摘要 :
Deep neural networks (DNNs) have evolved remarkably over the last decade and achieved great success in many machine learning tasks. Along the evolution of deep learning (DL) methods, computational complexity and resource consumpti...
展开
Deep neural networks (DNNs) have evolved remarkably over the last decade and achieved great success in many machine learning tasks. Along the evolution of deep learning (DL) methods, computational complexity and resource consumption of DL models continue to increase, this makes efficient deployment challenging, especially in devices with low memory resources or in applications with strict latency requirements. In this paper, we will introduce a DL inference optimization pipeline, which consists of a series of model compression methods, including Tensor Decomposition (TD), Graph Adaptive Pruning (GAP), Intrinsic Sparse Structures (ISS) in Long Short-Term Memory (LSTM), Knowledge Distillation (KD) and low-bit model quantization. We use different modeling scenarios to test our inference optimization pipeline with above mentioned methods, and it shows promising results to make inference more efficient with marginal loss of model accuracy.
收起
摘要 :
Deep neural networks (DNNs) have evolved remarkably over the last decade and achieved great success in many machine learning tasks. Along the evolution of deep learning (DL) methods, computational complexity and resource consumpti...
展开
Deep neural networks (DNNs) have evolved remarkably over the last decade and achieved great success in many machine learning tasks. Along the evolution of deep learning (DL) methods, computational complexity and resource consumption of DL models continue to increase, this makes efficient deployment challenging, especially in devices with low memory resources or in applications with strict latency requirements. In this paper, we will introduce a DL inference optimization pipeline, which consists of a series of model compression methods, including Tensor Decomposition (TD), Graph Adaptive Pruning (GAP), Intrinsic Sparse Structures (ISS) in Long Short-Term Memory (LSTM), Knowledge Distillation (KD) and low-bit model quantization. We use different modeling scenarios to test our inference optimization pipeline with above mentioned methods, and it shows promising results to make inference more efficient with marginal loss of model accuracy.
收起
摘要 :
We present the idea of peer indexing - indexing an image by semantically correlated images - and its application in image retrieval. A learning strategy is suggested for automatic acquisition of peer indices from user feedback, an...
展开
We present the idea of peer indexing - indexing an image by semantically correlated images - and its application in image retrieval. A learning strategy is suggested for automatic acquisition of peer indices from user feedback, and the similarity metric for the peer index is formulated. A cooperative framework is proposed under which the peer index is integrated with low-level features for image retrieval and relevance feedback. Encouraging results on both short-term and long-term retrieval performance of our approach are shown by experiments.
收起
摘要 :
We present the idea of peer indexing - indexing an image by semantically correlated images - and its application in image retrieval. A learning strategy is suggested for automatic acquisition of peer indices from user feedback, an...
展开
We present the idea of peer indexing - indexing an image by semantically correlated images - and its application in image retrieval. A learning strategy is suggested for automatic acquisition of peer indices from user feedback, and the similarity metric for the peer index is formulated. A cooperative framework is proposed under which the peer index is integrated with low-level features for image retrieval and relevance feedback. Encouraging results on both short-term and long-term retrieval performance of our approach are shown by experiments.
收起
摘要 :
A refractive index insensitive temperature sensor based on coaxial dual-waveguide optical fiber was proposed and demonstrated. The coaxial fiber contains a central core along the fiber axis and an annular core between the inner/ou...
展开
A refractive index insensitive temperature sensor based on coaxial dual-waveguide optical fiber was proposed and demonstrated. The coaxial fiber contains a central core along the fiber axis and an annular core between the inner/outer claddings. By inserting the coaxial fiber in between two single mode fibers through core-offset splicing, cladding modes are excited at the splice point and therefore a modal Mach-Zehnder interferometer is achieved. The effective refractive index of the inner cladding mode is independent of the external refractive index due to the existence of the annular core. Owing to the large thermo-optic coefficient difference between the coaxial fiber's core and cladding, the modal interferometer has high temperature sensitivity. Such an interferometer is extremely suitable for temperature measurement in wet or liquid environment.
