Search

EIE proposed to accelerate pruned and compressed neural networks, exploiting weight sparsity, activation sparsity, and 4-bit weight-sharing in neural network accelerators. Since published in ISCA'16, it opened a new design space to accelerate pruned and sparse neural networks and spawned many algorithm-hardware co-designs for model compression and acceleration, both in academia and commercial AI chips. In retrospect, we review the background of this project, summarize the pros and cons, and discuss new opportunities where pruning, sparsity, and low precision can accelerate emerging deep learning workloads., Invited retrospective paper at ISCA 2023

Domain-specific accelerators are used in various computing systems ranging from edge devices to data centers. Coarse-grained reconfigurable arrays (CGRAs) represent an architectural midpoint between the flexibility of an FPGA and the efficiency of an ASIC and are a promising candidate for servicing multi-tasked workloads within an application domain. Unfortunately, scheduling multiple tasks onto a CGRA is challenging. CGRAs lack abstractions that capture hardware resources, leaving workload schedulers unable to reason about performance, energy, and utilization for different schedules. This work first proposes a CGRA architecture that can flexibly partition key resources, including the global buffer memory capacity, the global buffer memory bandwidth, and the compute resources. Partitioned resources serve as hardware abstractions that decouple compilation and resource allocation. The compiler uses these abstractions for coarse-grained resource mapping, and the scheduler uses them for flexible resource allocation at run time. We then propose two hardware mechanisms to support multi-task execution. A flexible-shape execution region increases the overall resource utilization by mapping multiple tasks with different resource requirements. Dynamic partial reconfiguration (DPR) enables a CGRA to update the hardware configuration as the scheduler makes decisions rapidly. We show that our abstraction can help automatic and efficient scheduling of multi-tasked workloads onto our target CGRA with high utilization, resulting in 1.05x-1.24x higher throughput and a 23-28% lower latency in a multi-tasked cloud workload and 60.8% reduced latency in an autonomous system workload when compared to a baseline CGRA running single tasks at a time.

Additional file 1. PRISMA-P 2015 Checklist

Additional file 2. Search strategies for ‘Venlafaxine or Mirtazapine for major depressive disorder’ Preliminary search strategies prepared 4 March 2022

Combination treatment with psychiatric medication and psychological therapy is generally considered to be more effective than psychotherapy alone for common mental health disorders, that is depression and some anxiety disorders, especially when the condition is more severe (National Collaborating Centre for Mental Health, 2010; NICE, 2011; Cuijpers et al., 2020) . However, this has not been tested in routine treatment settings. We will analyse a large dataset of routinely collected information from four Improving Access to Psychological Therapies (IAPT) services (National Health Service community and primary care mental health services) where patients are treated for common mental health disorders. Around half of IAPT patients are prescribed a psychiatric medication when initially assessed by services. We will analyse the effect of combination treatment versus psychological therapy alone on symptom score change and a variety of clinical outcomes, whilst controlling for baseline severity of condition and other confounders.

While coarse-grained reconfigurable arrays (CGRAs) have emerged as promising programmable accelerator architectures, pipelining applications running on CGRAs is required to ensure high maximum clock frequencies. Current CGRA compilers either lack pipelining techniques resulting in low performance or perform exhaustive pipelining resulting in high energy and resource consumption. We introduce Cascade, an application pipelining toolkit for CGRAs, including a CGRA application frequency model, automated pipelining techniques for CGRA application compilers that work with both dense and sparse applications, and hardware optimizations for improving application frequency. Cascade enables 7 - 34x lower critical path delays and 7 - 190x lower EDP across a variety of dense image processing and machine learning workloads, and 2 - 4.4x lower critical path delays and 1.5 - 4.2x lower EDP on sparse workloads, compared to a compiler without pipelining., Comment: Preprint version

Achieving high code reuse in physical design flows is challenging but increasingly necessary to build complex systems. Unfortunately, existing approaches based on parameterized Tcl generators support very limited reuse and struggle to preserve reusable code as designers customize flows for specific designs and technologies. We present a vision and framework based on modular flow generators that encapsulates coarse-grain and fine-grain reusable code in modular nodes and assembles them into complete flows. The key feature is a flow consistency and instrumentation layer embedded in Python, which supports mechanisms for rapid and early feedback on inconsistent composition. The approach gradually types the Tcl language and allows both automatic and user-annotated static assertion checks. We evaluate the design flows of successive generations of silicon prototypes designed in TSMC16, TSMC28, TSMC40, SKY130, and IBM180 technologies, showing how our approach can enable significant code reuse in future flows.

