Integrated visual perception architecture for robotic clothes perception and manipulation

This thesis proposes a generic visual perception architecture for robotic clothes perception and manipulation. This proposed architecture is fully integrated with a stereo vision system and a dual-arm robot and is able to perform a number of autonomous laundering tasks. Clothes perception and manipulation is a novel research topic in robotics and has experienced rapid development in recent years. Compared to the task of perceiving and manipulating rigid objects, clothes perception and manipulation poses a greater challenge. This can be attributed to two reasons: firstly, deformable clothing requires precise (high-acuity) visual perception and dexterous manipulation; secondly, as clothing approximates a non-rigid 2-manifold in 3-space, that can adopt a quasi-infinite configuration space, the potential variability in the appearance of clothing items makes them difficult to understand, identify uniquely, and interact with by machine. From an applications perspective, and as part of EU CloPeMa project, the integrated visual perception architecture refines a pre-existing clothing manipulation pipeline by completing pre-wash clothes (category) sorting (using single-shot or interactive perception for garment categorisation and manipulation) and post-wash dual-arm flattening. To the best of the author’s knowledge, as investigated in this thesis, the autonomous clothing perception and manipulation solutions presented here were first proposed and reported by the author. All of the reported robot demonstrations in this work follow a perception-manipulation method- ology where visual and tactile feedback (in the form of surface wrinkledness captured by the high accuracy depth sensor i.e. CloPeMa stereo head or the predictive confidence modelled by Gaussian Processing) serve as the halting criteria in the flattening and sorting tasks, respectively. From scientific perspective, the proposed visual perception architecture addresses the above challenges by parsing and grouping 3D clothing configurations hierarchically from low-level curvatures, through mid-level surface shape representations (providing topological descriptions and 3D texture representations), to high-level semantic structures and statistical descriptions. A range of visual features such as Shape Index, Surface Topologies Analysis and Local Binary Patterns have been adapted within this work to parse clothing surfaces and textures and several novel features have been devised, including B-Spline Patches with Locality-Constrained Linear coding, and Topology Spatial Distance to describe and quantify generic landmarks (wrinkles and folds). The essence of this proposed architecture comprises 3D generic surface parsing and interpretation, which is critical to underpinning a number of laundering tasks and has the potential to be extended to other rigid and non-rigid object perception and manipulation tasks. The experimental results presented in this thesis demonstrate that: firstly, the proposed grasp- ing approach achieves on-average 84.7% accuracy; secondly, the proposed flattening approach is able to flatten towels, t-shirts and pants (shorts) within 9 iterations on-average; thirdly, the proposed clothes recognition pipeline can recognise clothes categories from highly wrinkled configurations and advances the state-of-the-art by 36% in terms of classification accuracy, achieving an 83.2% true-positive classification rate when discriminating between five categories of clothes; finally the Gaussian Process based interactive perception approach exhibits a substantial improvement over single-shot perception. Accordingly, this thesis has advanced the state-of-the-art of robot clothes perception and manipulation.

Interaction techniques with novel multimodal feedback for addressing gesture-sensing systems

Users need to be able to address in-air gesture systems, which means finding where to perform gestures and how to direct them towards the intended system. This is necessary for input to be sensed correctly and without unintentionally affecting other systems. This thesis investigates novel interaction techniques which allow users to address gesture systems properly, helping them find where and how to gesture. It also investigates audio, tactile and interactive light displays for multimodal gesture feedback; these can be used by gesture systems with limited output capabilities (like mobile phones and small household controls), allowing the interaction techniques to be used by a variety of device types. It investigates tactile and interactive light displays in greater detail, as these are not as well understood as audio displays. Experiments 1 and 2 explored tactile feedback for gesture systems, comparing an ultrasound haptic display to wearable tactile displays at different body locations and investigating feedback designs. These experiments found that tactile feedback improves the user experience of gesturing by reassuring users that their movements are being sensed. Experiment 3 investigated interactive light displays for gesture systems, finding this novel display type effective for giving feedback and presenting information. It also found that interactive light feedback is enhanced by audio and tactile feedback. These feedback modalities were then used alongside audio feedback in two interaction techniques for addressing gesture systems: sensor strength feedback and rhythmic gestures. Sensor strength feedback is multimodal feedback that tells users how well they can be sensed, encouraging them to find where to gesture through active exploration. Experiment 4 found that they can do this with 51mm accuracy, with combinations of audio and interactive light feedback leading to the best performance. Rhythmic gestures are continuously repeated gesture movements which can be used to direct input. Experiment 5 investigated the usability of this technique, finding that users can match rhythmic gestures well and with ease. Finally, these interaction techniques were combined, resulting in a new single interaction for addressing gesture systems. Using this interaction, users could direct their input with rhythmic gestures while using the sensor strength feedback to find a good location for addressing the system. Experiment 6 studied the effectiveness and usability of this technique, as well as the design space for combining the two types of feedback. It found that this interaction was successful, with users matching 99.9% of rhythmic gestures, with 80mm accuracy from target points. The findings show that gesture systems could successfully use this interaction technique to allow users to address them. Novel design recommendations for using rhythmic gestures and sensor strength feedback were created, informed by the experiment findings.

