Open Access. Powered by Scholars. Published by Universities.®

Digital Commons Network™

Open Access. Powered by Scholars. Published by Universities.®

Singapore Management University

Discipline
Keyword
Publication Year
Publication
Publication Type

Articles 2761 - 2790 of 8706

Full-Text Articles in Entire DC Network

Privacy-Preserving Outsourced Clinical Decision Support System In The Cloud, Ximeng Liu, Robert H. Deng, Kim-Kwang Raymond Choo, Yang Yang Jan 2021

Privacy-Preserving Outsourced Clinical Decision Support System In The Cloud, Ximeng Liu, Robert H. Deng, Kim-Kwang Raymond Choo, Yang Yang

Research Collection School Of Computing and Information Systems

In this paper, we propose a privacy-preserving clinical decision support system using Naïve Bayesian (NB) classifier, hereafter referred to as Peneus, designed for the outsourced cloud computing environment. Peneus allows one to use patient health information to train the NB classifier privately, which can then be used to predict a patient's (undiagnosed) disease based on his/her symptoms in a single communication round. Specifically, we design secure Single Instruction Multiple Data (SIMD) integer circuits using the fully homomorphic encryption scheme, which can greatly increase the performance compared with the original secure integer circuit. Then, we present a privacy-preserving historical Personal Health …

Go to article

Novel Techniques In Recovering, Embedding, And Enforcing Policies For Control-Flow Integrity, Yan Lin Jan 2021

Novel Techniques In Recovering, Embedding, And Enforcing Policies For Control-Flow Integrity, Yan Lin

Dissertations and Theses Collection (Open Access)

Control-Flow Integrity (CFI) is an attractive security property with which most injected and code-reuse attacks can be defeated, including advanced attacking techniques like Return-Oriented Programming (ROP). CFI extracts a control-flow graph (CFG) for a given program and instruments the program to respect the CFG. Specifically, checks are inserted before indirect branch instructions. Before these instructions are executed during runtime, the checks consult the CFG to ensure that the indirect branch is allowed to reach the intended target. Hence, any sort of controlflow hijacking would be prevented. There are three fundamental components in CFI enforcement. The first component is accurately recovering …

Go to article

Does Ict Result In Dematerialization? The Case Of Europe, 2005-2017, Annika Marie Rieger Jan 2021

Does Ict Result In Dematerialization? The Case Of Europe, 2005-2017, Annika Marie Rieger

Research Collection School of Social Sciences

Current levels of resource use are unsustainable, but there is a debate about the most feasible way to reduce them. One proposed mechanism is technological innovation: specifically, the implementation of information and communication technologies (ICTs) could result in significant reductions in material consumption by substituting virtual for material goods, increasing resource efficiency, and replacing more resource-intensive sectors. Critics of this view argue that dematerialization due to ICTs is unlikely: they consume large amounts of resources and encourage additional consumption. Additionally, increased efficiency resulting from ICT use could lead to rebound effects, reducing their environmentally beneficial impact. This paper uses a …

Go to article

Consumption, Annika Marie Rieger, Juliet B. Schor Jan 2021

Consumption, Annika Marie Rieger, Juliet B. Schor

Research Collection School of Social Sciences

Consumption is a major contributor to environmental degradation and change. However, it was not until 1992—at the United Nations Conference on Environment and Development—that consumption was seriously addressed by the global community. The consensus that emerged was that the global South had a “population” problem and the global North had a “consumption,” or more correctly, an “overconsumption” problem. It proved to be a durable formulation. Within environmental sociology, the prominence of the IPAT (Impact = Population × Affluence × Technology) equation (Ehrlich & Holdren, 1971) has contributed to this framing of the environment/consumption relation, although the rise of a global …

Go to article

Becoming Citizens: Policy Feedback And The Transformation Of The Thai Rice Farmer, Jacob Ricks, Thanapan Laiprakobsub Jan 2021

Becoming Citizens: Policy Feedback And The Transformation Of The Thai Rice Farmer, Jacob Ricks, Thanapan Laiprakobsub

Research Collection School of Social Sciences

Over the past twenty years, Thailand’s rice farmers have become one of the country’s most important and active political constituencies, a sharp contrast from the previous decades wherein they were treated with neglect or even derision by the Thai political elite. These “political peasants” now actively advocate for and successfully receive extensive subsidies from both authoritarian and democratic governments. What has driven this change? In this essay, we draw on theories of the policy feedback loop wherein policies yield both material and cognitive benefits, which change the political behavior of populations. We argue that the Thaksin Shinawatra government’s (2001-2006) paddy …