收起
摘要 :
This paper documents the features of compensation peer groups and demonstrates their significant role in understanding variations in CEO compensation. We hand-collect a sample of 373 of the S&P 500 firms and 235 of the S&P Mid-Cap 400 firms that provided explicit lists of compensation peer companies in their proxy statements in the 2006 fiscal year. Results show that the median pay level of the compensation peer group dominates the median pay level of the industry peers, industry/size peers, and the firm’s performance peers, as well as the pay level of the firm’s CEO in the previous year, in explaining CEO compensation (after controlling for size, firm performance, and CEO characteristics). Examinations of factors explaining compensation peer group composition demonstrate that even after controlling for industry, size, and equity performance, peer group composition is significantly affected by the level of CEO compensation at the potential peers. Firms appear to select highly paid peers to justify greater CEO compensation and this effect is stronger in firms where the CEO is the Chairman of the Board, the board is busier or older, and Towers Perrin is the compensation consultant. On the other hand, the selection bias is lower in firms where active institutional shareholders have greater ownership, the CEO is newly hired, and shareholders have expressed concerns on executive compensation....
展开
This paper documents the features of compensation peer groups and demonstrates their significant role in understanding variations in CEO compensation. We hand-collect a sample of 373 of the S&P 500 firms and 235 of the S&P Mid-Cap 400 firms that provided explicit lists of compensation peer companies in their proxy statements in the 2006 fiscal year. Results show that the median pay level of the compensation peer group dominates the median pay level of the industry peers, industry/size peers, and the firm’s performance peers, as well as the pay level of the firm’s CEO in the previous year, in explaining CEO compensation (after controlling for size, firm performance, and CEO characteristics). Examinations of factors explaining compensation peer group composition demonstrate that even after controlling for industry, size, and equity performance, peer group composition is significantly affected by the level of CEO compensation at the potential peers. Firms appear to select highly paid peers to justify greater CEO compensation and this effect is stronger in firms where the CEO is the Chairman of the Board, the board is busier or older, and Towers Perrin is the compensation consultant. On the other hand, the selection bias is lower in firms where active institutional shareholders have greater ownership, the CEO is newly hired, and shareholders have expressed concerns on executive compensation.
收起
摘要 :
This paper documents the features of compensation peer groups and demonstrates their significant role in understanding variations in CEO compensation. We hand-collect a sample of 373 of the S&P 500 firms and 235 of the S&P Mid-Cap 400 firms that provided explicit lists of compensation peer companies in their proxy statements in the 2006 fiscal year. Results show that the median pay level of the compensation peer group dominates the median pay level of the industry peers, industry/size peers, and the firm’s performance peers, as well as the pay level of the firm’s CEO in the previous year, in explaining CEO compensation (after controlling for size, firm performance, and CEO characteristics). Examinations of factors explaining compensation peer group composition demonstrate that even after controlling for industry, size, and equity performance, peer group composition is significantly affected by the level of CEO compensation at the potential peers. Firms appear to select highly paid peers to justify greater CEO compensation and this effect is stronger in firms where the CEO is the Chairman of the Board, the board is busier or older, and Towers Perrin is the compensation consultant. On the other hand, the selection bias is lower in firms where active institutional shareholders have greater ownership, the CEO is newly hired, and shareholders have expressed concerns on executive compensation....