Phobos: The design and implementation of embedded software for a low cost radar warning receiver

This portfolio thesis describes work undertaken by the author under the Engineering Doctorate program of the Institute for System Level Integration. It was carried out in conjunction with the sponsor company Teledyne Defence Limited. A radar warning receiver is a device used to detect and identify the emissions of radars. They were originally developed during the Second World War and are found today on a variety of military platforms as part of the platform’s defensive systems. Teledyne Defence has designed and built components and electronic subsystems for the defence industry since the 1970s. This thesis documents part of the work carried out to create Phobos, Teledyne Defence’s first complete radar warning receiver. Phobos was designed to be the first low cost radar warning receiver. This was made possible by the reuse of existing Teledyne Defence products, commercial off the shelf hardware and advanced UK government algorithms. The challenges of this integration are described and discussed, with detail given of the software architecture and the development of the embedded application. Performance of the embedded system as a whole is described and qualified within the context of a low cost system.

GPRM: a high performance programming framework for manycore processors

Processors with large numbers of cores are becoming commonplace. In order to utilise the available resources in such systems, the programming paradigm has to move towards increased parallelism. However, increased parallelism does not necessarily lead to better performance. Parallel programming models have to provide not only flexible ways of defining parallel tasks, but also efficient methods to manage the created tasks. Moreover, in a general-purpose system, applications residing in the system compete for the shared resources. Thread and task scheduling in such a multiprogrammed multithreaded environment is a significant challenge. In this thesis, we introduce a new task-based parallel reduction model, called the Glasgow Parallel Reduction Machine (GPRM). Our main objective is to provide high performance while maintaining ease of programming. GPRM supports native parallelism; it provides a modular way of expressing parallel tasks and the communication patterns between them. Compiling a GPRM program results in an Intermediate Representation (IR) containing useful information about tasks, their dependencies, as well as the initial mapping information. This compile-time information helps reduce the overhead of runtime task scheduling and is key to high performance. Generally speaking, the granularity and the number of tasks are major factors in achieving high performance. These factors are even more important in the case of GPRM, as it is highly dependent on tasks, rather than threads. We use three basic benchmarks to provide a detailed comparison of GPRM with Intel OpenMP, Cilk Plus, and Threading Building Blocks (TBB) on the Intel Xeon Phi, and with GNU OpenMP on the Tilera TILEPro64. GPRM shows superior performance in almost all cases, only by controlling the number of tasks. GPRM also provides a low-overhead mechanism, called “Global Sharing”, which improves performance in multiprogramming situations. We use OpenMP, as the most popular model for shared-memory parallel programming as the main GPRM competitor for solving three well-known problems on both platforms: LU factorisation of Sparse Matrices, Image Convolution, and Linked List Processing. We focus on proposing solutions that best fit into the GPRM’s model of execution. GPRM outperforms OpenMP in all cases on the TILEPro64. On the Xeon Phi, our solution for the LU Factorisation results in notable performance improvement for sparse matrices with large numbers of small blocks. We investigate the overhead of GPRM’s task creation and distribution for very short computations using the Image Convolution benchmark. We show that this overhead can be mitigated by combining smaller tasks into larger ones. As a result, GPRM can outperform OpenMP for convolving large 2D matrices on the Xeon Phi. Finally, we demonstrate that our parallel worksharing construct provides an efficient solution for Linked List processing and performs better than OpenMP implementations on the Xeon Phi. The results are very promising, as they verify that our parallel programming framework for manycore processors is flexible and scalable, and can provide high performance without sacrificing productivity.