Go to article

Wisdom, Not Veritism, Shane Ryan Jan 2021

Wisdom, Not Veritism, Shane Ryan

Research Collection School of Social Sciences

In this response to Pritchard's "In Defence of Veritism", I defend the view that it is wisdom rather than truth that is fundamental in epistemology. Given that recent philosophical discussions of the nature of wisdom may be unfamiliar to some epistemologists, a brief overview of these discussions is provided and that which is relevant for the subsequent discussion in this piece is highlighted. I explain that scholars working on the topic tend to accept that wisdom comprises at least one familiar epistemic standing and that, unlike other epistemic goods, wisdom is thought to be tied to a narrowly specified content. …

Go to article

Subjective Social Status And Inflammation: The Role Of Culture And Anger Control, Jose C. Yong, Andree Hartanto, Jacinth Jia Xin Tan Jan 2021

Subjective Social Status And Inflammation: The Role Of Culture And Anger Control, Jose C. Yong, Andree Hartanto, Jacinth Jia Xin Tan

Research Collection School of Social Sciences

Objective: Research on subjective social status (SSS) and inflammation risk suffers from a lack of cross-cultural data as well as inconsistent findings between SSS and the biomarker C-reactive protein (CRP). The current study addressed these issues by examining possible cultural differences in the SSS-CRP link with anger control as an underlying mechanism while controlling for potential confounds such as wealth, education, and health factors. Method: Participants comprised 1,435 adults from the Biomarker Project of the MIDUS (American) and MIDJA (Japanese) studies. Participants’ SSS and tendency to control anger were assessed through surveys, and their CRP levels were measured through fasting …

Go to article

Humanist But Not Radical: The Educational Philosophy Of Thiruvalluvar Kural, Devin K. Joshi Jan 2021

Humanist But Not Radical: The Educational Philosophy Of Thiruvalluvar Kural, Devin K. Joshi

Research Collection School of Social Sciences

Humanist ideas in education have been promoted by both Western thinkers and classical wisdom texts of Asia. Exploring this connection, I examine the educational philosophy of an iconic ancient Tamil (Indian) text, the Thiruvalluvar Kural, by juxtaposing it with a contemporary humanist classic, Paulo Freire’s Pedagogy of the Oppressed. As this comparative study reveals, both texts offer humanist visions of relevance to education, politics, and society. Notably, however, the Kural takes what might be described as a more mainstream humanist stance vis-à-vis Freire’s radical humanist approach. Nevertheless, both educational philosophies share a common humanist bond representing important breakthroughs …

Go to article

Styles Of Policy Advice: A Typology For Comparing The Standard Operating Procedures For The Provision Of Policy Advice, David Aubin, Marleen Brans Jan 2021

Styles Of Policy Advice: A Typology For Comparing The Standard Operating Procedures For The Provision Of Policy Advice, David Aubin, Marleen Brans

Research Collection School of Social Sciences

Policy scholars have long studied the phenomenon of policy advice, with the aim of understanding variations in the professional practice of policy analysts and other civil servants. The quest for theorizing the selection and use of advice and information in the policy process has triggered several attempts of typologizing the many facets of policy advice: the location and autonomy of the advisors, their roles, the content of advice, as well as the styles of policy advisors. The styles of policy advice bring attention to the standard operating procedures which influence the content and use of policy advice. This chapter develops …

Go to article

A Fluctuating Sense Of Power Is Associated With Reduced Well-Being, Eric M. Anicich, Michael Schaerer, Jake Gale, Trevor A. Foulk Jan 2021

A Fluctuating Sense Of Power Is Associated With Reduced Well-Being, Eric M. Anicich, Michael Schaerer, Jake Gale, Trevor A. Foulk

Research Collection Lee Kong Chian School Of Business

Social power research has been limited by theoretical and methodological traditions that prioritize static comparisons of high and low-power states. This is a crucial limitation given power’s inherently dynamic nature. Accordingly, Anicich and Hirsh (2017a) recently developed a theoretical framework related to the consequences of vertical code-switching – i.e., the act of alternating between behavioral patterns directed toward higher-power and lower-power interaction partners – known as the approach-inhibition-avoidance (AIA) theory of power. Across five main studies and two supplemental studies, we present the first empirical test of this theory using a mix of survey, experimental, and experience-sampling methods. We demonstrate …

Go to article

Public Health Insurance And Pharmaceutical Innovation: Evidence From China, Xuan Zhang, Huihua Nie Jan 2021

Public Health Insurance And Pharmaceutical Innovation: Evidence From China, Xuan Zhang, Huihua Nie

Research Collection School Of Economics

Developing countries are characterized by low levels of pharmaceutical innovation. A likely reason is their small market size, which is not because of the population size but because of low levels of income and lack of health insurance coverage. This study exploits a natural experiment from the implementation of a public health insurance program for rural residents in China (New Cooperative Medical Scheme [NCMS]) to examine whether the pharmaceutical industry increases innovation regarding diseases covered by the NCMS that are prevalent in rural areas. We examine the 1993–2009 patent data to gauge pharmaceutical innovation in China. Diseases with a 10% …

