Model selection and error estimation

We study model selection strategies based on penalized empirical loss minimization. We point out a tight relationship between error estimation and data-based complexity penalization: any good error estimate may be converted into a data-based penalty function and the performance of the estimate is governed by the quality of the error estimate. We consider several penalty functions, involving error estimates on independent test data, empirical VC dimension, empirical VC entropy, and margin-based quantities. We also consider the maximal difference between the error on the first half of the training data and the second half, and the expected maximal discrepancy, a closely related capacity estimate that can be calculated by Monte Carlo integration. Maximal discrepancy penalty functions are appealing for pattern classification problems, since their computation is equivalent to empirical risk minimization over the training data with some labels flipped.

Fast rates for estimation error and oracle inequalities for model section

We consider complexity penalization methods for model selection. These methods aim to choose a model to optimally trade off estimation and approximation errors by minimizing the sum of an empirical risk term and a complexity penalty. It is well known that if we use a bound on the maximal deviation between empirical and true risks as a complexity penalty, then the risk of our choice is no more than the approximation error plus twice the complexity penalty. There are many cases, however, where complexity penalties like this give loose upper bounds on the estimation error. In particular, if we choose a function from a suitably simple convex function class with a strictly convex loss function, then the estimation error (the difference between the risk of the empirical risk minimizer and the minimal risk in the class) approaches zero at a faster rate than the maximal deviation between empirical and true risks. In this paper, we address the question of whether it is possible to design a complexity penalized model selection method for these situations. We show that, provided the sequence of models is ordered by inclusion, in these cases we can use tight upper bounds on estimation error as a complexity penalty. Surprisingly, this is the case even in situations when the difference between the empirical risk and true risk (and indeed the error of any estimate of the approximation error) decreases much more slowly than the complexity penalty. We give an oracle inequality showing that the resulting model selection method chooses a function with risk no more than the approximation error plus a constant times the complexity penalty.

A gray-box neural network-based model identification and fault estimation scheme for nonlinear dynamic systems

A novel gray-box neural network model (GBNNM), including multi-layer perception (MLP) neural network (NN) and integrators, is proposed for a model identification and fault estimation (MIFE) scheme. With the GBNNM, both the nonlinearity and dynamics of a class of nonlinear dynamic systems can be approximated. Unlike previous NN-based model identification methods, the GBNNM directly inherits system dynamics and separately models system nonlinearities. This model corresponds well with the object system and is easy to build. The GBNNM is embedded online as a normal model reference to obtain the quantitative residual between the object system output and the GBNNM output. This residual can accurately indicate the fault offset value, so it is suitable for differing fault severities. To further estimate the fault parameters (FPs), an improved extended state observer (ESO) using the same NNs (IESONN) from the GBNNM is proposed to avoid requiring the knowledge of ESO nonlinearity. Then, the proposed MIFE scheme is applied for reaction wheels (RW) in a satellite attitude control system (SACS). The scheme using the GBNNM is compared with other NNs in the same fault scenario, and several partial loss of effect (LOE) faults with different severities are considered to validate the effectiveness of the FP estimation and its superiority.

Three Carrier Ambiguity Resolutions : Generalised problems, models and solutions

In this paper, the problems of three carrier phase ambiguity resolution (TCAR) and position estimation (PE) are generalized as real time GNSS data processing problems for a continuously observing network on large scale. In order to describe these problems, a general linear equation system is presented to uniform various geometry-free, geometry-based and geometry-constrained TCAR models, along with state transition questions between observation times. With this general formulation, generalized TCAR solutions are given to cover different real time GNSS data processing scenarios, and various simplified integer solutions, such as geometry-free rounding and geometry-based LAMBDA solutions with single and multiple-epoch measurements. In fact, various ambiguity resolution (AR) solutions differ in the floating ambiguity estimation and integer ambiguity search processes, but their theoretical equivalence remains under the same observational systems models and statistical assumptions. TCAR performance benefits as outlined from the data analyses in some recent literatures are reviewed, showing profound implications for the future GNSS development from both technology and application perspectives.

Sensing and control on the sphere

The advantages of a spherical imaging model are increasingly well recognized within the robotics community. Perhaps less well known is the use of the sphere for attitude estimation, control and scene structure estimation. This paper proposes the sphere as a unifying concept, not just for cameras, but for sensor fusion, estimation and control. We review and summarize relevant work in these areas and illustrate this with relevant simulation examples for spherical visual servoing and scene structure estimation.