Utilising presence in places to support mobile interaction

Physical places are given contextual meaning by the objects and people that make up the space. Presence in physical places can be utilised to support mobile interaction by making access to media and notifications on a smartphone easier and more visible to other people. Smartphone interfaces can be extended into the physical world in a meaningful way by anchoring digital content to artefacts, and interactions situated around physical artefacts can provide contextual meaning to private manipulations with a mobile device. Additionally, places themselves are designed to support a set of tasks, and the logical structure of places can be used to organise content on the smartphone. Menus that adapt the functionality of a smartphone can support the user by presenting the tools most likely to be needed just-in-time, so that information needs can be satisfied quickly and with little cognitive effort. Furthermore, places are often shared with people whom the user knows, and the smartphone can facilitate social situations by providing access to content that stimulates conversation. However, the smartphone can disrupt a collaborative environment, by alerting the user with unimportant notifications, or sucking the user in to the digital world with attractive content that is only shown on a private screen. Sharing smartphone content on a situated display creates an inclusive and unobtrusive user experience, and can increase focus on a primary task by allowing content to be read at a glance. Mobile interaction situated around artefacts of personal places is investigated as a way to support users to access content from their smartphone while managing their physical presence. A menu that adapts to personal places is evaluated to reduce the time and effort of app navigation, and coordinating smartphone content on a situated display is found to support social engagement and the negotiation of notifications. Improving the sensing of smartphone users in places is a challenge that is out-with the scope of this thesis. Instead, interaction designers and developers should be provided with low-cost positioning tools that utilise presence in places, and enable quantitative and qualitative data to be collected in user evaluations. Two lightweight positioning tools are developed with the low-cost sensors that are currently available: The Microsoft Kinect depth sensor allows movements of a smartphone user to be tracked in a limited area of a place, and Bluetooth beacons enable the larger context of a place to be detected. Positioning experiments with each sensor are performed to highlight the capabilities and limitations of current sensing techniques for designing interactions with a smartphone. Both tools enable prototypes to be built with a rapid prototyping approach, and mobile interactions can be tested with more advanced sensing techniques as they become available. Sensing technologies are becoming pervasive, and it will soon be possible to perform reliable place detection in-the-wild. Novel interactions that utilise presence in places can support smartphone users by making access to useful functionality easy and more visible to the people who matter most in everyday life.

Garbage collection optimization for non uniform memory access architectures

Cache-coherent non uniform memory access (ccNUMA) architecture is a standard design pattern for contemporary multicore processors, and future generations of architectures are likely to be NUMA. NUMA architectures create new challenges for managed runtime systems. Memory-intensive applications use the system’s distributed memory banks to allocate data, and the automatic memory manager collects garbage left in these memory banks. The garbage collector may need to access remote memory banks, which entails access latency overhead and potential bandwidth saturation for the interconnection between memory banks. This dissertation makes five significant contributions to garbage collection on NUMA systems, with a case study implementation using the Hotspot Java Virtual Machine. It empirically studies data locality for a Stop-The-World garbage collector when tracing connected objects in NUMA heaps. First, it identifies a locality richness which exists naturally in connected objects that contain a root object and its reachable set— ‘rooted sub-graphs’. Second, this dissertation leverages the locality characteristic of rooted sub-graphs to develop a new NUMA-aware garbage collection mechanism. A garbage collector thread processes a local root and its reachable set, which is likely to have a large number of objects in the same NUMA node. Third, a garbage collector thread steals references from sibling threads that run on the same NUMA node to improve data locality. This research evaluates the new NUMA-aware garbage collector using seven benchmarks of an established real-world DaCapo benchmark suite. In addition, evaluation involves a widely used SPECjbb benchmark and Neo4J graph database Java benchmark, as well as an artificial benchmark. The results of the NUMA-aware garbage collector on a multi-hop NUMA architecture show an average of 15% performance improvement. Furthermore, this performance gain is shown to be as a result of an improved NUMA memory access in a ccNUMA system. Fourth, the existing Hotspot JVM adaptive policy for configuring the number of garbage collection threads is shown to be suboptimal for current NUMA machines. The policy uses outdated assumptions and it generates a constant thread count. In fact, the Hotspot JVM still uses this policy in the production version. This research shows that the optimal number of garbage collection threads is application-specific and configuring the optimal number of garbage collection threads yields better collection throughput than the default policy. Fifth, this dissertation designs and implements a runtime technique, which involves heuristics from dynamic collection behavior to calculate an optimal number of garbage collector threads for each collection cycle. The results show an average of 21% improvements to the garbage collection performance for DaCapo benchmarks.

