Search

Pruning unimportant parameters can allow deep neural networks (DNNs) to reduce their heavy computation and memory requirements. A saliency metric estimates which parameters can be safely pruned with little impact on the classification performance of the DNN. Many saliency metrics have been proposed, each within the context of a wider pruning algorithm. The result is that it is difficult to separate the effectiveness of the saliency metric from the wider pruning algorithm that surrounds it. Similar-looking saliency metrics can yield very different results because of apparently minor design choices. We propose a novel taxonomy of saliency metrics based on four mostly-orthogonal principal components. We show that a broad range of metrics from the pruning literature can be grouped according to these components. Our taxonomy serves as a guide to prior work, and allows us to construct new saliency metrics by exploring novel combinations of our taxonomic components. We perform the first in-depth experimental investigation of more than 300 saliency metrics made up of existing techniques and new combinations of components. Our results provide decisive answers to open research questions. In particular, we demonstrate the importance of reduction and scaling when pruning groups of weights. We also propose a novel scaling method based on the number of weights transitively removed. We find that some of our constructed metrics can outperform the best existing state-of-the-art metrics for convolutional neural network channel pruning. We find further that our novel scaling method improves existing saliency metrics.

Porosity, one of the important quality attributes of pharmaceutical tablets, directly affects the mechanical properties, the mass transport and hence tablet disintegration, dissolution and ultimately the bioavailability of an orally administered drug. The ability to accurately and quickly monitor the porosity of tablets during manufacture or during the manufacturing process will enable a greater assurance of product quality. This tutorial systematically outlines the steps involved in the terahertz-based measurement method that can be used to quantify the porosity of a tablet within seconds in a non-destructive and non-invasive manner. The terahertz-based porosity measurement can be performed using one of the three main methods, which are (i) the zero-porosity approximation (ZPA); (ii) the traditional Bruggeman effective medium approximation (TB-EMA); and (iii) the anisotropic Bruggeman effective medium approximation (AB-EMA). By using a set of batches of flat-faced and biconvex tablets as a case study, the three main methods are compared and contrasted. Overall, frequency-domain signal processing coupled with the AB-EMA method was found to be most suitable approach in terms of accuracy and robustness when predicting the porosity of tablets over a range of complexities and geometries. This tutorial aims to concisely outline all the necessary steps, precautions and unique advantages associated with the terahertz-based porosity measurement method.

Background Treatment of HIV with antiretroviral therapy (ART) can improve the health of people living with HIV (PLHIV), stop onward transmission of HIV and effectively prevent the spread of the virus. In 2016, we conducted a systematic review to assess the feasibility of treatment monitoring for PLHIV on ART in low and middle-income countries (LMICs), in line with the 90-90-90 treatment target. By 2020, global estimates suggest the 90-90-90 target remains unattainable in many LMICs. This study aims to review the progress and identify needs for public health interventions to improve viral load monitoring and viral suppression for PLHIV in LMICs. Methods A literature search was conducted using an update of the initial search strategy developed for the 2016 review with key search terms relevant to HIV treatment and care, decentralization and viral load monitoring. Electronic databases (Medline and PubMed) were searched to identify relevant literature published in English between Dec 2015 and August 2021. The primary outcome was initial viral load (VL) monitoring (the proportion of PLHIV on ART and eligible for VL monitoring who received a VL test). Secondary outcomes included follow-up VL monitoring (the proportion of PLHIV who received a follow-up VL after an initial elevated VL test), confirmation of treatment failure (the proportion of PLHIV who had two consecutive elevated VL results) and switching treatment regimen rates (the proportion of PLHIV who switched treatment regimen after confirmation of treatment failure). Results The search strategy identified 1984 non-duplicate records, of which 34 studies were included in the review. More than 85% (29/34) of included studies were conducted in 11 sub-Saharan African countries (SSA) using routinely collected program data; two studies were conducted among key populations (KPs) attending research clinics. Sixty per cent (20/34) of these studies were designed to evaluate VL monitoring and/or VL cascade among PLHIV on ART, and most were published in 2019–2021. Marked variations in initial VL monitoring coverage were reported across study settings/countries (range: 12–93% median: 74% IQR: 46-82%) and study populations (adults (range: 25–96%, median: 67% IQR: 50-84%), children, adolescents/young people (range: 2–94%, median: 72% IQR: 47-85%), and pregnant women (range: 32–82%, median: 57% IQR: 43-71%)). Community-based models reported higher VL monitoring (median: 85%, IQR: 82%-88%) compared to decentralised care at primary health facility (median: 64%, IRQ: 48%-82%). Suboptimal uptake of follow-up VL monitoring and low regimen switching rates were observed. Conclusions There was a marked increase in the number of studies of VL monitoring for PLHIV on ART in LMICs over the past five years. Substantial gaps in VL coverage across study settings and study populations were evident, with limited data availability outside of SSA and in KPs. Further research is needed to fill the data gaps. Development and implementation of innovative, community-based interventions are required to improve VL monitoring and address the “failure cascade” in PLHIV on ART who fail to achieve viral suppression.

