One Proxy Device Is Enough for Hardware-Aware Neural Architecture Search

Convolutional neural networks (CNNs) are used in numerous real-world applications such as vision-based autonomous driving and video content analysis. To run CNN inference on various target devices, hardware-aware neural architecture search (NAS) is crucial. A key requirement of efficient hardware-aware NAS is the fast evaluation of inference latencies in order to rank different architectures. While building a latency predictor for each target device has been commonly used in state of the art, this is a very time-consuming process, lacking scalability in the presence of extremely diverse devices. In this work, we address the scalability challenge by exploiting latency monotonicity --- the architecture latency rankings on different devices are often correlated. When strong latency monotonicity exists, we can re-use architectures searched for one proxy device on new target devices, without losing optimality. In the absence of strong latency monotonicity, we propose an efficient proxy adaptation technique to significantly boost the latency monotonicity. Finally, we validate our approach and conduct experiments with devices of different platforms on multiple mainstream search spaces, including MobileNet-V2, MobileNet-V3, NAS-Bench-201, ProxylessNAS and FBNet. Our results highlight that, by using just one proxy device, we can find almost the same Pareto-optimal architectures as the existing per-device NAS, while avoiding the prohibitive cost of building a latency predictor for each device.

U.S-Based Community Oncologists/Hematologists' Perceptions Regarding Future Use of Gene Therapies for Patients with Hematological Disorders

Introduction Gene therapies, defined as the introduction, removal, or change in genetic material into a patient's cells to treat a specific disease, represent a significant advance in medicine, with the potential to cure or significantly improve outcomes of various benign and malignant hematologic disorders (American Society of Gene Therapy). Gene therapies have received approvals for subsets of patients with spinal muscular atrophy or retinal dystrophy, and almost a thousand studies are actively recruiting patients for various gene therapy trials. Aside from CAR-T therapies, several gene therapies are in advanced clinical development in the U.S. for disorders such as hemophilia, hemoglobinopathies, and congenital immunodeficiency syndromes. Results of many such trials were presented at the 2020 ASH annual meeting. While approvals for some gene therapy products are expected in the near future, the complexity of treatment, knowledge gaps among providers, or barriers with accessibility or cost may limit the integration of these treatments into routine clinical practice. Therefore, the present study surveyed U.S.-based community oncologists/hematologists (cO/H) to evaluate their perceptions of the utility of gene therapies and barriers to adoption or integration into clinical practice. Methods Between February and April 2021, cO/H from across the U.S. were invited to complete a web-based survey about gene therapies. Physician demographics and practice characteristics were also captured in the survey. Responses were aggregated and analyzed using descriptive statistics. Results A total of 369 cO/H completed the survey; 36% identified as a medical oncologist and 63% as a hematologist/oncologist. cO/H had an average of 19 years of clinical experience and spent an average of 86% of their working time in direct patient care and saw an average of 17 patients per day on clinic days. Half of cO/H stated that they were not aware of recent data of gene therapies for adult hematology/oncology indications (35% "not very aware"; 15% "not at all aware"). When asked to report the number of approved gene therapy products (excluding CAR-T products) in the U.S. in early 2021, 27% of participants reported 0, 24% reported 1, and 20% reported 2 products were currently available. Regarding gene therapy use for adult hematologic/oncologic indications in the next 2 years, 53% of cO/H reported that they expect gene therapies to mostly be administered and managed by academic centers to which they will refer their patients; 27% reported that indications will be limited and unlikely to affect their practice (Table 1). cO/H perceived cost as the greatest barrier to adopting gene therapies into their clinical practice; specifically, cO/H cited cost limitations by payers (49%), the prohibitive cost to patients (46%), and the prohibitive cost to practices/hospitals (37%). Other barriers to adoption were limited real-world efficacy data (18%) and long-term complications (13%). Most cO/H reported a moderate (39%) or high (43%) comfort level with prescribing a gene therapy for adult hematology/oncology indications if it were reimbursed. Conclusions Many cO/H were not aware of recent data of gene therapy products, with over half expecting to refer their patients to large academic centers for gene therapies. While most cO/H would be comfortable prescribing gene therapies for their patients, cost was perceived as prohibitive. This information can inform various stakeholders, including patients, advocacy groups, pharmaceutical manufacturers, payers, and professional societies, in laying the foundations for gene therapy products in hematology. Future work should focus on enhancing education of gene therapy products to community providers as well as identifying support programs that lessen the burden of cost. Figure 1 Figure 1. Disclosures Gajra: Cardinal Health: Current Employment, Current equity holder in publicly-traded company. Fortier: Cardinal Health: Current Employment. Jeune-Smith: Cardinal Health: Current Employment. Feinberg: Cardinal Health: Current Employment.