Dynamically managing sensed context data

This thesis reports on an investigation of the feasibility and usefulness of incorporating dynamic management facilities for managing sensed context data in a distributed contextaware mobile application. The investigation focuses on reducing the work required to integrate new sensed context streams in an existing context aware architecture. Current architectures require integration work for new streams and new contexts that are encountered. This means of operation is acceptable for current fixed architectures. However, as systems become more mobile the number of discoverable streams increases. Without the ability to discover and use these new streams the functionality of any given device will be limited to the streams that it knows how to decode. The integration of new streams requires that the sensed context data be understood by the current application. If the new source provides data of a type that an application currently requires then the new source should be connected to the application without any prior knowledge of the new source. If the type is similar and can be converted then this stream too should be appropriated by the application. Such applications are based on portable devices (phones, PDAs) for semi-autonomous services that use data from sensors connected to the devices, plus data exchanged with other such devices and remote servers. Such applications must handle input from a variety of sensors, refining the data locally and managing its communication from the device in volatile and unpredictable network conditions. The choice to focus on locally connected sensory input allows for the introduction of privacy and access controls. This local control can determine how the information is communicated to others. This investigation focuses on the evaluation of three approaches to sensor data management. The first system is characterised by its static management based on the pre-pended metadata. This was the reference system. Developed for a mobile system, the data was processed based on the attached metadata. The code that performed the processing was static. The second system was developed to move away from the static processing and introduce a greater freedom of handling for the data stream, this resulted in a heavy weight approach. The approach focused on pushing the processing of the data into a number of networked nodes rather than the monolithic design of the previous system. By creating a separate communication channel for the metadata it is possible to be more flexible with the amount and type of data transmitted. The final system pulled the benefits of the other systems together. By providing a small management class that would load a separate handler based on the incoming data, Dynamism was maximised whilst maintaining ease of code understanding. The three systems were then compared to highlight their ability to dynamically manage new sensed context. The evaluation took two approaches, the first is a quantitative analysis of the code to understand the complexity of the relative three systems. This was done by evaluating what changes to the system were involved for the new context. The second approach takes a qualitative view of the work required by the software engineer to reconfigure the systems to provide support for a new data stream. The evaluation highlights the various scenarios in which the three systems are most suited. There is always a trade-o↵ in the development of a system. The three approaches highlight this fact. The creation of a statically bound system can be quick to develop but may need to be completely re-written if the requirements move too far. Alternatively a highly dynamic system may be able to cope with new requirements but the developer time to create such a system may be greater than the creation of several simpler systems.

Data structures for SIMD logic simulation

Due to the growth of design size and complexity, design verification is an important aspect of the Logic Circuit development process. The purpose of verification is to validate that the design meets the system requirements and specification. This is done by either functional or formal verification. The most popular approach to functional verification is the use of simulation based techniques. Using models to replicate the behaviour of an actual system is called simulation. In this thesis, a software/data structure architecture without explicit locks is proposed to accelerate logic gate circuit simulation. We call thus system ZSIM. The ZSIM software architecture simulator targets low cost SIMD multi-core machines. Its performance is evaluated on the Intel Xeon Phi and 2 other machines (Intel Xeon and AMD Opteron). The aim of these experiments is to: • Verify that the data structure used allows SIMD acceleration, particularly on machines with gather instructions ( section 5.3.1). • Verify that, on sufficiently large circuits, substantial gains could be made from multicore parallelism ( section 5.3.2 ). • Show that a simulator using this approach out-performs an existing commercial simulator on a standard workstation ( section 5.3.3 ). • Show that the performance on a cheap Xeon Phi card is competitive with results reported elsewhere on much more expensive super-computers ( section 5.3.5 ). To evaluate the ZSIM, two types of test circuits were used: 1. Circuits from the IWLS benchmark suit [1] which allow direct comparison with other published studies of parallel simulators.2. Circuits generated by a parametrised circuit synthesizer. The synthesizer used an algorithm that has been shown to generate circuits that are statistically representative of real logic circuits. The synthesizer allowed testing of a range of very large circuits, larger than the ones for which it was possible to obtain open source files. The experimental results show that with SIMD acceleration and multicore, ZSIM gained a peak parallelisation factor of 300 on Intel Xeon Phi and 11 on Intel Xeon. With only SIMD enabled, ZSIM achieved a maximum parallelistion gain of 10 on Intel Xeon Phi and 4 on Intel Xeon. Furthermore, it was shown that this software architecture simulator running on a SIMD machine is much faster than, and can handle much bigger circuits than a widely used commercial simulator (Xilinx) running on a workstation. The performance achieved by ZSIM was also compared with similar pre-existing work on logic simulation targeting GPUs and supercomputers. It was shown that ZSIM simulator running on a Xeon Phi machine gives comparable simulation performance to the IBM Blue Gene supercomputer at very much lower cost. The experimental results have shown that the Xeon Phi is competitive with simulation on GPUs and allows the handling of much larger circuits than have been reported for GPU simulation. When targeting Xeon Phi architecture, the automatic cache management of the Xeon Phi, handles and manages the on-chip local store without any explicit mention of the local store being made in the architecture of the simulator itself. However, targeting GPUs, explicit cache management in program increases the complexity of the software architecture. Furthermore, one of the strongest points of the ZSIM simulator is its portability. Note that the same code was tested on both AMD and Xeon Phi machines. The same architecture that efficiently performs on Xeon Phi, was ported into a 64 core NUMA AMD Opteron. To conclude, the two main achievements are restated as following: The primary achievement of this work was proving that the ZSIM architecture was faster than previously published logic simulators on low cost platforms. The secondary achievement was the development of a synthetic testing suite that went beyond the scale range that was previously publicly available, based on prior work that showed the synthesis technique is valid.