“In what circumstances would you want this AI to make decisions on your behalf?” We have been investigating how to enable a user of an Artificial Intelligence‐powered system to answer questions like this through a series of empirical studies, a group of which we summarize here. We began the series by (a) comparing four explanation configurations of saliency explanations and/or reward explanations. From this study we learned that, although some configurations had significant strengths, no one configuration was a clear “winner.” This result led us to hypothesize that one reason for the low success rates Explainable AI (XAI) research has in enabling users to create a coherent mental model is that the AI itself does not have a coherent model. This hypothesis led us to (b) build a model‐based agent, to compare explaining it with explaining a model‐free agent. Our results were encouraging, but we then realized that participants' cognitive energy was being sapped by having to create not only a mental model, but also a process by which to create that mental model. This realization led us to (c) create such a process (which we term After‐Action Review for AI or “AAR/AI”) for them, integrate it into the explanation environment, and compare participants' success with AAR/AI scaffolding vs without it. Our AAR/AI studies' results showed that AAR/AI participants were more effective assessing the AI than non‐AAR/AI participants, with significantly better precision and significantly better recall at finding the AI's reasoning flaws.

Data mining is a process of finding correlations and collecting and analysing a huge amount of data in a database to discover patterns or relationships. Flight delay creates significant problems in the present aviation system. Data mining techniques are desired for analysing the performance in which micro-level causes propagate to make system-level patterns of delay. Analysing flight delays is very difficult – both when looking from a historical view as well as when estimating delays with forecast demand. This paper proposes using Decision Tree (DT), Support Vector Machine (SVM), Naive Bayesian (NB), K-nearest neighbour (KNN) and Artificial Neural Network (ANN) to study and analyse delays among aircrafts. The performance of different data mining methods is found in the different regions of the updated datasets on these classifiers. Finally, the result shows a significant variation in the performance of different data mining methods and feature selection for this problem. This paper aims to deal with how data mining techniques can be used to understand difficult aircraft system delays in aviation. Our aim is to develop a classification model for studying and reducing delay using different data mining methods and, in this manner, to show that DT has a greater classification accuracy. The different feature selectors are used in this study in order to reduce the number of initial attributes. Our results clearly demonstrate the value of DT for analysing and visualising how system-level effects happen from subsystem-level causes.

How should empirical researchers conduct controlled, remote “lab” studies in the uncontrolled, noisy conditions of each participant's own home? Volatility in participant home environments, hardware, internet connection, and surrounding distractions takes the “controlled” out of controlled studies. This paper recounts our in-the-trenches mitigations for designing and conducting two complex controlled studies under COVID, in which participants, from home, interactively localized faults in an AI system. The studies with our COVID-era mitigations in 5 categories—Privacy/Security, Data Collection, Control, Technology Issues, Payment-ultimately produced crisp results beyond what we thought possible under such uncontrolled circumstances.

The cost of Convolutional Neural Network (CNN) inference can be reduced by pruning weights from a trained network, eliminating computations while preserving the predictive accuracy up to some threshold. While many heuristic saliency metrics have been proposed to guide this process, the quality of pruning decisions made by any one metric is highly contextsensitive. Some metrics make excellent pruning decisions for one network, while making poor decisions for other networks. Traditionally, a single heuristic saliency metric is used for the entire pruning process. We show how to compose a set of these saliency metrics to form a much more robust (albeit still heuristic) saliency. The key idea is to exploit the cases where the different base metrics do well, and avoid the cases where they do poorly by switching to a different metric. With an experimental evaluation of channel pruning on several popular CNNs on the CIFAR-10 and CIFAR-100 datasets, we show that the composite saliency metrics derived by our method consistently outperform all of the individual constituent metrics.

