2,497 publications from this institution
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.
We study the problem of optimizing the mapping of LDPC codes on parallel machines to minimize the communication cost. To reduce the search space, the problem is solved in two stages: clustering, and cluster allocation. We propose a simplified clustering technique based on a modified min-cut algorithm that reduces the search complexity from O(n/sup 2/) to O(n). It was found that most of the locality is exploited by the clustering operation, which results in a 40-52% improvement in the total communication cost over random mapping. For large networks, cluster allocation is much more costly and results in only 1-8% additional improvement in unidirectional and bi-directional torus topologies. We compared the performance of two different approaches for cluster allocation. The first one is bused on min-cut algorithm, and the second one is based on a genetic algorithm. It was found that the min-cut based approach is better for small network sizes. For large network sizes with the number of clusters /spl ges/64, the genetic based approach becomes more attractive.
Abstract Background E-values are a recently introduced approach to evaluate confounding in observational studies. We aimed to empirically assess the current use of E-values in published literature. Methods We conducted a systematic literature search for all publications, published up till the end of 2018, which cited at least one of two inceptive E-value papers and presented E-values for original data. For these case publications we identified control publications, matched by journal and issue, where the authors had not calculated E-values. Results In total, 87 papers presented 516 E-values. Of the 87 papers, 14 concluded that residual confounding likely threatens at least some of the main conclusions. Seven of these 14 named potential uncontrolled confounders. 19 of 87 papers related E-value magnitudes to expected strengths of field-specific confounders. The median E-value was 1.88, 1.82, and 2.02 for the 43, 348, and 125 E-values where confounding was felt likely to affect the results, unlikely to affect the results, or not commented upon, respectively. The 69 case-control publication pairs dealt with effect sizes of similar magnitude. Of 69 control publications, 52 did not comment on unmeasured confounding and 44/69 case publications concluded that confounding was unlikely to affect study conclusions. Conclusions Few papers using E-values conclude that confounding threatens their results, and their E-values overlap in magnitude with those of papers acknowledging susceptibility to confounding. Facile automation in calculating E-values may compound the already poor handling of confounding. E-values should not be a substitute for careful consideration of potential sources of unmeasured confounding. If used, they should be interpreted in the context of expected confounding in specific fields.
The authors introduce a two-port BiCMOS (bipolar complementary metal-oxide semiconductor) static memory cell that combines ECL (emitter-coupled-logic)-level word-line voltage swings and emitter-follower bit line coupling with a static CMOS latch to achieve access times comparable to those of high-speed bipolar SRAMs (static random-access memories), which preserving the high density and low power of CMOS memory arrays. The memory can be accessed for read and write independently and simultaneously, making it especially attractive for the design of video, cache, and other application-specific memories. An experimental 4 K*1 bit two-port memory integrated in a 1.5- mu m-5-GHz BiCMOS technology exhibits a read access time of 4 ns and a power dissipation of 550 mW.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">></ETX>
ii Chapter One: Introduction 1 Chapter Two: The HKM derivatives in the Qur’an 4 Ḥakīm 4 Ḥikmah 9 Ḥakam 12 Ḥakama and other verbs 14 Conclusion 20 Chapter Three: Hukam in the compositions of Kabir 23 Conclusion 30 Chapter Four: Guru Nanak and hukam 33 Historical and social context 33 Hukam in the compositions of Nanak 37 Conclusion 49 Chapter Five: The Sikh Gurus and hukam 52 The Sikh community as an historical entity 52 Hukam in the compositions of the Sikh Gurus 57 Conclusion 65 Chapter Six: The development of hukam 68 References 70 Bibliography 73
Promising evidence from clinical studies of drug effects does not always translate to improvements in patient outcomes. In this review, we discuss why early evidence is often ill suited to the task of predicting the clinical utility of drugs. The current gap between initially described drug effects and their subsequent clinical utility results from deficits in the design, conduct, analysis, reporting, and synthesis of clinical studies—often creating conditions that generate favorable, but ultimately incorrect, conclusions regarding drug effects. There are potential solutions that could improve the relevance of clinical evidence in predicting the real-world effectiveness of drugs. What is needed is a new emphasis on clinical utility, with nonconflicted entities playing a greater role in the generation, synthesis, and interpretation of clinical evidence. Clinical studies should adopt strong design features, reflect clinical practice, and evaluate outcomes and comparisons that are meaningful to patients. Transformative changes to the research agenda may generate more meaningful and accurate evidence on drug effects to guide clinical decision making.