Determining the Composition of a Mixed Material with Synthetic Data

Determining the composition of a mixed material is an open problem that has attracted the interest of researchers in many fields. In our recent work, we proposed a novel approach to determine the composition of a mixed material using convolutional neural networks (CNNs). In machine learning, a model “learns” a specific task for which it is designed through data. Hence, obtaining a dataset of mixed materials is required to develop CNNs for the task of estimating the composition. However, the proposed method instead creates the synthetic data of mixed materials generated from using only images of pure materials present in those mixtures. Thus, it eliminates the prohibitive cost and tedious process of collecting images of mixed materials. The motivation for this study is to provide mathematical details of the proposed approach in addition to extensive experiments and analyses. We examine the approach on two datasets to demonstrate the ease of extending the proposed approach to any mixtures. We perform experiments to demonstrate that the proposed approach can accurately determine the presence of the materials, and sufficiently estimate the precise composition of a mixed material. Moreover, we provide analyses to strengthen the validation and benefits of the proposed approach.

The semantics of shared memory in Intel CPU/FPGA systems

Heterogeneous CPU/FPGA devices, in which a CPU and an FPGA can execute together while sharing memory, are becoming popular in several computing sectors. In this paper, we study the shared-memory semantics of these devices, with a view to providing a firm foundation for reasoning about the programs that run on them. Our focus is on Intel platforms that combine an Intel FPGA with a multicore Xeon CPU. We describe the weak-memory behaviours that are allowed (and observable) on these devices when CPU threads and an FPGA thread access common memory locations in a fine-grained manner through multiple channels. Some of these behaviours are familiar from well-studied CPU and GPU concurrency; others are weaker still. We encode these behaviours in two formal memory models: one operational, one axiomatic. We develop executable implementations of both models, using the CBMC bounded model-checking tool for our operational model and the Alloy modelling language for our axiomatic model. Using these, we cross-check our models against each other via a translator that converts Alloy-generated executions into queries for the CBMC model. We also validate our models against actual hardware by translating 583 Alloy-generated executions into litmus tests that we run on CPU/FPGA devices; when doing this, we avoid the prohibitive cost of synthesising a hardware design per litmus test by creating our own 'litmus-test processor' in hardware. We expect that our models will be useful for low-level programmers, compiler writers, and designers of analysis tools. Indeed, as a demonstration of the utility of our work, we use our operational model to reason about a producer/consumer buffer implemented across the CPU and the FPGA. When the buffer uses insufficient synchronisation -- a situation that our model is able to detect -- we observe that its performance improves at the cost of occasional data corruption.

Principles and Practices of Plant-based Irrigation Management

Many consider tools for plant-based irrigation management methods to be the most precise way to manage irrigation in either a research or a commercial settings. Although many types of tools are available, they all measure some aspect of water movement along the soil–plant–atmosphere continuum. This article presents some of the more commonly used tools and the methods involved to properly employ them. In addition, recent literature is reviewed to provide context to the methods themselves and also to highlight each one’s specific advantages and disadvantages. Ultimately, there is no clear winner or “best” tool as all have disadvantages, either due to prohibitive cost, the amount of data output, the difficulty of data interpretation, lack of signal resolution, or lack of dynamic ability to provide decision support. Therefore, we conclude that the user should carefully weigh these varied advantages and disadvantages in the context of their production goals before deciding on a given tool for irrigation management.

A Glider Simulation Model Based on Optimized Support Vector Regression for Efficient Coordinated Observation

Multi-gliders have been widely deployed as an array in nowadays ocean observation for fine and long-term ocean research, especially in deep-sea exploration. However, the strong, variable ocean currents and the delayed information feedback of gliders are remaining huge challenges for the deployment of glider arrays which may cause that the observed data cannot meet the study needs and bring a prohibitive cost. In this paper, we develop a Glider Simulation Model (GSM) based on the support vector regression with the particle swarm optimization (PSO)-SVR algorithm to integrate the information feedback from gliders and ocean current data for rapid modeling to effectively predict the gliders’ trajectories. Based on the real-time predictive information of the trajectories, each glider can select future movement strategies. We utilize the in-suit datasets obtained by sea-wing gliders in ocean observation to train and test the simulation model. The results show that GSM has an effective and stable performance. The information obtained from the modeling approaches can be utilized for the optimization of the deployment of the glider arrays.

Nobody can Really Afford Legal Services: The Price of Justice in Namibia

Nobody (except for the privileged few) can afford legal services in Namibia. In the light of this dawning awareness, how should the government and other stakeholders design the legal profession so that the greatest number of Namibians can access legal services and, ultimately, justice while preserving the profession's financial viability? The predominantly economic nature of this question means that its solutions lie less in the field of law than in the field of economics. Thus, this article adopts a methodology that reflects that insight. As a primary purpose, this article works towards solving the high cost of legal services in Namibia. It utilises a literature-review methodology that searches the scholarship on the legal profession for practical, down-to-earth solutions put forward in other countries to take the edge off the prohibitive cost of legal services. The article mainly finds that, if structured as a compulsory salary deduction, legal insurance promises the greatest positive impact on costs. And it concludes that the optimal solutions should consist of measures aimed at heightening competition in the legal profession and measures that broaden cost-sharing in providing legal assistance to the public. The article argues that competition can be effectively increased by lubricating the flow of information about prices and services, and by having more public entities bear the burden of expanding the system of legal assistance.