Convolutional neural networks (CNNs) are used in many embedded applications, from industrial robotics and automation systems to biometric identification on mobile devices. State-of-the-art classification is typically achieved by large networks, which are prohibitively expensive to run on mobile and embedded devices with tightly constrained memory and energy budgets. We propose an approach for ahead-of-time domain specific optimization of CNN models, based on an integer linear programming (ILP) for selecting primitive operations to implement convolutional layers. We optimize the trade-off between execution time and memory consumption by: 1) attempting to minimize execution time across the whole network by selecting data layouts and primitive operations to implement each layer; and 2) allocating an appropriate workspace that reflects the upper bound of memory footprint per layer. These two optimization strategies can be used to run any CNN on any platform with a C compiler. Our evaluation with a range of popular ImageNet neural architectures (GoogleNet, AlexNet, VGG, ResNet and SqueezeNet) on the ARM Cortex-A15 yields speedups of 8x compared to a greedy algorithm based primitive selection, reduces memory requirement by 2.2x while sacrificing only 15% of inference time compared to a solver that considers inference time only. In addition, our optimization approach exposes a range of optimal points for different configurations across the Pareto frontier of memory and latency trade-off, which can be used under arbitrary system constraints.

Next generation of embedded Information and Communication Technology (ICT) systems are interconnected and collaborative systems able to perform autonomous tasks. The remarkable expansion of the embedded ICT market, together with the rise and breakthroughs of Artificial Intelligence (AI), have put the focus on the Edge as it stands as one of the keys for the next technological revolution: the seamless integration of AI in our daily life. However, training and deployment of custom AI solutions on embedded devices require a fine-grained integration of data, algorithms, and tools to achieve high accuracy and overcome functional and non-functional requirements. Such integration requires a high level of expertise that becomes a real bottleneck for small and medium enterprises wanting to deploy AI solutions on the Edge , which, ultimately, slows down the adoption of AI on applications in our daily life. In this work, we present a modular AI pipeline as an integrating framework to bring data, algorithms, and deployment tools together. By removing the integration barriers and lowering the required expertise, we can interconnect the different stages of particular tools and provide a modular end-to-end development of AI products for embedded devices. Our AI pipeline consists of four modular main steps: (i) data ingestion, (ii) model training, (iii) deployment optimization, and (iv) the IoT hub integration. To show the effectiveness of our pipeline, we provide examples of different AI applications during each of the steps. Besides, we integrate our deployment framework, Low-Power Deep Neural Network (LPDNN), into the AI pipeline and present its lightweight architecture and deployment capabilities for embedded devices. Finally, we demonstrate the results of the AI pipeline by showing the deployment of several AI applications such as keyword spotting, image classification, and object detection on a set of well-known embedded platforms, where LPDNN consistently outperforms all other popular deployment frameworks.

A novel process was developed for the biorefining of floatable wastewater scum and other waste oils from water treatment facilities into biodiesel and other value-added bio-products. To test the scalability and commercial potential of the technology, a 7000 l/year pilot-scale system was designed and built. Scum from a wastewater treatment facility, located in St. Paul, Mn, was collected and converted into methyl esters (biodiesel) according to the process chemistry. All of the incoming and outgoing process streams were sampled, tested, weighed and recorded to calculate both the process efficiency and product quality. Data from the pilot-scale system operation was compared to laboratory results and the theoretically expected values for each individual unit operation. The biodiesel was tested using a third party laboratory and confirmed it met all of the US EPA’s test requirements for commercial-grade biodiesel.