The increasing complexity of modern configurable systems makes it critical to improve the level of automation in the process of system configuration. Such automation can also improve the agility of the development cycle, allowing for rapid and automated integration of decoupled workflows. In this paper, we present a new framework for automated configuration of systems representable as state machines. The framework leverages model checking and satisfiability modulo theories (SMT) and can be applied to any application domain representable using SMT formulas. Our approach can also be applied modularly, improving its scalability. Furthermore, we show how optimization can be used to produce configurations that are best according to some metric and also more likely to be understandable to humans. We showcase this framework and its flexibility by using it to configure a CGRA memory tile for various image processing applications.

The architecture of a coarse-grained reconfigurable array (CGRA) processing element (PE) has a significant effect on the performance and energy efficiency of an application running on the CGRA. This paper presents an automated approach for generating specialized PE architectures for an application or an application domain. Frequent subgraphs mined from a set of applications are merged to form a PE architecture specialized to that application domain. For the image processing and machine learning domains, we generate specialized PEs that are up to 10.5x more energy efficient and consume 9.1x less area than a baseline PE.

Image processing and machine learning applications benefit tremendously from hardware acceleration, but existing compilers target either FPGAs, which sacrifice power and performance for flexible hardware, or ASICs, which rapidly become obsolete as applications change. Programmable domain-specific accelerators have emerged as a promising middle-ground between these two extremes, but such architectures have traditionally been difficult compiler targets. The main obstacle is that these accelerators often use a different memory abstraction than CPUs and GPUs: push memories that send a data stream from one computation kernel to other kernels, possibly reordered. To address the compilation challenges caused by push memories, we propose that the representation of memory in the middle and backend of the compiler be altered to combine storage with address generation and control logic in a single structure -- a unified buffer. We show that this compiler abstraction can be implemented efficiently on a programmable accelerator, and design a memory mapping algorithm that combines polyhedral analysis and software vectorization techniques to target our accelerator. Our evaluation shows that the compiler supports programmability while maintaining high performance. It can compile a wide range of image processing and machine learning applications to our accelerator with 4.7x better runtime and 4.3x better energy-efficiency as compared to an FPGA.

While hardware generators have drastically improved design productivity, they have introduced new challenges for the task of verification. To effectively cover the functionality of a sophisticated generator, verification engineers require tools that provide the flexibility of metaprogramming. However, flexibility alone is not enough; components must also be portable in order to encourage the proliferation of verification libraries as well as enable new methodologies. This paper introduces fault, a Python embedded hardware verification language that aims to empower design teams to realize the full potential of generators., CAV 2020: 32nd International Conference on Computer-Aided Verification

Oxidized lipids derived from omega-6 (n-6) and omega-3 (n-3) polyunsaturated fatty acids, collectively known as oxylipins, are bioactive signaling molecules that play diverse roles in human health and disease. Supplementation with n-3 docosahexaenoic acid (DHA) during pregnancy has been reported to decrease the risk of preterm birth in singleton pregnancies, which may be due to effects of DHA supplementation on oxylipins or their precursor n-6 and n-3 fatty acids. There is only limited understanding of the levels and trajectory of changes in plasma oxylipins during pregnancy, effects of DHA supplementation on oxylipins and unesterified fatty acids, and whether and how oxylipins and their unesterified precursor fatty acids influence preterm birth. In the present study we used liquid chromatography-tandem mass spectrometry to profile oxylipins and their precursor fatty acids in the unesterified pool using plasma samples collected from a subset of pregnant Australian women who participated in the ORIP (Omega-3 fats to Reduce the Incidence of Prematurity) study. ORIP is a large randomized controlled trial testing whether daily supplementation with n-3 DHA can reduce the incidence of early preterm birth compared to control. Plasma was collected at study entry (≈pregnancy week 14) and again at ≈week 24, in a subgroup of 48 ORIP participants-12 cases with spontaneous preterm (

Supplemental material, Figure_1 for The ‘patient voice’: patients who experience antidepressant withdrawal symptoms are often dismissed, or misdiagnosed with relapse, or a new medical condition by Anne Guy, Marion Brown, Stevie Lewis and Mark Horowitz in Therapeutic Advances in Psychopharmacology

Supplemental material, Figure_2 for The ‘patient voice’: patients who experience antidepressant withdrawal symptoms are often dismissed, or misdiagnosed with relapse, or a new medical condition by Anne Guy, Marion Brown, Stevie Lewis and Mark Horowitz in Therapeutic Advances in Psychopharmacology