A distributed, compact routing protocol for the Internet

The Internet has grown in size at rapid rates since BGP records began, and continues to do so. This has raised concerns about the scalability of the current BGP routing system, as the routing state at each router in a shortest-path routing protocol will grow at a supra-linearly rate as the network grows. The concerns are that the memory capacity of routers will not be able to keep up with demands, and that the growth of the Internet will become ever more cramped as more and more of the world seeks the benefits of being connected. Compact routing schemes, where the routing state grows only sub-linearly relative to the growth of the network, could solve this problem and ensure that router memory would not be a bottleneck to Internet growth. These schemes trade away shortest-path routing for scalable memory state, by allowing some paths to have a certain amount of bounded “stretch”. The most promising such scheme is Cowen Routing, which can provide scalable, compact routing state for Internet routing, while still providing shortest-path routing to nearly all other nodes, with only slightly stretched paths to a very small subset of the network. Currently, there is no fully distributed form of Cowen Routing that would be practical for the Internet. This dissertation describes a fully distributed and compact protocol for Cowen routing, using the k-core graph decomposition. Previous compact routing work showed the k-core graph decomposition is useful for Cowen Routing on the Internet, but no distributed form existed. This dissertation gives a distributed k-core algorithm optimised to be efficient on dynamic graphs, along with with proofs of its correctness. The performance and efficiency of this distributed k-core algorithm is evaluated on large, Internet AS graphs, with excellent results. This dissertation then goes on to describe a fully distributed and compact Cowen Routing protocol. This protocol being comprised of a landmark selection process for Cowen Routing using the k-core algorithm, with mechanisms to ensure compact state at all times, including at bootstrap; a local cluster routing process, with mechanisms for policy application and control of cluster sizes, ensuring again that state can remain compact at all times; and a landmark routing process is described with a prioritisation mechanism for announcements that ensures compact state at all times.

NETCODE: an XOR-based warning dissemination scheme for vehicular wireless networks

The next generation of vehicles will be equipped with automated Accident Warning Systems (AWSs) capable of warning neighbouring vehicles about hazards that might lead to accidents. The key enabling technology for these systems is the Vehicular Ad-hoc Networks (VANET) but the dynamics of such networks make the crucial timely delivery of warning messages challenging. While most previously attempted implementations have used broadcast-based data dissemination schemes, these do not cope well as data traffic load or network density increases. This problem of sending warning messages in a timely manner is addressed by employing a network coding technique in this thesis. The proposed NETwork COded DissEmination (NETCODE) is a VANET-based AWS responsible for generating and sending warnings to the vehicles on the road. NETCODE offers an XOR-based data dissemination scheme that sends multiple warning in a single transmission and therefore, reduces the total number of transmissions required to send the same number of warnings that broadcast schemes send. Hence, it reduces contention and collisions in the network improving the delivery time of the warnings. The first part of this research (Chapters 3 and 4) asserts that in order to build a warning system, it is needful to ascertain the system requirements, information to be exchanged, and protocols best suited for communication between vehicles. Therefore, a study of these factors along with a review of existing proposals identifying their strength and weakness is carried out. Then an analysis of existing broadcast-based warning is conducted which concludes that although this is the most straightforward scheme, loading can result an effective collapse, resulting in unacceptably long transmission delays. The second part of this research (Chapter 5) proposes the NETCODE design, including the main contribution of this thesis, a pair of encoding and decoding algorithms that makes the use of an XOR-based technique to reduce transmission overheads and thus allows warnings to get delivered in time. The final part of this research (Chapters 6--8) evaluates the performance of the proposed scheme as to how it reduces the number of transmissions in the network in response to growing data traffic load and network density and investigates its capacity to detect potential accidents. The evaluations use a custom-built simulator to model real-world scenarios such as city areas, junctions, roundabouts, motorways and so on. The study shows that the reduction in the number of transmissions helps reduce competition in the network significantly and this allows vehicles to deliver warning messages more rapidly to their neighbours. It also examines the relative performance of NETCODE when handling both sudden event-driven and longer-term periodic messages in diverse scenarios under stress caused by increasing numbers of vehicles and transmissions per vehicle. This work confirms the thesis' primary contention that XOR-based network coding provides a potential solution on which a more efficient AWS data dissemination scheme can be built.