Hardware-Software Co-Design is a highly successful strategy for improving performance of domain-specific computing systems. We argue for the application of the same methodology to deep learning; specifically, we propose to extend neural architecture search with information about the hardware to ensure that the model designs produced are highly efficient in addition to the typical criteria around accuracy. Using the task of keyword spotting in audio on edge computing devices, we demonstrate that our approach results in neural architecture that is not only highly accurate, but also efficiently mapped to the computing platform which will perform the inference. Using our modified neural architecture search, we demonstrate $0.88\%$ increase in TOP-1 accuracy with $1.85\times$ reduction in latency for keyword spotting in audio on an embedded SoC, and $1.59\times$ on a high-end GPU., Comment: The 2019 International Conference on High Performance Computing & Simulation

In recent years, deep neural networks (DNNs) have emerged as the most successful technology for many difficult problems in image, video, voice and text processing. DNNs are resource hungry and require very large amounts of computation and memory, which is a particular challenge on IoT, mobile and embedded systems. In this chapter, we outline some major performance challenges of DNNs such as computation, parallelism, data locality and memory requirements. We describe research on these problems, such as the use of existing high-performance linear algebra libraries, hardware acceleration, reduced-precision storage and arithmetic and sparse data representations. Finally, we discuss recent trends in adapting compiler and domain-specific program generation techniques to create high-performance parallel DNN programs.

Quantization of weights and activations in Deep Neural Networks (DNNs) is a powerful technique for network compression, and has enjoyed significant attention and success. However, much of the inference-time benefit of quantization is accessible only through the use of customized hardware accelerators or by providing an FPGA implementation of quantized arithmetic. Building on prior work, we show how to construct arbitrary bit-precise signed and unsigned integer operations using a software technique which logically \emph{embeds} a vector architecture with custom bit-width lanes in universally available fixed-width scalar arithmetic. We evaluate our approach on a high-end Intel Haswell processor, and an embedded ARM processor. Our approach yields very fast implementations of bit-precise custom DNN operations, which often match or exceed the performance of operations quantized to the sizes supported in native arithmetic. At the strongest level of quantization, our approach yields a maximum speedup of $\thicksim6\times$ on the Intel platform, and $\thicksim10\times$ on the ARM platform versus quantization to native 8-bit integers.

The large-scale adoption of composite materials in industry has allowed for a greater freedom in design and function of structures and their respective components. However, the freedom of material choice has resulted in increased complexity in manufacturing. Machine learning (ML) and Artificial Intelligence (AI) are currently being explored for a number of advanced manufacturing applications, and their applicability has begun to extend into the composites manufacturing realm. In this document, a comprehensive overview of machine learning applications in composites manufacturing will be presented with discussions on a novel inspection software developed for the Automated Fiber Placement (AFP) process at the University of South Carolina utilizing an ML vision system. This vision system allows for defect data to be fully integrated into the manufacturing process, allowing for the ML inspection system to influence several chains in the composites product lifecycle management.

Building software systems is hard work, with challenges ranging from technical issues to usability issues. If the technical issues are not addressed, the software cannot work - but if the usability issues are not addressed, many potential users and customers are not even interested in whether it works. Further, usability must be inclusive: software needs to support diverse sorts of users. To help software professionals address gender-inclusive usability, we have created the GenderMag Recorder's Assistant tool. This Open Source tool is the first to semi-automate evaluating gender biases in software that is being designed, developed, or maintained. In this showpiece, we will demo the tool and encourage attendees to get involved in using it and improving upon it.

Convolutional neural networks (CNNs) have emerged as one of the most successful machine learning technologies for image and video processing. The most computationally intensive parts of CNNs are the convolutional layers, which convolve multi-channel images with multiple kernels. A common approach to implementing convolutional layers is to expand the image into a column matrix (im2col) and perform Multiple Channel Multiple Kernel (MCMK) convolution using an existing parallel General Matrix Multiplication (GEMM) library. This im2col conversion greatly increases the memory footprint of the input matrix and reduces data locality. In this paper we propose a new approach to MCMK convolution that is based on General Matrix Multiplication (GEMM), but not on im2col. Our algorithm eliminates the need for data replication on the input thereby enabling us to apply the convolution kernels on the input images directly. We have implemented several variants of our algorithm on a CPU processor and an embedded ARM processor. On the CPU, our algorithm is faster than im2col in most cases., Camera ready version to be published at ASAP 2017 - The 28th Annual IEEE International Conference on Application-specific Systems, Architectures and Processors. 6 pages