Highlights • Inflammation and reduced neurogenesis are associated with the pathophysiology of depression. • IL-1β decreased neurogenesis in human hippocampal progenitor cells. • EPA, DHA, sertraline and venlafaxine prevented the IL-1β-induced reduction in neurogenesis. • EPA and DHA reversed the IL-1β-induced increase in kynurenine levels. • EPA, DHA, sertraline and venlafaxine decreased the upregulation of IDO and KMO mRNA., Both increased inflammation and reduced neurogenesis have been associated with the pathophysiology of major depression. We have previously described how interleukin-1 (IL-1) β, a pro-inflammatory cytokine increased in depressed patients, decreases neurogenesis in human hippocampal progenitor cells. Here, using the same human in vitro model, we show how omega-3 (ω-3) polyunsaturated fatty acids and conventional antidepressants reverse this reduction in neurogenesis, while differentially affecting the kynurenine pathway. We allowed neural cells to proliferate for 3 days and further differentiate for 7 days in the presence of IL-1β (10 ng/ml) and either the selective serotonin reuptake inhibitor sertraline (1 µM), the serotonin and norepinephrine reuptake inhibitor venlafaxine (1 µM), or the ω-3 fatty acids eicosapentaenoic acid (EPA, 10 µM) or docosahexaenoic acid (DHA, 10 µM). Co-incubation with each of these compounds reversed the IL-1β-induced reduction in neurogenesis (DCX- and MAP2-positive neurons), indicative of a protective effect. Moreover, EPA and DHA also reversed the IL-1β-induced increase in kynurenine, as well as mRNA levels of indolamine-2,3-dioxygenase (IDO); while DHA and sertraline reverted the IL-1β-induced increase in quinolinic acid and mRNA levels of kynurenine 3-monooxygenase (KMO). Our results show common effects of monoaminergic antidepressants and ω-3 fatty acids on the reduction of neurogenesis caused by IL-1β, but acting through both common and different kynurenine pathway-related mechanisms. Further characterization of their individual properties will be of benefit towards improving a future personalized medicine approach.

BackgroundOxidized derivatives of polyunsaturated fatty acids, collectively known as oxylipins, are labile bioactive mediators with diverse roles in human physiology and pathology. Oxylipins are increasingly being measured in plasma collected in clinical studies to investigate biological mechanisms and as pharmacodynamic biomarkers for nutrient-based and drug-based interventions. Whole blood is generally stored either on ice or at room temperature prior to processing. However, the potential impacts of delays in processing, and of temperature prior to processing, on oxylipin concentrations are incompletely understood.ObjectiveTo evaluate the effects of delayed processing of blood samples in a timeframe that is typical of a clinical laboratory setting, using typical storage temperatures, on concentrations of representative unesterified oxylipins measured by liquid chromatography-tandem mass spectrometry.DesignWhole blood (drawn on three separate occasions from a single person) was collected into 5 mL purple-top potassium-EDTA tubes and stored for 0, 10, 20, 30, 60 or 120min at room temperature or on wet ice, followed by centrifugation at 4 °C for 10min with plasma collection. Each sample was run in duplicate, therefore there were six tubes and up to six data points at each time point for each oxylipin at each condition (ice/room temperature). Representative oxylipins derived from arachidonic acid, docosahexaenoic acid, and linoleic acid were quantified by liquid chromatography tandem mass spectrometry. Longitudinal models were used to estimate differences between temperature groups 2h after blood draw.ResultsWe found that most oxylipins measured in human plasma in traditional potassium-EDTA tubes are reasonably stable when stored on ice for up to 2h prior to processing, with little evidence of auto-oxidation in either condition. By contrast, in whole blood stored at room temperature, substantial time-dependent increases in the 12-lipoxygenase-derived (12-HETE, 14-HDHA) and platelet-derived (thromboxane B2) oxylipins were observed.ConclusionThese findings suggest that certain plasma oxylipins can be measured with reasonable accuracy despite delayed processing for up to 2h when blood is stored on ice prior to centrifugation. 12-Lipoxygenase- and platelet-derived oxylipins may be particularly sensitive to post-collection artifact with delayed processing at room temperature. Future studies are needed to determine impacts of duration and temperature of centrifugation on oxylipin concentrations.