Probabilistic verification of satellite systems for mission critical applications

In this thesis, we present a quantitative approach using probabilistic verification techniques for the analysis of reliability, availability, maintainability, and safety (RAMS) properties of satellite systems. The subject of our research is satellites used in mission critical industrial applications. A strong case for using probabilistic model checking to support RAMS analysis of satellite systems is made by our verification results. This study is intended to build a foundation to help reliability engineers with a basic background in model checking to apply probabilistic model checking to small satellite systems. We make two major contributions. One of these is the approach of RAMS analysis to satellite systems. In the past, RAMS analysis has been extensively applied to the field of electrical and electronics engineering. It allows system designers and reliability engineers to predict the likelihood of failures from the indication of historical or current operational data. There is a high potential for the application of RAMS analysis in the field of space science and engineering. However, there is a lack of standardisation and suitable procedures for the correct study of RAMS characteristics for satellite systems. This thesis considers the promising application of RAMS analysis to the case of satellite design, use, and maintenance, focusing on its system segments. Data collection and verification procedures are discussed, and a number of considerations are also presented on how to predict the probability of failure. Our second contribution is leveraging the power of probabilistic model checking to analyse satellite systems. We present techniques for analysing satellite systems that differ from the more common quantitative approaches based on traditional simulation and testing. These techniques have not been applied in this context before. We present the use of probabilistic techniques via a suite of detailed examples, together with their analysis. Our presentation is done in an incremental manner: in terms of complexity of application domains and system models, and a detailed PRISM model of each scenario. We also provide results from practical work together with a discussion about future improvements.

The role of context in image annotation and recommendation

With the rise of smart phones, lifelogging devices (e.g. Google Glass) and popularity of image sharing websites (e.g. Flickr), users are capturing and sharing every aspect of their life online producing a wealth of visual content. Of these uploaded images, the majority are poorly annotated or exist in complete semantic isolation making the process of building retrieval systems difficult as one must firstly understand the meaning of an image in order to retrieve it. To alleviate this problem, many image sharing websites offer manual annotation tools which allow the user to “tag” their photos, however, these techniques are laborious and as a result have been poorly adopted; Sigurbjörnsson and van Zwol (2008) showed that 64% of images uploaded to Flickr are annotated with < 4 tags. Due to this, an entire body of research has focused on the automatic annotation of images (Hanbury, 2008; Smeulders et al., 2000; Zhang et al., 2012a) where one attempts to bridge the semantic gap between an image’s appearance and meaning e.g. the objects present. Despite two decades of research the semantic gap still largely exists and as a result automatic annotation models often offer unsatisfactory performance for industrial implementation. Further, these techniques can only annotate what they see, thus ignoring the “bigger picture” surrounding an image (e.g. its location, the event, the people present etc). Much work has therefore focused on building photo tag recommendation (PTR) methods which aid the user in the annotation process by suggesting tags related to those already present. These works have mainly focused on computing relationships between tags based on historical images e.g. that NY and timessquare co-exist in many images and are therefore highly correlated. However, tags are inherently noisy, sparse and ill-defined often resulting in poor PTR accuracy e.g. does NY refer to New York or New Year? This thesis proposes the exploitation of an image’s context which, unlike textual evidences, is always present, in order to alleviate this ambiguity in the tag recommendation process. Specifically we exploit the “what, who, where, when and how” of the image capture process in order to complement textual evidences in various photo tag recommendation and retrieval scenarios. In part II, we combine text, content-based (e.g. # of faces present) and contextual (e.g. day-of-the-week taken) signals for tag recommendation purposes, achieving up to a 75% improvement to precision@5 in comparison to a text-only TF-IDF baseline. We then consider external knowledge sources (i.e. Wikipedia & Twitter) as an alternative to (slower moving) Flickr in order to build recommendation models on, showing that similar accuracy could be achieved on these faster moving, yet entirely textual, datasets. In part II, we also highlight the merits of diversifying tag recommendation lists before discussing at length various problems with existing automatic image annotation and photo tag recommendation evaluation collections. In part III, we propose three new image retrieval scenarios, namely “visual event summarisation”, “image popularity prediction” and “lifelog summarisation”. In the first scenario, we attempt to produce a rank of relevant and diverse images for various news events by (i) removing irrelevant images such memes and visual duplicates (ii) before semantically clustering images based on the tweets in which they were originally posted. Using this approach, we were able to achieve over 50% precision for images in the top 5 ranks. In the second retrieval scenario, we show that by combining contextual and content-based features from images, we are able to predict if it will become “popular” (or not) with 74% accuracy, using an SVM classifier. Finally, in chapter 9 we employ blur detection and perceptual-hash clustering in order to remove noisy images from lifelogs, before combining visual and geo-temporal signals in order to capture a user’s “key moments” within their day. We believe that the results of this thesis show an important step towards building effective image retrieval models when there lacks sufficient textual content (i.e. a cold start).