The purpose of this study was to evaluate the minimum amount of contrast material for coronary imaging with aortic root catheter-directed enhancement and 64-MDCT angiography (MDCTA).A 64-MDCT scanner was used after animal institutional review board approval to study four swine (40-60 kg). Heart rate reduction to 65 beats per minute was achieved with atenolol by mouth and i.v. Cardizem. Common femoral artery access was obtained with a 5-French micropuncture kit and sonographic guidance. A diffusiontip (640 side holes), 5-French pigtail catheter was positioned in the aortic root on the CT table with a retrofitted C-arm fluoroscopy unit and connected to an arterial power injector. Aortic root MDCTA (retrospective ECG gating; collimation, 0.6 mm; tube rotation time, 0.33 second; scanning time, 10-12 seconds; tube voltage, 120 kVp; effective mAs, 850 mAs; pitch, 0.2; field of view, 109-123 mm; slice thickness and increment, 0.6 and 0.3 mm) was begun 1 second after the injection of 100 mL of various Visipaque (iodixanol) concentrations (10%, 20%, 30%, 40%) at 10 mL/s. Coronary mean and peak densities, 3D maximum intensity projections, and 4D projections were obtained.The mean pooled coronary attenuation values (H +/- SD) for the right (RCA), left anterior descending (LAD), and left circumflex (LCx) coronary arteries at various concentrations (10%, 20%, 30%, 40%) were as follows: 10% (RCA [232.6 +/- 64.0], LAD [180.4 +/- 45.1], and LCx [176.6 +/- 56.2]); 20% (RCA [383.0 +/- 98.7], LAD [324.3 +/- 60.1], and LCx [331.8 +/- 105.5]); 30% (RCA [441.8 +/- 137.6], LAD [401.3 +/- 125.8], and LCx [418.5 +/- 173.0]); and 40% (RCA [717.3 +/- 377.7], LAD [573.3 +/- 233.3], and LCx [584.8 +/- 189.0]). Coronary imaging with aortic root MDCTA was feasible at all concentrations, and the attenuation values were statistically significantly greater than 250 H at 20%, 30%, and 40% (p0.05). The attenuation values with aortic root MDCTA using one fifth of the volume of contrast material are comparable to those currently achieved both clinically and experimentally with peripheral i.v. MDCTA.Aortic root MDCTA can depict the coronary arteries with as little as 20 mL of contrast material. This may provide an alternative means of coronary evaluation in patients with renal insufficiency.

Objective To determine the effect of using experienced general practitioners (GPs) to review the advice given by call handlers in NHS 111, a national service giving telephone advice to people seeking medical care. Design Observational study following the introduction of GPs to review call handlers’ decisions which had been made using decision support software. Setting NHS 111 call centre covering Cambridgeshire and Peterborough. Intervention When a call handler using standard NHS 111 decision support software would have advised the caller to attend the hospital accident and emergency (A&E) department, the decision was reviewed by an experienced GP. Main outcome measures Percentage of calls where an outcome other than A&E attendance was recommended by the GP. Results Of 1474 cases reviewed, the GP recommended A&E attendance in 400 cases (27.1%). In the remainder of cases, the GP recommended attendance at a primary care out-of-hours centre or minor injury unit in 665 cases (45.2%) and self-management or some alternative strategy in 409 (27.8%). Conclusions Fewer callers to NHS 111 would be sent to emergency departments if the decision was reviewed by an experienced GP. Telephone triage services need to consider whether using relatively unskilled call handlers supported by computer software is the most cost-effective way to handle requests for medical care.