Developing and middle-income countries increasingly empha-size higher education and entrepreneurship in their long-term develop-ment strategy. Our work focuses on the influence of higher education institutions (HEIs) on startup ecosystems in Brazil, an emerging economy. First, we describe regional variability in entrepreneurial network characteristics. Then we examine the influence of elite HEIs in economic hubs on entrepreneur networks. Second, we investigate the influence ofthe academic trajectories of startup founders, including their courses of study and HEIs of origin, on the fundraising capacity of startups. Given the growing capability of social media databases such as Crunchbase and LinkedIn to provide startup and individual-level data, we draw on computational methods to mine data for social network analysis. We find that HEI quality and the maturity of the ecosystem influence startup success. Our network analysis illustrates that elite HEIs have powerful influences on local entrepreneur ecosystems. Surprisingly, while the most nationally prestigious HEIs in the South and Southeast have the longest geographical reach, their network influence still remains local.

Real-time CNN-based object detection models for applications like surveillance can achieve high accuracy but are computationally expensive. Recent works have shown 10 to 100x reduction in computation cost for inference by using domain-specific networks. However, prior works have focused on inference only. If the domain model requires frequent retraining, training costs can pose a significant bottleneck. To address this, we propose Dataset Culling: a pipeline to reduce the size of the dataset for training, based on the prediction difficulty. Images that are easy to classify are filtered out since they contribute little to improving the accuracy. The difficulty is measured using our proposed confidence loss metric with little computational overhead. Dataset Culling is extended to optimize the image resolution to further improve training and inference costs. We develop fixed-angle, long-duration video datasets across several domains, and we show that the dataset size can be culled by a factor of 300x to reduce the total training time by 47x with no accuracy loss or even with slight improvement. Codes are available: https://github.com/kentaroy47/DatasetCulling, Comment: accepted to IEEE ICIP 2019. 5 pages

We propose an end-to-end framework for training domain specific models (DSMs) to obtain both high accuracy and computational efficiency for object detection tasks. DSMs are trained with distillation \cite{hinton2015distilling} and focus on achieving high accuracy at a limited domain (e.g. fixed view of an intersection). We argue that DSMs can capture essential features well even with a small model size, enabling higher accuracy and efficiency than traditional techniques. In addition, we improve the training efficiency by reducing the dataset size by culling easy to classify images from the training set. For the limited domain, we observed that compact DSMs significantly surpass the accuracy of COCO trained models of the same size. By training on a compact dataset, we show that with an accuracy drop of only 3.6\%, the training time can be reduced by 93\%. The codes are uploaded in https://github.com/kentaroy47/training-domain-specific-models.

Both increased inflammation and reduced neurogenesis have been associated with the pathophysiology of major depression. We have previously described how interleukin-1 (IL-1) β a pro-inflammatory cytokine increased in depressed patients, decreases neurogenesis in human hippocampal progenitor cells. Here, using the same human in vitro model, we show how omega-3 (ω-3) polyunsaturated fatty acids and conventional antidepressants reverse this reduction in neurogenesis, while differentially affecting the kynurenine pathway. We allowed neural cells to proliferate for 3 days and further differentiate for 7 days in the presence of IL-1β (10 ng/ml) and either the selective serotonin reuptake inhibitor sertraline (1 µM), the serotonin and norepinephrine reuptake inhibitor venlafaxine (1 µM), or the ω-3 fatty acids eicosapentaenoic acid (EPA, 10 µM) or docosahexaenoic acid (DHA, 10 µM). Co-incubation with each of these compounds reversed the IL-1β-induced reduction in neurogenesis (DCX- and MAP2-positive neurons), indicative of a protective effect. Moreover, EPA and DHA also reversed the IL-1β-induced increase in kynurenine, as well as mRNA levels of indolamine-2,3-dioxygenase (IDO); while DHA and sertraline reverted the IL-1β-induced increase in quinolinic acid and mRNA levels of kynurenine 3-monooxygenase (KMO). Our results show common effects of monoaminergic antidepressants and ω-3 fatty acids on the reduction of neurogenesis caused by IL-1β, but acting through both common and different kynurenine pathway-related mechanisms. Further characterization of their individual properties will be of benefit towards improving a future personalized medicine approach.