Victor A. van de Graaf, MD; Julia C. A. Noorduyn, MSc; Nienke W. Willigenburg, PhD; Ise K. Butter, MSc; Arthur de Gast, MD, PhD; Ben W. Mol, MD, PhD; Daniel B. F. Saris, MD, PhD; Jos W. R. Twisk, PhD; Rudolf W. Poolman, MD, PhD; for the ESCAPE Research Group
Clinicians, when trying to apply trial results to patient care, need to individualize patient care and, potentially, manage patients based on results of subgroup analyses. Apparently compelling subgroup effects often prove spurious, and guidance is needed to differentiate credible from less credible subgroup claims. We therefore provide 5 criteria to use when assessing the validity of subgroup analyses: (1) Can chance explain the apparent subgroup effect; (2) Is the effect consistent across studies; (3) Was the subgroup hypothesis one of a small number of hypotheses developed a priori with direction specified; (4) Is there strong preexisting biological support; and (5) Is the evidence supporting the effect based on within- or between-study comparisons. The first 4 criteria are applicable to individual studies or systematic reviews, the last only to systematic reviews of multiple studies. These criteria will help clinicians deciding whether to use subgroup analyses to guide their patient care.
Abstract Aims The European Medicines Agency (EMA) and the US Food and Drug Administration (FDA) produce guidelines for the design of pivotal psychiatric drug trials used in new drug applications. It is unknown who are involved in the guideline development and what specific trial design recommendations they give. Methods Cross-sectional study of EMA Clinical Efficacy and Safety Guidelines and FDA Guidance Documents. Study outcomes: (1) guideline committee members and declared conflicts of interest; (2) guideline development and organisation of commenting phases; (3) categorisation of stakeholders who comment on draft and final guidelines according to conflicts of interest (‘industry’, ‘not-industry but with industry-related conflicts’, ‘independent’, ‘unclear’); and (4) trial design recommendations (trial duration, psychiatric comorbidity, ‘enriched design’, efficacy outcomes, comparator choice). Protocol registration https://doi.org/10.1101/2020.01.22.20018499 (27 January 2020). Results We included 13 EMA and five FDA guidelines covering 15 psychiatric indications. Eleven months after submission, the EMA had not processed our request regarding committee member disclosures. FDA offices draft the Guidance Documents, but the Agency is not in possession of employee conflicts of interest declarations because FDA employees generally may not hold financial interests (although some employees may hold interests up to $15,000). The EMA and FDA guideline development phases are similar; drafts and final versions are publicly announced and everybody can submit comments. Seventy stakeholders commented on ten guidelines: 38 (54%) ‘industry’, 18 (26%) ‘not-industry but with industry-related conflicts’, six (9%) ‘independent’ and eight (11%) ‘unclear’. They submitted 1014 comments: 640 (68%) ‘industry’, 243 (26%) ‘not-industry but with industry-related conflicts’, 44 (5%) ‘independent’ and 20 (2%) ‘unclear’ (67 could not be assigned to a specific stakeholder). The recommended designs were generally for trials of short duration; with restricted trial populations; allowing previous exposure to the drug; and often recommending rating scale efficacy outcomes. EMA mainly recommended three arm designs (both placebo and active comparators), whereas FDA mainly recommended placebo-controlled designs. There were also other important differences and FDA's recommendations regarding the exclusion of psychiatric comorbidity seemed less restrictive. Conclusions The EMA and FDA clinical research guidelines for psychiatric pivotal trials recommend designs that tend to have limited generalisability. Independent and non-conflicted stakeholders are underrepresented in the guideline development. It seems warranted with more active involvement of scientists and independent organisations without conflicts of interest in the guideline development process.
University leaders aim to protect, shape, and promote the missions of their institutions. I evaluated whether top highly cited scientists are likely to occupy these positions. Of the current leaders of 96 U.S. high research activity universities, only 6 presidents or chancellors were found among the 4009 U.S. scientists listed in the ISIHighlyCited.com database. Of the current leaders of 77 UK universities, only 2 vice-chancellors were found among the 483 UK scientists listed in the same database. In a sample of 100 top-cited clinical medicine scientists and 100 top-cited biology and biochemistry scientists, only 1 and 1, respectively, had served at any time as president of a university. Among the leaders of 25 U.S. universities with the highest citation volumes, only 12 had doctoral degrees in life, natural, physical or computer sciences, and 5 of these 12 had a Hirsch citation index m < 1.0. The participation of highly cited scientists in the top leadership of universities is limited. This could have consequences for the research and overall mission of universities.
We describe the Student Electronic Notebook and the process of porting IBM's AIX 1.1 to run on it. We believe that portable workstation-class machines connected by wireless networks and dependent on a computational and informational infrastructure raise a number of important issues in operating systems and distributed computation (e.g., the partitioning of tasks between workstations and infrastructure), and therefore the development of such machines and their software is important. We conclude by summarizing our activites, itemizing the lessons we learned and identifying the key criteria for the design of the successor machines.