Transmission electron tomography: quality assessment and enhancement for three-dimensional imaging of nanostructures

Nanotechnology has revolutionised humanity's capability in building microscopic systems by manipulating materials on a molecular and atomic scale. Nan-osystems are becoming increasingly smaller and more complex from the chemical perspective which increases the demand for microscopic characterisation techniques. Among others, transmission electron microscopy (TEM) is an indispensable tool that is increasingly used to study the structures of nanosystems down to the molecular and atomic scale. However, despite the effectivity of this tool, it can only provide 2-dimensional projection (shadow) images of the 3D structure, leaving the 3-dimensional information hidden which can lead to incomplete or erroneous characterization. One very promising inspection method is Electron Tomography (ET), which is rapidly becoming an important tool to explore the 3D nano-world. ET provides (sub-)nanometer resolution in all three dimensions of the sample under investigation. However, the fidelity of the ET tomogram that is achieved by current ET reconstruction procedures remains a major challenge. This thesis addresses the assessment and advancement of electron tomographic methods to enable high-fidelity three-dimensional investigations. A quality assessment investigation was conducted to provide a quality quantitative analysis of the main established ET reconstruction algorithms and to study the influence of the experimental conditions on the quality of the reconstructed ET tomogram. Regular shaped nanoparticles were used as a ground-truth for this study. It is concluded that the fidelity of the post-reconstruction quantitative analysis and segmentation is limited, mainly by the fidelity of the reconstructed ET tomogram. This motivates the development of an improved tomographic reconstruction process. In this thesis, a novel ET method was proposed, named dictionary learning electron tomography (DLET). DLET is based on the recent mathematical theorem of compressed sensing (CS) which employs the sparsity of ET tomograms to enable accurate reconstruction from undersampled (S)TEM tilt series. DLET learns the sparsifying transform (dictionary) in an adaptive way and reconstructs the tomogram simultaneously from highly undersampled tilt series. In this method, the sparsity is applied on overlapping image patches favouring local structures. Furthermore, the dictionary is adapted to the specific tomogram instance, thereby favouring better sparsity and consequently higher quality reconstructions. The reconstruction algorithm is based on an alternating procedure that learns the sparsifying dictionary and employs it to remove artifacts and noise in one step, and then restores the tomogram data in the other step. Simulation and real ET experiments of several morphologies are performed with a variety of setups. Reconstruction results validate its efficiency in both noiseless and noisy cases and show that it yields an improved reconstruction quality with fast convergence. The proposed method enables the recovery of high-fidelity information without the need to worry about what sparsifying transform to select or whether the images used strictly follow the pre-conditions of a certain transform (e.g. strictly piecewise constant for Total Variation minimisation). This can also avoid artifacts that can be introduced by specific sparsifying transforms (e.g. the staircase artifacts the may result when using Total Variation minimisation). Moreover, this thesis shows how reliable elementally sensitive tomography using EELS is possible with the aid of both appropriate use of Dual electron energy loss spectroscopy (DualEELS) and the DLET compressed sensing algorithm to make the best use of the limited data volume and signal to noise inherent in core-loss electron energy loss spectroscopy (EELS) from nanoparticles of an industrially important material. Taken together, the results presented in this thesis demonstrates how high-fidelity ET reconstructions can be achieved using a compressed sensing approach.

Using interaction data for improving the offline and online evaluation of search engines

This thesis investigates how web search evaluation can be improved using historical interaction data. Modern search engines combine offline and online evaluation approaches in a sequence of steps that a tested change needs to pass through to be accepted as an improvement and subsequently deployed. We refer to such a sequence of steps as an evaluation pipeline. In this thesis, we consider the evaluation pipeline to contain three sequential steps: an offline evaluation step, an online evaluation scheduling step, and an online evaluation step. In this thesis we show that historical user interaction data can aid in improving the accuracy or efficiency of each of the steps of the web search evaluation pipeline. As a result of these improvements, the overall efficiency of the entire evaluation pipeline is increased. Firstly, we investigate how user interaction data can be used to build accurate offline evaluation methods for query auto-completion mechanisms. We propose a family of offline evaluation metrics for query auto-completion that represents the effort the user has to spend in order to submit their query. The parameters of our proposed metrics are trained against a set of user interactions recorded in the search engine’s query logs. From our experimental study, we observe that our proposed metrics are significantly more correlated with an online user satisfaction indicator than the metrics proposed in the existing literature. Hence, fewer changes will pass the offline evaluation step to be rejected after the online evaluation step. As a result, this would allow us to achieve a higher efficiency of the entire evaluation pipeline. Secondly, we state the problem of the optimised scheduling of online experiments. We tackle this problem by considering a greedy scheduler that prioritises the evaluation queue according to the predicted likelihood of success of a particular experiment. This predictor is trained on a set of online experiments, and uses a diverse set of features to represent an online experiment. Our study demonstrates that a higher number of successful experiments per unit of time can be achieved by deploying such a scheduler on the second step of the evaluation pipeline. Consequently, we argue that the efficiency of the evaluation pipeline can be increased. Next, to improve the efficiency of the online evaluation step, we propose the Generalised Team Draft interleaving framework. Generalised Team Draft considers both the interleaving policy (how often a particular combination of results is shown) and click scoring (how important each click is) as parameters in a data-driven optimisation of the interleaving sensitivity. Further, Generalised Team Draft is applicable beyond domains with a list-based representation of results, i.e. in domains with a grid-based representation, such as image search. Our study using datasets of interleaving experiments performed both in document and image search domains demonstrates that Generalised Team Draft achieves the highest sensitivity. A higher sensitivity indicates that the interleaving experiments can be deployed for a shorter period of time or use a smaller sample of users. Importantly, Generalised Team Draft optimises the interleaving parameters w.r.t. historical interaction data recorded in the interleaving experiments. Finally, we propose to apply the sequential testing methods to reduce the mean deployment time for the interleaving experiments. We adapt two sequential tests for the interleaving experimentation. We demonstrate that one can achieve a significant decrease in experiment duration by using such sequential testing methods. The highest efficiency is achieved by the sequential tests that adjust their stopping thresholds using historical interaction data recorded in diagnostic experiments. Our further experimental study demonstrates that cumulative gains in the online experimentation efficiency can be achieved by combining the interleaving sensitivity optimisation approaches, including Generalised Team Draft, and the sequential testing approaches. Overall, the central contributions of this thesis are the proposed approaches to improve the accuracy or efficiency of the steps of the evaluation pipeline: the offline evaluation frameworks for the query auto-completion, an approach for the optimised scheduling of online experiments, a general framework for the efficient online interleaving evaluation, and a sequential testing approach for the online search evaluation. The experiments in this thesis are based on massive real-life datasets obtained from Yandex, a leading commercial search engine. These experiments demonstrate the potential of the proposed approaches to improve the efficiency of the evaluation pipeline.