Data analysis pipeline of growth videos. Features from training images are used to extract clusters from each frame. Clusters in adjacent images are then associated with each other. During this process, we account for the possibility of cell growth and division. Temporal association is used to construct cell lineages from which growth statistics are generated. (PDF 77Â kb)

Synechocystis cells are ellipsoidal and most lineages grow in microfluidic chambers. (a) Two images representing the same field of view of a microfluidic chamber at an early time point (green) and 40Â h later (maroon). Growth is evident for most (if not all) Synechocystis lineages. (b) Bright-field image of Synechocystis cells imaged on a glass slide, illustrating their ellipsoidal shape. (c) Quantification of the ellipticity of cells and cell doublets in (b), defined as the ratio of minor axis length to major axis length, demonstrates that most cells are not spherical. (PDF 2 MB)

Microfluidic system setup. (a) Schematic of the microfluidic cell culture chip. This device is a two-layer push-up device. Blue lines show outline of flow channels containing cell culture chambers and reagent inlets. Yellow lines outline the control layer containing push-up valves, multiplexer, and a peristaltic pump. Right: image of an actual cell culture chip. The dimensions of each chamber are 1.85 × 900 mm, and the height is 25 μm. Due to the higher magnification (20×) used to image bacteria, only the center portion (0.3 × 0.3 mm) of the chip is monitored. (b) Illustration of control and operation of the microfluidic cell culture chip. The device is surrounded by an environmental chamber and secured on an automated stage of a Leica DMI 6000 microscope. Each control channel is connected with an independent pressure reservoir located above the microscope. All associated equipment, including CO2 and temperature control, is connected to a computer and the operation of each component is controlled automatically via custom MATLAB software. (c) The cell culture chip contains 16 inputs that are routed to a central pump and distributed to 96 chambers. Each chamber is independently addressable. (PDF 6 MB)

Robust sister-sister generation time correlation under continuous illumination. (a, b) The generation times of sister cells were highly correlated for cells that divided in the first 30Â h of the experiment (a) and in the last 30Â h (b). The similarity between the two correlation coefficients demonstrates that the correlation between sister-cell generation times is not an artifact of temporal variations in cell growth. (c) There was no significant correlation between the generation times of mother and daughter cells. (PDF 340Â kb)

Probabilistic image analysis pipeline. (aâ d) Outline features extracted from the training dataset. These distributions are used to generate probability distributions for classifying single cells, doublets, and clusters of more than two cells. (a) Distributions of cluster areas in the training set corresponding to single cells (blue), doublets (red), or more than two cells (green). (b) Distributions of cluster circularity (Eq. 1 in Additional file 4). A perfect circle and a line have circularities of 1 and 0, respectively. (c) Distributions of cluster eccentricity, defined as the ratio of the distance between the two foci of an ellipse to the major axis. A perfect circle and a line have eccentricities of 0 and 1, respectively. (d) Distribution of distance between cell centers for doublets. (e) Cluster correlations between adjacent images in time. Distance metric (Eq. 3 in Additional file 4) guarantees that closer clusters are more correlated with each other. White circles show features such as movement and division between adjacent temporal frames. (PDF 499Â kb)

Method of cell volume extraction from bright-field images. Since the image of each cell represents its two-dimensional projection, we assume that each cell is rotationally symmetric with respect to its major axis. Based on this assumption, to compute the volume of each cell (or doublet), we first extract the orientation of its major axis. Then, we add up the volumes of circular disks perpendicular to the major axis with thicknesses of one pixel. (PDF 354Â kb)

Cells were slightly smaller when grown under light-dark cycles. (a, b) Left: birth (a) and division (b) volumes as a function of time for cells grown under continuous illumination (blue) and light-dark cycles (orange). Right: volume distributions. There is no birth-volume data in the dark since cell division did not occur in the dark (Fig. 3d). (PDF 400 kb)

Data analysis pipeline of growth videos. Features from training images are used to extract clusters from each frame. Clusters in adjacent images are then associated with each other. During this process, we account for the possibility of cell growth and division. Temporal association is used to construct cell lineages from which growth statistics are generated. (PDF 77Â kb)

Representative Synechocystis single-cell growth curves. (a) A representative single-cell growth curve under continuous illumination, showing birth volume (V b ), division volume (V d ), and generation time (T). (b) A representative single-cell growth curve during light-dark growth. To compute generation time, the 12-h dark periods were removed. (PDF 347Â kb)

Method of cell volume extraction from bright-field images. Since the image of each cell represents its two-dimensional projection, we assume that each cell is rotationally symmetric with respect to its major axis. Based on this assumption, to compute the volume of each cell (or doublet), we first extract the orientation of its major axis. Then, we add up the volumes of circular disks perpendicular to the major axis with thicknesses of one pixel. (PDF 354Â kb)