Social consequences of energetically costly nest construction in a facultatively social bee

Social groups form when the costs of breeding independently exceed fitness costs imposed by group living. The costs of independent breeding can often be energetic, especially for animals performing expensive behaviours, such as nest construction. To test the hypothesis that nesting costs can drive sociality by disincentivizing independent nest founding, we measured the energetics of nest construction and inheritance in a facultatively social carpenter bee ( Xylocopa sonorina Smith), which bores tunnel nests in wood. We measured metabolic rates of bees excavating wood and used computerized tomography images of nesting logs to measure excavation volumes. From these data, we demonstrate costly energetic investments in nest excavation of a minimum 4.3 kJ per offspring provisioned, an expense equivalent to nearly 7 h of flight. This high, potentially prohibitive cost of nest founding may explain why females compete for existing nests rather than constructing new ones, often leading to the formation of social groups. Further, we found that nest inheritors varied considerably in their investment in nest renovation, with costs ranging more than 12-fold (from 7.08 to 89.1 kJ energy), probably reflecting differences in inherited nest quality. On average, renovation costs were lower than estimated new nest construction costs, with some nests providing major savings. These results suggest that females may join social groups to avoid steep energetic costs, but that the benefits of this strategy are not experienced equally.

Moving towards a local testing solution for undetermined thyroid fine-needle aspirates: validation of a novel custom DNA-based NGS panel

AimsIn thyroid cytopathology, the undetermined diagnostic categories still pose diagnostic challenges. Although next-generation sequencing (NGS) is a promising technique for the molecular testing of thyroid fine-needle aspiration (FNA) specimens, access to such technology can be difficult because of its prohibitive cost and lack of reimbursement in countries with universal health coverage. To overcome these issues, we developed and validated a novel custom NGS panel, Nexthyro, specifically designed to target 264 clinically relevant mutations involved in thyroid tumourigenesis. Moreover, in this study, we compared its analytical performance with that of our previous molecular testing strategy.MethodsThe panel, which includes 15 genes (BRAF, EIF1AX, GNAS, HRAS, IDH1, KRAS, NF2, NRAS, PIK3CA, PPM1D, PTEN, RET, DICER1, CHEK2, TERT promoter), was validated with a cell-line derived reference standard and 72 FNA archival samples previously tested with the 7-gene test.ResultsNexthyro yielded 100% specificity and detected mutant alleles at levels as low as 2%. Moreover, in 5/72 (7%) FNAs, it detected more clinically relevant mutations in BRAF and RAS genes compared with the 7-gene test. Nexthyro also revealed better postsequencing metrics than the previously adopted commercial ‘generic’ NGS panel.ConclusionOur comparative analysis indicates that Nexthyro is a reliable NGS panel. The study also implies that a custom-based solution for routine thyroid FNA is sustainable at the local level, allowing patients with undetermined thyroid nodules affordable access to NGS.

Testcrosses are an efficient strategy for identifying cis-regulatory variation: Bayesian analysis of allele-specific expression (BayesASE)

Allelic imbalance (AI) occurs when alleles in a diploid individual are differentially expressed and indicates cis acting regulatory variation. What is the distribution of allelic effects in a natural population? Are all alleles the same? Are all alleles distinct? The approach described applies to any technology generating allele-specific sequence counts, for example for chromatin accessibility and can be applied generally including to comparisons between tissues or environments for the same genotype. Tests of allelic effect are generally performed by crossing individuals and comparing expression between alleles directly in the F1. However, a crossing scheme that compares alleles pairwise is a prohibitive cost for more than a handful of alleles as the number of crosses is at least (n2-n)/2 where n is the number of alleles. We show here that a testcross design followed by a hypothesis test of AI between testcrosses can be used to infer differences between nontester alleles, allowing n alleles to be compared with n crosses. Using a mouse data set where both testcrosses and direct comparisons have been performed, we show that the predicted differences between nontester alleles are validated at levels of over 90% when a parent-of-origin effect is present and of 60%−80% overall. Power considerations for a testcross, are similar to those in a reciprocal cross. In all applications, the testing for AI involves several complex bioinformatics steps. BayesASE is a complete bioinformatics pipeline that incorporates state-of-the-art error reduction techniques and a flexible Bayesian approach to estimating AI and formally comparing levels of AI between conditions. The modular structure of BayesASE has been packaged in Galaxy, made available in Nextflow and as a collection of scripts for the SLURM workload manager on github (https://github.com/McIntyre-Lab/BayesASE).