An ontology for risk management of digital collections

Maintaining accessibility to and understanding of digital information over time is a complex challenge that often requires contributions and interventions from a variety of individuals and organizations. The processes of preservation planning and evaluation are fundamentally implicit and share similar complexity. Both demand comprehensive knowledge and understanding of every aspect of to-be-preserved content and the contexts within which preservation is undertaken. Consequently, means are required for the identification, documentation and association of those properties of data, representation and management mechanisms that in combination lend value, facilitate interaction and influence the preservation process. These properties may be almost limitless in terms of diversity, but are integral to the establishment of classes of risk exposure, and the planning and deployment of appropriate preservation strategies. We explore several research objectives within the course of this thesis. Our main objective is the conception of an ontology for risk management of digital collections. Incorporated within this are our aims to survey the contexts within which preservation has been undertaken successfully, the development of an appropriate methodology for risk management, the evaluation of existing preservation evaluation approaches and metrics, the structuring of best practice knowledge and lastly the demonstration of a range of tools that utilise our findings. We describe a mixed methodology that uses interview and survey, extensive content analysis, practical case study and iterative software and ontology development. We build on a robust foundation, the development of the Digital Repository Audit Method Based on Risk Assessment. We summarise the extent of the challenge facing the digital preservation community (and by extension users and creators of digital materials from many disciplines and operational contexts) and present the case for a comprehensive and extensible knowledge base of best practice. These challenges are manifested in the scale of data growth, the increasing complexity and the increasing onus on communities with no formal training to offer assurances of data management and sustainability. These collectively imply a challenge that demands an intuitive and adaptable means of evaluating digital preservation efforts. The need for individuals and organisations to validate the legitimacy of their own efforts is particularly prioritised. We introduce our approach, based on risk management. Risk is an expression of the likelihood of a negative outcome, and an expression of the impact of such an occurrence. We describe how risk management may be considered synonymous with preservation activity, a persistent effort to negate the dangers posed to information availability, usability and sustainability. Risk can be characterised according to associated goals, activities, responsibilities and policies in terms of both their manifestation and mitigation. They have the capacity to be deconstructed into their atomic units and responsibility for their resolution delegated appropriately. We continue to describe how the manifestation of risks typically spans an entire organisational environment, and as the focus of our analysis risk safeguards against omissions that may occur when pursuing functional, departmental or role-based assessment. We discuss the importance of relating risk-factors, through the risks themselves or associated system elements. To do so will yield the preservation best-practice knowledge base that is conspicuously lacking within the international digital preservation community. We present as research outcomes an encapsulation of preservation practice (and explicitly defined best practice) as a series of case studies, in turn distilled into atomic, related information elements. We conduct our analyses in the formal evaluation of memory institutions in the UK, US and continental Europe. Furthermore we showcase a series of applications that use the fruits of this research as their intellectual foundation. Finally we document our results in a range of technical reports and conference and journal articles. We present evidence of preservation approaches and infrastructures from a series of case studies conducted in a range of international preservation environments. We then aggregate this into a linked data structure entitled PORRO, an ontology relating preservation repository, object and risk characteristics, intended to support preservation decision making and evaluation. The methodology leading to this ontology is outlined, and lessons are exposed by revisiting legacy studies and exposing the resource and associated applications to evaluation by the digital preservation community.