Microfluidic cell culture chip fabrication process. To make the cell culture chip, photoresist was used to build structures on silicon wafers, using standard photolithography to create molds. PDMS was poured over the flow-layer mold and spun on the control-layer mold. Both molds were then bonded together on a glass slide. (PDF 286Â kb)

Synechocystis cells are ellipsoidal and most lineages grow in microfluidic chambers. (a) Two images representing the same field of view of a microfluidic chamber at an early time point (green) and 40Â h later (maroon). Growth is evident for most (if not all) Synechocystis lineages. (b) Bright-field image of Synechocystis cells imaged on a glass slide, illustrating their ellipsoidal shape. (c) Quantification of the ellipticity of cells and cell doublets in (b), defined as the ratio of minor axis length to major axis length, demonstrates that most cells are not spherical. (PDF 2 MB)

Additional Methods and Tables S1â S3. Table S1. Summary of simulations of cells grown under continuous illumination. Table S2. Summary of simulations of cells grown under 12-h light-dark illumination cycles. Table S3. Parameters used for cell-size regulation simulations. (DOCX 56Â kb)

Probabilistic image analysis pipeline. (aâ d) Outline features extracted from the training dataset. These distributions are used to generate probability distributions for classifying single cells, doublets, and clusters of more than two cells. (a) Distributions of cluster areas in the training set corresponding to single cells (blue), doublets (red), or more than two cells (green). (b) Distributions of cluster circularity (Eq. 1 in Additional file 4). A perfect circle and a line have circularities of 1 and 0, respectively. (c) Distributions of cluster eccentricity, defined as the ratio of the distance between the two foci of an ellipse to the major axis. A perfect circle and a line have eccentricities of 0 and 1, respectively. (d) Distribution of distance between cell centers for doublets. (e) Cluster correlations between adjacent images in time. Distance metric (Eq. 3 in Additional file 4) guarantees that closer clusters are more correlated with each other. White circles show features such as movement and division between adjacent temporal frames. (PDF 499Â kb)

Growth behavior is similar across all chambers in light-dark cycle experiment. (a) Total growth in different chambers under light-dark cycles. Substantial growth is observed during illuminated periods across all microfluidic chambers. In the dark, minimal growth is detected. (b) Residual errors (gray, with mean shown in black) of exponential fits to lineage growth curves during the illumination periods L1 and L2 of Fig. 3a. (PDF 366 kb)

FPMax implements four FPUs optimized for latency or throughput workloads in two precisions, fabricated in 28nm UTBB FDSOI. Each unit's parameters, e.g pipeline stages, booth encoding etc., were optimized to yield 1.42ns latency at 110GLOPS/W (SP) and 1.39ns latency at 36GFLOPS/W (DP). At 100% activity, body-bias control improves the energy efficiency by about 20%; at 10% activity this saving is almost 2x. Keywords: FPU, energy efficiency, hardware generator, SOI

Convolutional Neural Networks (CNNs) are the state of the art solution for many computer vision problems, and many researchers have explored optimized implementations. Most implementations heuristically block the computation to deal with the large data sizes and high data reuse of CNNs. This paper explores how to block CNN computations for memory locality by creating an analytical model for CNN-like loop nests. Using this model we automatically derive optimized blockings for common networks that improve the energy efficiency of custom hardware implementations by up to an order of magnitude. Compared to traditional CNN CPU implementations based on highly-tuned, hand-optimized BLAS libraries,our x86 programs implementing the optimal blocking reduce the number of memory accesses by up to 90%.

Emerging data suggest that megakaryocytes (MKs) play a significant role in skeletal homeostasis. Indeed, osteosclerosis observed in several MK-related disorders may be a result of increased numbers of MKs. In support of this idea, we have previously demonstrated that MKs increase osteoblast (OB) proliferation by a direct cell-cell contact mechanism and that MKs also inhibit osteoclast (OC) formation. As MKs and OCs are derived from the same hematopoietic precursor, in these osteoclastogenesis studies we examined the role of the main MK growth factor, thrombopoietin (TPO) on OC formation and bone resorption. Here we show that TPO directly increases OC formation and differentiation in vitro. Specifically, we demonstrate the TPO receptor (c-mpl or CD110) is expressed on cells of the OC lineage, c-mpl is required for TPO to enhance OC formation in vitro, and TPO activates the mitogen-activated protein kinases, Janus kinase/signal transducer and activator of transcription, and nuclear factor-kappaB signaling pathways, but does not activate the PI3K/AKT pathway. Further, we found TPO enhances OC resorption in CD14+CD110+ human OC progenitors derived from peripheral blood mononuclear cells, and further separating OC progenitors based on CD110 expression enriches for mature OC development. The regulation of OCs by TPO highlights a novel therapeutic target for bone loss diseases and may be important to consider in the numerous hematologic disorders associated with alterations in TPO/c-mpl signaling as well as in patients suffering from bone disorders.