Unaided stereo vision based pose estimation

This paper presents the development of a low-cost sensor platform for use in ground-based visual pose estimation and scene mapping tasks. We seek to develop a technical solution using low-cost vision hardware that allows us to accurately estimate robot position for SLAM tasks. We present results from the application of a vision based pose estimation technique to simultaneously determine camera poses and scene structure. The results are generated from a dataset gathered traversing a local road at the St Lucia Campus of the University of Queensland. We show the accuracy of the pose estimation over a 1.6km trajectory in relation to GPS ground truth.

A robust structure and motion replacement for bundle adjustment

This paper demonstrates the application of a robust form of pose estimation and scene reconstruction using data from camera images. We demonstrate results that suggest the ability of the algorithm to rival methods of RANSAC based pose estimation polished by bundle adjustment in terms of solution robustness, speed and accuracy, even when given poor initialisations. Our simulated results show the behaviour of the algorithm in a number of novel simulated scenarios reflective of real world cases that show the ability of the algorithm to handle large observation noise and difficult reconstruction scenes. These results have a number of implications for the vision and robotics community, and show that the application of visual motion estimation on robotic platforms in an online fashion is approaching real-world feasibility.

Arterial travel time estimation : revisiting the classical procedure

Travel time is an important network performance measure and it quantifies congestion in a manner easily understood by all transport users. In urban networks, travel time estimation is challenging due to number of reasons such as, fluctuations in traffic flow due to traffic signals, significant flow to/from mid link sinks/sources, etc. The classical analytical procedure utilizes cumulative plots at upstream and downstream locations for estimating travel time between the two locations. In this paper, we discuss about the issues and challenges with classical analytical procedure such as its vulnerability to non conservation of flow between the two locations. The complexity with respect to exit movement specific travel time is discussed. Recently, we have developed a methodology utilising classical procedure to estimate average travel time and its statistic on urban links (Bhaskar, Chung et al. 2010). Where, detector, signal and probe vehicle data is fused. In this paper we extend the methodology for route travel time estimation and test its performance using simulation. The originality is defining cumulative plots for each exit turning movement utilising historical database which is self updated after each estimation. The performance is also compared with a method solely based on probe (Probe-only). The performance of the proposed methodology has been found insensitive to different route flow, with average accuracy of more than 94% given a probe per estimation interval which is more than 5% increment in accuracy with respect to Probe-only method.

An innovative approach to the assessment of nutrient intake in adults with type 2 diabetes : the development, trial and evaluation of a mobile phone photo/voice dietary record