The ideal procedural sedation drug for emergency department (ED) use would be easily titrated, rapid in onset, brief in duration, and provide sedation and analgesia without respiratory or hemodynamic compromise. Although many agents have been tried, no single drug fits this profile. The authors evaluated the comparative effectiveness and safety of propofol versus propofol/ketamine combination for procedural sedation using bispectral index monitoring for measuring depth of sedation. A prospective, randomized case series of patients undergoing procedural sedation for fracture manipulation was studied in a Level 1 trauma center emergency department. Patients were randomized to a propofol (P) group with a target dose of 0.5 to 1.5 mg/kg or a propofol/ketamine (P/K) group with a target dose of both ketamine and propofol of 0.75 mg/kg. Procedural success, bispectral index (BIS) score, adverse effects, recovery time, and vital signs were measured. Twenty-eight patients were enrolled. The P/K group experienced a smaller decline in systolic blood pressure (1.6% versus 12.5%) and BIS score at goal sedation (77 versus 61), a smaller difference between baseline and goal sedation BIS score (18.78 ± 10 versus 34.64 ± 11) and a lower mean propofol dose (92.5 ± 58 versus 177.27 ± 11 mg). No patient in either group experienced respiratory depression or required any intervention. The combination of propofol and ketamine provides an attractive combination for procedural sedation in the emergency department. Compared to propofol alone, "ketofol" results in less hypotension, better sedation, and enhanced patient comfort and safety.

To study the effect of blur adaptation on accommodative variability, accommodative responses and pupil diameters in myopes (n = 22) and emmetropes (n = 19) were continuously measured before, during, and after exposure to defocus blur. Accommodative and pupillary response measurements were made by an autorefractor during a monocular reading exercise. The text was presented on a computer screen at 33 cm viewing distance with a rapid serial visual presentation paradigm. After baseline testing and a 5-min rest, blur was induced by wearing either an optimally refractive lens, or a +1.0 DS or a +3.0 DS defocus lens. Responses were continuously measured during a 5-min period of adaptation. The lens was then removed, and measurements were again made during a 5-min post-adaptation period. After a second 5-min rest, a final post-adaptation period was measured. No significant change of baseline accommodative responses was found after the 5-min period of adaptation to the blurring lenses (p > 0.05). Compared to the pre-adaptation level, both refractive groups had similar and significant increases in accommodative variability right after blur adaptation to both defocus lenses. After the second rest period, the accommodative variability in both groups returned to the pre-adaptation level. The results indicate that blur adaptation has a short-term effect on the accommodative system to elevate instability of the accommodative response. Mechanisms underlying the increase in accommodative variability by blur adaptation and possible influences of the accommodation stability on myopia development were discussed.

We demonstrated a liquid lens whose focal length can be controlled by an actuator. The lens cell is composed of elastic membrane, planar glass plate, a periphery sealing ring, and a liquid with a fixed volume in the lens chamber. Part of the periphery sealing ring is excavated to form a hollow chamber which functions as a reservoir. This hollowed periphery is surrounded by an exterior rubber membrane. The shaft of an actuator is used to deform the elastic rubber. Squeezing the liquid contained in the reservoir into the lens chamber. Excess liquid in the lens chamber will push the lens membrane to outward, resulting in a lens shape change. Due to the compact structure and easy operation, this liquid lens has potential applications in zoom lenses, auto beam steering, and eyeglasses.

Sensitivity, selectivity, and speed are the most important analytical parameters in sensing chemical warfare agents (CWAs) and toxic industrial chemicals (TICs). For homeland security applications, where protection from chemical releases in public facilities such as at airports or subway stations is desired, selectivity becomes even more critical since false alarms can cause costly societal disruptions and trigger public panic. Currently, ion mobility spectrometry (IMS) has been widely deployed in the battlefield for CWA detection because of its excellent sensitivity and speed. However, IMS alone has suffered from low selectivity, especially for lower molecular weight TICs, which results in false alarms. This article describes an emerging technology that combines differential mobility spectrometry (DMS) and IMS in one package. In this configuration, the DMS serves as an ion filter for the IMS, prescreening targets from background with a separation mechanism that is complementary to the IMS. The tandem analyzer consists of a microfabricated DMS and two miniaturized IMS detectors arranged in an orthogonal configuration, enabling the detection of positive and negative ions simultaneously. The combination of these two technologies can significantly reduce false alarm rates. The tandem DMS–IMS provides a real-time, powerful technology to counter terrorist threats in infrastructure protection scenarios. Keywords: ion mobility spectrometry; differential mobility spectrometry; homeland security; chemical warfare agent; toxic industrial chemical; sensing and detection

The addition of multimedia resources in educational activities is increasingly common in our daily lives. In this paper we describe an experience of creation and publication of digital content using Web multiuser technology. Our intention is to run a participative design, which conceives the virtual environment as a space dedicated to the primary schools, and whose construction and modeling involves the educators and children. This content participative design is a practice encouraged by educators and software engineers as a way to achieve success in applications development. Finally, we present the developed virtual space.