展开
This paper documents the features of compensation peer groups and demonstrates their significant role in understanding variations in CEO compensation. We hand-collect a sample of 373 of the S&P 500 firms and 235 of the S&P Mid-Cap 400 firms that provided explicit lists of compensation peer companies in their proxy statements in the 2006 fiscal year. Results show that the median pay level of the compensation peer group dominates the median pay level of the industry peers, industry/size peers, and the firm’s performance peers, as well as the pay level of the firm’s CEO in the previous year, in explaining CEO compensation (after controlling for size, firm performance, and CEO characteristics). Examinations of factors explaining compensation peer group composition demonstrate that even after controlling for industry, size, and equity performance, peer group composition is significantly affected by the level of CEO compensation at the potential peers. Firms appear to select highly paid peers to justify greater CEO compensation and this effect is stronger in firms where the CEO is the Chairman of the Board, the board is busier or older, and Towers Perrin is the compensation consultant. On the other hand, the selection bias is lower in firms where active institutional shareholders have greater ownership, the CEO is newly hired, and shareholders have expressed concerns on executive compensation.
收起
摘要 :
We study an agency model with a novel combination of features - agents (CEOs) differ in their ability, firms choose both the scope of the CEO's activities and their incentives, and there is free entry by firms. The outcome is an i...
展开
We study an agency model with a novel combination of features - agents (CEOs) differ in their ability, firms choose both the scope of the CEO's activities and their incentives, and there is free entry by firms. The outcome is an industry equilibrium in which firms are heterogenous in scope and output. That is, firms hiring more able CEOs complement higher ability with greater scope and stronger incentives, resulting in greater output. Pay has a strong "superstars" element in the sense that motivating higher ability CEOs to accept a job involving more effort and greater risk of managing greater scope, requires much greater rewards. The model is a simple one that makes strong assumptions; this allows us to analyze it very completely and arrive at sharp conclusions. For example, we find that an increase in demand for the industry's product, e.g., a booming economy or opening of foreign economies, increases both the overall level and skewness of the cross section distribution of CEO compensation. The model suggests a variety of other empirical predictions. Some preliminary empirical work suggests the model may prove quite useful for un- derstanding some interesting trends in compensation. For example, our model provides an explanation for the recent increased level and dispersion in CEO compensation that is rooted in product market competition and rational board reaction to changes in the firm's environment.
收起
摘要 :
It is often taken for granted that: 1) capital markets and institutions allocate funds to firms with high returns; 2) the net gains to the economy from investments by corporations have improved in the last 30-50 years due to techn...
展开
It is often taken for granted that: 1) capital markets and institutions allocate funds to firms with high returns; 2) the net gains to the economy from investments by corporations have improved in the last 30-50 years due to technological innovations; and 3) the discipline role of markets and institutions ensures that corporate assets funded with external funds earn higher returns. However, corporate real assets are long lived, and realized returns have to be tracked over a long period to verify these assertions. In this study, we perform large-scale calculations of the realized returns on assets to all firms available in the Compustat database for periods of 10, 20, 30, 40, and 50 years. Our methodology relies only on realized, not expected, cash flows between the firms and all their fund providers. We found several new and surprising results. Realized returns on corporate assets over long periods are, on the whole, lower than expected by the fund providers. They also suffer a long-term decline, and have been below the yields of 10- year Treasury Bonds since 1973. Additionally, firms that received more external financing (from capital markets and institutions) report even lower realized long-term returns. A wealth transfer from an increasingly important class of non-interest bearing liabilities augments the realized returns on equity. These unexpected results may stimulate a fresh debate on the role and long-term performance of capital markets and institutions.
收起
摘要 :
The paper examines three equity-based structural models to study the nonlinear relationship between equity and credit default swap (CDS) prices. These models di.er in the specification of the default barrier. With cross-firm CDS p...
展开
The paper examines three equity-based structural models to study the nonlinear relationship between equity and credit default swap (CDS) prices. These models di.er in the specification of the default barrier. With cross-firm CDS premia and equity information, we are able to estimate and compare the threemodels. We find that the stochastic barrier model performs better than the constant and uncertain barrier models in terms of both in-sample fit and out-of-sample forecasting of CDS premia. In addition, we demonstrate a linkage between the default barrier, jump intensity, and barrier volatility estimated from our models and firm-specific variables related to default risk, such as credit ratings, equity volatility, and leverage ratios.
收起