bases. The method applies equally to nominal, continuous and mixed domains, and does not depend upon an embedding n-dimensional space. In continuous Euclidean problem domains, the method is shown to be a generalisation of the Shepard's Interpolation method. We term the retrieval algorithm the Generalised Shepard Nearest Neighbour (GSNN) method. A novel aspect of GSNN is that it provides a general method for interpolation over nominal solution domains. The performance of the retrieval method is examined with reference to the Iris classification problem,and to a simulated sparse nominal value test problem. The introducion of a solution-space metric is shown to out-perform conventional nearest neighbours methods on sparse case bases.

Case base reduction using solution-space metrics

In this paper we propose a case base reduction technique which uses a metric defined on the solution space. The technique utilises the Generalised Shepard Nearest Neighbour (GSNN) algorithm to estimate nominal or real valued solutions in case bases with solution space metrics. An overview of GSNN and a generalised reduction technique, which subsumes some existing decremental methods, such as the Shrink algorithm, are presented. The reduction technique is given for case bases in terms of a measure of the importance of each case to the predictive power of the case base. A trial test is performed on two case bases of different kinds, with several metrics proposed in the solution space. The tests show that GSNN can out-perform standard nearest neighbour methods on this set. Further test results show that a caseremoval order proposed based on a GSNN error function can produce a sparse case base with good predictive power.

A generic time and space accurate numerical approach to closely coupled fluid-structure interaction problems

Fluid structure interaction, as applied to flexible structures, has wide application in diverse areas such as flutter in aircraft, flow in elastic pipes and blood vessels and extrusion of metals through dies. However a comprehensive computational model of these multi-physics phenomena is a considerable challenge. Until recently work in this area focused on one phenomenon and represented the behaviour of the other more simply even to the extent in metal forming, for example, that the deformation of the die is totally ignored. More recently, strategies for solving the full coupling between the fluid and soild mechanics behaviour have developed. Conventionally, the computational modelling of fluid structure interaction is problematical since computational fluid dynamics (CFD) is solved using finite volume (FV) methods and computational structural mechanics (CSM) is based entirely on finite element (FE) methods. In the past the concurrent, but rather disparate, development paths for the finite element and finite volume methods have resulted in numerical software tools for CFD and CSM that are different in almost every respect. Hence, progress is frustrated in modelling the emerging multi-physics problem of fluid structure interaction in a consistent manner. Unless the fluid-structure coupling is either one way, very weak or both, transferring and filtering data from one mesh and solution procedure to another may lead to significant problems in computational convergence. Using a novel three phase technique the full interaction between the fluid and the dynamic structural response are represented. The procedure is demonstrated on some challenging applications in complex three dimensional geometries involving aircraft flutter, metal forming and blood flow in arteries.

Case-based retrieval of 3-dimensional shapes for the design of metal castings

This paper describes research into retrieval based on 3-dimensional shapes for use in the metal casting industry. The purpose of the system is to advise a casting engineer on the design aspects of a new casting by reference to similar castings which have been prototyped and tested in the past. The key aspects of the system are the orientation of the shape within the mould, the positions of feeders and chills, and particular advice concerning special problems and solutions, and possible redesign. The main focus of this research is the effectiveness of similarity measures based on 3-dimensional shapes. The approach adopted here is to construct similarity measures based on a graphical representation deriving from a shape decomposition used extensively by experienced casting design engineers. The paper explains the graphical representation and discusses similarity measures based on it. Performance measures for the CBR system are given, and the results for trials of the system are presented. The competence of the current case-base is discussed, with reference to a representation of cases as points in an n-dimensional feature space, and its principal components visualization. A refinement of the case base is performed as a result of the competence analysis and the performance of the case-base before and after refinement is compared.