OBJECTIVE. Conventional MDCT angiography uses a traditional peripheral IV approach for contrast injection; however, we describe our experience with a superior vena cava (SVC) catheter approach for coronary artery MDCT angiography as a potential means of decreasing iodinated contrast volume.CONCLUSION. Central SVC contrast injection can decrease the contrast volume by 50% while maintaining coronary attenuation similar to that of peripheral IV injection. This approach has potential in reducing the contrast volume on coronary MDCT angiography studies and therefore the risk of contrast-induced nephropathy in certain high-risk patients. Further studies with higher injection rates and faster scan acquisition are needed for defining a lower contrast volume threshold.

The food processing industry is Canada's second-largest manufacturing industry. It employed 226,579 people in 1986, and shipments were valued at CDN $47 billion, or 15 percent of the value of total manufactured output that year. More significantly, the food and beverage industries together ranked highest among all manufacturing industries in terms of value added, at CDN $15 billion or approximately 14 percent of total value added in Canadian manufacturing industries in 1986 (Statistics Canada). Given the high degree of competition in this industry in the United States, the history of “comfortable” competition in the food industry in Canada, and the significant contribution of this industry to the Canadian economy, it becomes important to look more carefully at how this industry has been and will be affected by the Canada-U.S. Free Trade Agreement (FTA).

A 22-month-old girl with a renal mass had multiple small pulmonary nodules on CT at her initial presentation. After biopsy and neoadjuvant chemotherapy, a Wilms tumor was resected and the pulmonary nodules were shown to have regressed on CT. Follow-up imaging 4 months after initial diagnosis demonstrated multiple new liver lesions and new pulmonary nodules with peripheral eosinophilia. Lung biopsy revealed granuloma formation with prominent eosinophils. The serum antibody titers for Toxocara canis were elevated. This case illustrates that toxocariasis should be considered as a rare differential diagnostic possibility for multiple liver lesions and multifocal peripheral pulmonary opacities in young children with Wilms tumor.

Slot waveguides allow for high optical confinement in a planar optical waveguide. Here we show a method for maintaining this high degree of confinement in slot waveguides with sharp bends. This high confinement can be achieved by using an asymmetric slot-based structure, where the mode in the bend remains localized in the slot region. We show that the relative power inside the slot can be as high as 28% for a 1 mum radius bend in an air-clad silicon waveguide.

The unlimited and pervasive use of the Internet by young people raises many concerns about child safety. What solutions are available and why aren't they being used?

A simple prescriptive model is developed to assess and improve economic relations between distributors and suppliers in a highly competitive market. Through a case study of the Canadian electrical distribution industry, it is shown that the present relations between distributors and their suppliers are very poorly structured. An alternative system of contractual relations is proposed which should improve long-run profitability of both suppliers and distributors.

The prospect of bilateral trade liberalization requires an understanding of contemporary work on the adjustment process. This survey of recent models by principally Canadian economists examines the methodological foundations of this research. It is found that the new trade adjustment models are just beginning to incorporate key elements of industrial organization effects such as intra-firm trade and non-tariff barriers to trade. More progress in these areas is required.

Introduction Although there are numerous brief reports concerning the metabolic effectiveness of human growth hormone,1-8the full details of a metabolic balance study in a young child have not been published. It is the purpose of this paper to report detailed results of a metabolic balance study carried out on a 3 10/12-year-old "hypopituitary" boy. To define what is meant by "hypopituitary" a case report describing the subject is included. The human growth hormone employed was obtained from Dr. Maurice Raben9and was administered intramuscularly in a dosage of 5 mg. per day. Report of Case The patient was a 310/12-year-old Japanese boy, referred because of gradual deceleration in linear growth.* At birth, his weight was 3,200 gm. and his length was 47 cm. The pregnancy was uncomplicated, but the delivery was notable in that a precipitous delivery was necessitated by a prolapse of the umbilical cord. There