Nutrition interventions in the form of both self-management education and individualised diet therapy are considered essential for the long-term management of type 2 diabetes mellitus (T2DM). The measurement of diet is essential to inform, support and evaluate nutrition interventions in the management of T2DM. Barriers inherent within health care settings and systems limit ongoing access to personnel and resources, while traditional prospective methods of assessing diet are burdensome for the individual and often result in changes in typical intake to facilitate recording. This thesis investigated the inclusion of information and communication technologies (ICT) to overcome limitations to current approaches in the nutritional management of T2DM, in particular the development, trial and evaluation of the Nutricam dietary assessment method (NuDAM) consisting of a mobile phone photo/voice application to assess nutrient intake in a free-living environment with older adults with T2DM. Study 1: Effectiveness of an automated telephone system in promoting change in dietary intake among adults with T2DM The effectiveness of an automated telephone system, Telephone-Linked Care (TLC) Diabetes, designed to deliver self-management education was evaluated in terms of promoting dietary change in adults with T2DM and sub-optimal glycaemic control. In this secondary data analysis independent of the larger randomised controlled trial, complete data was available for 95 adults (59 male; mean age(±SD)=56.8±8.1 years; mean(±SD)BMI=34.2±7.0kg/m2). The treatment effect showed a reduction in total fat of 1.4% and saturated fat of 0.9% energy intake, body weight of 0.7 kg and waist circumference of 2.0 cm. In addition, a significant increase in the nutrition self-efficacy score of 1.3 (p<0.05) was observed in the TLC group compared to the control group. The modest trends observed in this study indicate that the TLC Diabetes system does support the adoption of positive nutrition behaviours as a result of diabetes self-management education, however caution must be applied in the interpretation of results due to the inherent limitations of the dietary assessment method used. The decision to use a close-list FFQ with known bias may have influenced the accuracy of reporting dietary intake in this instance. This study provided an example of the methodological challenges experienced with measuring changes in absolute diet using a FFQ, and reaffirmed the need for novel prospective assessment methods capable of capturing natural variance in usual intakes. Study 2: The development and trial of NuDAM recording protocol The feasibility of the Nutricam mobile phone photo/voice dietary record was evaluated in 10 adults with T2DM (6 Male; age=64.7±3.8 years; BMI=33.9±7.0 kg/m2). Intake was recorded over a 3-day period using both Nutricam and a written estimated food record (EFR). Compared to the EFR, the Nutricam device was found to be acceptable among subjects, however, energy intake was under-recorded using Nutricam (-0.6±0.8 MJ/day; p<0.05). Beverages and snacks were the items most frequently not recorded using Nutricam; however forgotten meals contributed to the greatest difference in energy intake between records. In addition, the quality of dietary data recorded using Nutricam was unacceptable for just under one-third of entries. It was concluded that an additional mechanism was necessary to complement dietary information collected via Nutricam. Modifications to the method were made to allow for clarification of Nutricam entries and probing forgotten foods during a brief phone call to the subject the following morning. The revised recording protocol was evaluated in Study 4. Study 3: The development and trial of the NuDAM analysis protocol Part A explored the effect of the type of portion size estimation aid (PSEA) on the error associated with quantifying four portions of 15 single foods items contained in photographs. Seventeen dietetic students (1 male; age=24.7±9.1 years; BMI=21.1±1.9 kg/m2) estimated all food portions on two occasions: without aids and with aids (food models or reference food photographs). Overall, the use of a PSEA significantly reduced mean (±SD) group error between estimates compared to no aid (-2.5±11.5% vs. 19.0±28.8%; p<0.05). The type of PSEA (i.e. food models vs. reference food photograph) did not have a notable effect on the group estimation error (-6.7±14.9% vs. 1.4±5.9%, respectively; p=0.321). This exploratory study provided evidence that the use of aids in general, rather than the type, was more effective in reducing estimation error. Findings guided the development of the Dietary Estimation and Assessment Tool (DEAT) for use in the analysis of the Nutricam dietary record. Part B evaluated the effect of the DEAT on the error associated with the quantification of two 3-day Nutricam dietary records in a sample of 29 dietetic students (2 males; age=23.3±5.1 years; BMI=20.6±1.9 kg/m2). Subjects were randomised into two groups: Group A and Group B. For Record 1, the use of the DEAT (Group A) resulted in a smaller error compared to estimations made without the tool (Group B) (17.7±15.8%/day vs. 34.0±22.6%/day, p=0.331; respectively). In comparison, all subjects used the DEAT to estimate Record 2, with resultant error similar between Group A and B (21.2±19.2%/day vs. 25.8±13.6%/day; p=0.377 respectively). In general, the moderate estimation error associated with quantifying food items did not translate into clinically significant differences in the nutrient profile of the Nutricam dietary records, only amorphous foods were notably over-estimated in energy content without the use of the DEAT (57kJ/day vs. 274kJ/day; p<0.001). A large proportion (89.6%) of the group found the DEAT helpful when quantifying food items contained in the Nutricam dietary records. The use of the DEAT reduced quantification error, minimising any potential effect on the estimation of energy and macronutrient intake. Study 4: Evaluation of the NuDAM The accuracy and inter-rater reliability of the NuDAM to assess energy and macronutrient intake was evaluated in a sample of 10 adults (6 males; age=61.2±6.9 years; BMI=31.0±4.5 kg/m2). Intake recorded using both the NuDAM and a weighed food record (WFR) was coded by three dietitians and compared with an objective measure of total energy expenditure (TEE) obtained using the doubly labelled water technique. At the group level, energy intake (EI) was under-reported to a similar extent using both methods, with the ratio of EI:TEE was 0.76±0.20 for the NuDAM and 0.76±0.17 for the WFR. At the individual level, four subjects reported implausible levels of energy intake using the WFR method, compared to three using the NuDAM. Overall, moderate to high correlation coefficients (r=0.57-0.85) were found across energy and macronutrients except fat (r=0.24) between the two dietary measures. High agreement was observed between dietitians for estimates of energy and macronutrient derived for both the NuDAM (ICC=0.77-0.99; p<0.001) and WFR (ICC=0.82-0.99; p<0.001). All subjects preferred using the NuDAM over the WFR to record intake and were willing to use the novel method again over longer recording periods. This research program explored two novel approaches which utilised distinct technologies to aid in the nutritional management of adults with T2DM. In particular, this thesis makes a significant contribution to the evidence base surrounding the use of PhRs through the development, trial and evaluation of a novel mobile phone photo/voice dietary record. The NuDAM is an extremely promising advancement in the nutritional management of individuals with diabetes and other chronic conditions. Future applications lie in integrating the NuDAM with other technologies to facilitate practice across the remaining stages of the nutrition care process.