GATA-1(low/low) mice have an increase in megakaryocytes (MKs) and trabecular bone. The latter is thought to result from MKs directly stimulating osteoblastic bone formation while simultaneously inhibiting osteoclastogenesis. Osteoprotegerin (OPG) is known to inhibit osteoclastogenesis and OPG(-/-) mice have reduced trabecular and cortical bone due to increased osteoclastogenesis. Interestingly, GATA-1(low/low) mice have increased OPG levels. Here, we sought to determine whether GATA-1 knockdown in OPG(-/-) mice could rescue the observed osteoporotic bone phenotype. GATA-1(low/low) mice were bred with OPG(-/-) mice and bone phenotype assessed. GATA-1(low/low) × OPG(-/-) mice have increased cortical bone porosity, similar to OPG(-/-) mice. Both OPG(-/-) and GATA-1(low/low) × OPG(-/-) mice, were found to have increased osteoclasts localized to cortical bone, possibly producing the observed elevated porosity. Biomechanical assessment indicates that OPG(-/-) and GATA-1(low/low) × OPG(-/-) femurs are weaker and less stiff than C57BL/6 or GATA-1(low/low) femurs. Notably, GATA-1(low/low) × OPG(-/-) mice had trabecular bone parameters that were not different from C57BL/6 values, suggesting that GATA-1 deficiency can partially rescue the trabecular bone loss observed with OPG deficiency. The fact that GATA-1 deficiency appears to be able to partially rescue the trabecular, but not the cortical bone phenotype suggests that MKs can locally enhance trabecular bone volume, but that MK secreted factors cannot access cortical bone sufficiently to inhibit osteoclastogenesis or that OPG itself is required to inhibit osteoclastogenesis in cortical bone.

Recent studies suggest that megakaryocytes (MKs) may play a significant role in skeletal homeostasis, as evident by the occurrence of osteosclerosis in multiple MK related diseases (Thiele, et al., 1999, Lennert, et al., 1975, Chagraoui, et al., 2006). We previously reported a novel interaction whereby MKs enhanced proliferation of osteoblast lineage/osteoprogenitor cells (OBs) by a mechanism requiring direct cell-cell contact. However, the signal transduction pathways and the downstream effector molecules involved in this process have not been characterized. Here we show that MKs contact with OBs, via beta1 integrin, activate the p38/MAPKAPK2/p90RSK kinase cascade in the bone cells, which causes Mdm2 to neutralizes p53/Rb-mediated check point and allows progression through the G1/S. Interestingly, activation of MAPK (ERK1/2) and AKT, collateral pathways that regulate the cell cycle, remained unchanged with MK stimulation of OBs. The MK-to-OB signaling ultimately results in significant increases in the expression of c-fos and cyclin A, necessary for sustaining the OB proliferation. Overall, our findings show that OBs respond to the presence of MKs, in part, via an integrin-mediated signaling mechanism, activating a novel response axis that de-represses cell cycle activity. Understanding the mechanisms by which MKs enhance OB proliferation will facilitate the development of novel anabolic therapies to treat bone loss associated with osteoporosis and other bone-related diseases.

When setting rates, many utilities use rate surveys - regional compilations of utilities' rates - to gauge a fair price increase. However, each utility has a unique set of factors that affect its rate, so simple comparisons between two utility rates may lead to the wrong conclusion. This thesis describes regression models which provide better comparisons by incorporating factors that influence rates. Two types of bills - water only and combined water and sewer - are modeled at four consumption levels: 3000, 6000, 9000, and 12000 gallons per month. The models use the data from all the utilities in the sample to provide an estimated average bill, with a 95% confidence interval, for each utility. Then, each utility can compare its actual bill with this estimate. The models also show that high bills (both types) are associated with source water, recent rate changes, large grants, and large connection fees at most consumption levels.