Challenges to environmental toxicology and epidemiology : where do we stand and which way do we go?

Modern toxicology investigates a wide array of both old and new health hazards. Priority setting is needed to select agents for research from the plethora of exposure circumstances. The changing societies and a growing fraction of the aged have to be taken into consideration. A precise exposure assessment is of importance for risk estimation and regulation. Toxicology contributes to the exploration of pathomechanisms to specify the exposure metrics for risk estimation. Combined effects of co-existing agents are not yet sufficiently understood. Animal experiments allow a separate administration of agents which can not be disentangled by epidemiological means, but their value is limited for low exposure levels in many of today’s settings. As an experimental science, toxicology has to keep pace with the rapidly growing knowledge about the language of the genome and the changing paradigms in cancer development. During the pioneer era of assembling a working draft of the human genome, toxicogenomics has been developed. Gene and pathway complexity have to be considered when investigating gene–environment interactions. For a best conduct of studies, modern toxicology needs a close liaison with many other disciplines like epidemiology and bioinformatics.

The increased popularity of mopeds and motor scooters : exploring usage patterns and safety outcomes

Increased use of powered two-wheelers (PTWs) often underlies increases in the number of reported crashes, promoting research into PTW safety. PTW riders are overrepresented in crash and injury statistics relative to exposure and, as such, are considered vulnerable road users. PTW use has increased substantially over the last decade in many developed countries. One such country is Australia, where moped and scooter use has increased at a faster rate than motorcycle use in recent years. Increased moped use is particularly evident in the State of Queensland which is one of four Australian jurisdictions where moped riding is permitted for car licence holders and a motorcycle licence is not required. A moped is commonly a small motor scooter and is limited to a maximum design speed of 50 km/h and a maximum engine cylinder capacity of 50 cubic centimetres. Scooters exceeding either of these specifications are classed as motorcycles in all Australian jurisdictions. While an extensive body of knowledge exists on motorcycle safety, some of which is relevant to moped and scooter safety, the latter PTW types have received comparatively little focused research attention. Much of the research on moped safety to date has been conducted in Europe where they have been popular since the mid 20th century, while some studies have also been conducted in the United States. This research is of limited relevance to Australia due to socio-cultural, economic, regulatory and environmental differences. Moreover, while some studies have compared motorcycles to mopeds in terms of safety, no research to date has specifically examined the differences and similarities between mopeds and larger scooters, or between larger scooters and motorcycles. To address the need for a better understanding of moped and scooter use and safety, the current program of research involved three complementary studies designed to achieve the following aims: (1) develop better knowledge and understanding of moped and scooter usage trends and patterns; and (2) determine the factors leading to differences in moped, scooter and motorcycle safety. Study 1 involved six-monthly observations of PTW types in inner city parking areas of Queensland’s capital city, Brisbane, to monitor and quantify the types of PTW in use over a two year period. Study 2 involved an analysis of Queensland PTW crash and registration data, primarily comparing the police-reported crash involvement of mopeds, scooters and motorcycles over a five year period (N = 7,347). Study 3 employed both qualitative and quantitative methods to examine moped and scooter usage in two components: (a) four focus group discussions with Brisbane-based Queensland moped and scooter riders (N = 23); and (b) a state-wide survey of Queensland moped and scooter riders (N = 192). Study 1 found that of the PTW types parked in inner city Brisbane over the study period (N = 2,642), more than one third (36.1%) were mopeds or larger scooters. The number of PTWs observed increased at each six-monthly phase, but there were no significant changes in the proportions of PTW types observed across study phases. There were no significant differences in the proportions or numbers of PTW type observed by season. Study 2 revealed some important differences between mopeds, scooters and motorcycles in terms of safety and usage through analysis of crash and registration data. All Queensland PTW registrations doubled between 2001 and 2009, but there was an almost fifteen-fold increase in moped registrations. Mopeds subsequently increased as a proportion of Queensland registered PTWs from 1.2 percent to 8.8 percent over this nine year period. Moped and scooter crashes increased at a faster rate than motorcycle crashes over the five year study period from July 2003 to June 2008, reflecting their relatively greater increased usage. Crash rates per 10,000 registrations for the study period were only slightly higher for mopeds (133.4) than for motorcycles and scooters combined (124.8), but estimated crash rates per million vehicle kilometres travelled were higher for mopeds (6.3) than motorcycles and scooters (1.7). While the number of crashes increased for each PTW type over the study period, the rate of crashes per 10,000 registrations declined by 40 percent for mopeds compared with 22 percent for motorcycles and scooters combined. Moped and scooter crashes were generally less severe than motorcycle crashes and this was related to the particular crash characteristics of the PTW types rather than to the PTW types themselves. Compared to motorcycle and moped crashes, scooter crashes were less likely to be single vehicle crashes, to involve a speeding or impaired rider, to involve poor road conditions, or to be attributed to rider error. Scooter and moped crashes were more likely than motorcycle crashes to occur on weekdays, in lower speed zones and at intersections. Scooter riders were older on average (39) than moped (32) and motorcycle (35) riders, while moped riders were more likely to be female (36%) than scooter (22%) or motorcycle riders (7%). The licence characteristics of scooter and motorcycle riders were similar, with moped riders more likely to be licensed outside of Queensland and less likely to hold a full or open licence. The PTW type could not be identified in 15 percent of all cases, indicating a need for more complete recording of vehicle details in the registration data. The focus groups in Study 3a and the survey in Study 3b suggested that moped and scooter riders are a heterogeneous population in terms of demographic characteristics, riding experience, and knowledge and attitudes regarding safety and risk. The self-reported crash involvement of Study 3b respondents suggests that most moped and scooter crashes result in no injury or minor injury and are not reported to police. Study 3 provided some explanation for differences observed in Study 2 between mopeds and scooters in terms of crash involvement. On the whole, scooter riders were older, more experienced, more likely to have undertaken rider training and to value rider training programs. Scooter riders were also more likely to use protective clothing and to seek out safety-related information. This research has some important practical implications regarding moped and scooter use and safety. While mopeds and scooters are generally similar in terms of usage, and their usage has increased, scooter riders appear to be safer than moped riders due to some combination of superior skills and safer riding behaviour. It is reasonable to expect that mopeds and scooters will remain popular in Queensland in future and that their usage may further increase, along with that of motorcycles. Future policy and planning should consider potential options for encouraging moped riders to acquire better riding skills and greater safety awareness. While rider training and licensing appears an obvious potential countermeasure, the effectiveness of rider training has not been established and other options should also be strongly considered. Such options might include rider education and safety promotion, while interventions could also target other road users and urban infrastructure. Future research is warranted in regard to moped and scooter safety, particularly where the use of those PTWs has increased substantially from low levels. Research could address areas such as rider training and licensing (including program evaluations), the need for more detailed and reliable data (particularly crash and exposure data), protective clothing use, risks associated with lane splitting and filtering, and tourist use of mopeds. Some of this research would likely be relevant to motorcycle use and safety, as well as that of mopeds and scooters.

Machine prognostics based on health state estimation using SVM

The ability to accurately predict the remaining useful life of machine components is critical for machine continuous operation, and can also improve productivity and enhance system safety. In condition-based maintenance (CBM), maintenance is performed based on information collected through condition monitoring and an assessment of the machine health. Effective diagnostics and prognostics are important aspects of CBM for maintenance engineers to schedule a repair and to acquire replacement components before the components actually fail. All machine components are subjected to degradation processes in real environments and they have certain failure characteristics which can be related to the operating conditions. This paper describes a technique for accurate assessment of the remnant life of machines based on health state probability estimation and involving historical knowledge embedded in the closed loop diagnostics and prognostics systems. The technique uses a Support Vector Machine (SVM) classifier as a tool for estimating health state probability of machine degradation, which can affect the accuracy of prediction. To validate the feasibility of the proposed model, real life historical data from bearings of High Pressure Liquefied Natural Gas (HP-LNG) pumps were analysed and used to obtain the optimal prediction of remaining useful life. The results obtained were very encouraging and showed that the proposed prognostic system based on health state probability estimation has the potential to be used as an estimation tool for remnant life prediction in industrial machinery.

Freeway travel time estimation & prediction : a literature review

This report is the second deliverable of the Real Time and Predictive Traveller Information project and the first deliverable of the Freeway Travel Time Information sub-project in the Integrated Traveller Information research Domain of the Smart Transport Research Centre. The primary objective of the Freeway Travel Time Information sub-project is to develop algorithms for real-time travel time estimation and prediction models for Freeway traffic. The objective of this report is to review the literature pertaining to travel time estimation and prediction models for freeway traffic.

Arterial travel time estimation & prediction : a literature review

This report is the fourth deliverable of the Real Time and Predictive Traveller Information project and the first deliverable of the Arterial Travel Time Information sub-project in the Integrated Traveller Information research Domain of the Smart Transport Research Centre. The primary objective of the Arterial Travel Time Information sub-project is to develop algorithms for real-time travel time estimation and prediction models for arterial traffic. The objective of this report is to review the literature pertaining to travel time estimation and prediction models for arterial traffic.