In recent years there has been increasing interest in using cryo TEM tomography to study cells in close to their "native" environment. One limitation of this technique is the relatively low signal to noise ratio in each of the TEM images, since the total electron dose through the sample must be constrained to limit structure damage to the cell. Even with gold markers added to the sample, robust automatic alignment of the TEM slice data for reconstruction remains difficult. We have tried to address this problem by leveraging recent work in probabilistic analysis, and have constructed a prototype alignment system using Markov random fields (MRF s) for alignment, and robust optimization methods for projective model estimation. With markers, there are three basic steps required to align the TEM dataset: marker feature identification, correspondence and tracking of these features throughout the image set, and projective model estimation from these feature tracks. In our framework, features are extracted initially using standard template matching techniques like cross correlation. Feature correspondence and tracking is accomplished by constructing a Markov random field (MRF) probabilistic model where contour labels are random variables which take on values of candidate marker feature locations. We use mutual information and the relative geometric positions to estimate a priori marker correspondence probabilities between two images. An approximate probabilistic inference technique called loopy belief propagation (LBP) is then used to calculate the maximum a posteriori assignment of features to contours in the image set. In this technique, rather than a joint distribution (whose complexity is exponential in the number of random variables), a collection of singleton and pairwise distributions is maintained in a special data structure. This data structure contains cycles, and is called a cluster graph. The a priori estimates for these distributions (initial beliefs) are refined by belief propagation, until they converge to roughly the true pairwise distributions (final beliefs). The correspondences of candidate markers to contours are taken directly from these beliefs. Errors in the correspondence are possible due to feature location mistakes as well as inaccurate inference results. Therefore, the projective model estimation uses a robust fitting method as opposed to least squares (the traditionally applied fitting) and is tolerant to outliers. Once we have an estimate of the projective model, the model is iterated using expectation maximization (EM) to re-estimate perceived outliers with improved reprojection data from the current model. This iteration is performed as many times as necessary before a stopping criterion is satisfied, but in our example a small number of iterations is needed (often only one).This robust framework has allowed us to fully automatically recover dozens of contours (both complete and piecewise) with subpixel accuracy from several challenging cryo datasets of bacteria Caulobacter crescentus. The results were used to create 3D reconstructions comparable to results previously obtainable only by extensive manual intervention.

We present a color calibration pipeline for large camera arrays. We assume static lighting conditions for each camera, such as studio lighting or a stationary array outdoors. We also assume we can place a planar calibration target so it is visible from every camera. Our goal is uniform camera color responses, not absolute color accuracy, so we match the cameras to each other instead of to a color standard. We first iteratively adjust the color channel gains and offsets for each camera to make their responses as similar as possible. This step white balances the cameras, and for studio applications, ensures that the range of intensities in the scene are mapped to the usable output range of the cameras. Residual errors are then calibrated in post-processing. We present results calibrating an array of 100 CMOS image sensors in different physical configurations, including closely or widely spaced cameras with overlapping fields of views, and tightly packed cameras with non-overlapping fields of view. The process is entirely automatic, and the camera configuration runs in less than five minutes on the 100 camera array.Pre-2018 CSE ID: CS2005-0821

This paper presents a camera that samples the 4D light field on its sensor in a single photographic exposure. This is achieved by inserting a microlens array between the sensor and main lens, creating a plenoptic camera. Each microlens measures not just the total amount of light deposited at that location, but how much light arrives along each ray. By resorting the measured rays of light to where they would have terminated in slightly different, synthetic cameras, we can compute sharp photographs focused at different depths. We show that a linear increase in the resolution of images under each microlens results in a linear increase in the sharpness of the refocused photographs. This property allows us to extend the depth of field of the camera without reducing the aperture, enabling shorter exposures and lower image noise. Especially in the macrophotography regime, we demonstrate that we can also compute synthetic photographs from a range of different viewpoints. These capabilities argue for a different strategy in designing photographic imaging systems. To the photographer, the plenoptic camera operates exactly like an ordinary hand-held camera. We have used our prototype to take hundreds of light field photographs, and we present examples of portraits, high-speed action and macro close-ups.

The objective is to design and build a small, high-bandwidth switch., Comment: 13 pages, 10 figures

Two widely recognized economic theories attempt to explain the process of development in an interregional context. Trade theory (and traditional neoclassical growth theory in general) posits that economic growth is both the necessary and sufficient condition for development of a less developed region. The theory of unequal exchange, on the other hand, contends that while economic growth is necessary to the development of a region, it is not a sufficient condition to bring about true development, defined as increases in the overall welfare of the region's population. This article attempts to break down the determinants of wages into an economic growth component and an economic development component as suggested by these theories. It then shows the importance of the economic development component in explaining cross-state wage differentials. The state of North Carolina has attempted to further its development through growth-related policies. The analysis found herein suggests the need for a reassessment of North Carolina's existing development policies.