Go to article

Joint Tests For Dynamic And Spatial Effects In Short Dynamic Panel Data Models With Fixed Effects And Heteroskedasticity, Zhenlin Yang Jan 2021

Joint Tests For Dynamic And Spatial Effects In Short Dynamic Panel Data Models With Fixed Effects And Heteroskedasticity, Zhenlin Yang

Research Collection School Of Economics

Simple and reliable tests are proposed for testing the existence of dynamic and/or spatial effects in fixed-effects panel data models with small T and possibly heteroskedastic errors. The tests are constructed based on the adjusted quasi scores (AQS), which correct the conditional quasi scores given the initial differences to account for the effect of initial values. To improve the finite sample performance, standardized AQS tests are also derived, which are shown to have much improved finite sample properties. All the proposed tests are robust against nonnormality, but some are not robust against cross-sectional heteroskedasticity (CH). A different type of adjustments …

Go to article

Generalized Jump Regressions For Local Moments, Tim Bollerslev, Jia Li, Leonardo Salim Saker Chaves Jan 2021

Generalized Jump Regressions For Local Moments, Tim Bollerslev, Jia Li, Leonardo Salim Saker Chaves

Research Collection School Of Economics

We develop new high-frequency-based inference procedures for analyzing the relationship between jumps in instantaneous moments of stochastic processes. The estimation consists of two steps: the nonparametric determination of the jumps as differences in local averages, followed by a minimum-distance type estimation of the parameters of interest under general loss functions that include both least-square and more robust quantile regressions as special cases. The resulting asymptotic distribution of the estimator, derived under an infill asymptotic setting, is highly nonstandard and generally not mixed normal. In addition, we establish the validity of a novel bootstrap algorithm for making feasible inference including bias-correction. …

Go to article

Retirement Preparedness In Singapore, Junxing Chay, Seonghoon Kim Jan 2021

Retirement Preparedness In Singapore, Junxing Chay, Seonghoon Kim

Research Collection School Of Economics

This chapter presents empirical evidence on a variety of measures to assess the social and economic well-being of near-retirees and retirees in Singapore based on rich data from the Singapore Life Panel, a nationally representative monthly panel survey of 50- to 70-year-old adults in Singapore. Large differences in retirement behaviour and household financial conditions related to retirement preparedness across education and wealth groups are documented. This suggests the importance of a retirement-related policy that focuses on the low-skilled and low-wealth groups. In addition, we discuss and evaluate current public policy designed to increase the employment of older workers and propose …

Go to article

3d Dental Biometrics: Automatic Pose-Invariant Dental Arch Extraction And Matching, Xin Zhong, Zhiyuan Zhang Jan 2021

3d Dental Biometrics: Automatic Pose-Invariant Dental Arch Extraction And Matching, Xin Zhong, Zhiyuan Zhang

Research Collection School Of Computing and Information Systems

A novel automatic pose-invariant dental arch extraction and matching framework is developed for 3D dental identification using laser-scanned dental plasters. In our previous attempt [1-5], 3D point-based algorithms have been developed and they have shown a few advantages over existing 2D dental identifications. This study is a continuous effort in developing arch-based algorithms to extract and match dental arch feature in an automatic and pose-invariant way. As best as we know, this is the first attempt at automatic dental arch extraction and matching for 3D dental identification. A Radial Ray Algorithm (RRA) is proposed by projecting dental arch shape from …

Go to article

Deep Unsupervised Anomaly Detection, Tangqing Li, Zheng Wang, Siying Liu, Wen-Yan Lin Jan 2021

Deep Unsupervised Anomaly Detection, Tangqing Li, Zheng Wang, Siying Liu, Wen-Yan Lin

Research Collection School Of Computing and Information Systems

This paper proposes a novel method to detect anomalies in large datasets under a fully unsupervised setting. The key idea behind our algorithm is to learn the representation underlying normal data. To this end, we leverage the latest clustering technique suitable for handling high dimensional data. This hypothesis provides a reliable starting point for normal data selection. We train an autoencoder from the normal data subset, and iterate between hypothesizing normal candidate subset based on clustering and representation learning. The reconstruction error from the learned autoencoder serves as a scoring function to assess the normality of the data. Experimental results …

Go to article

Recalibrating Directors’ Liabilities Amidst Technological Flux, Pey Woan Lee, Susanna Leong Jan 2021

Recalibrating Directors’ Liabilities Amidst Technological Flux, Pey Woan Lee, Susanna Leong

Research Collection Yong Pung How School Of Law

In this chapter, we consider how to balance corporate innovation and personal liability in the specific context of directors’ exposure to personal liability for corporate copyright infringements. Directors may incur personal liability in the statutory context if they have ‘authorised’ the company’s infringement, or at common law if they acted as the company’s joint tortfeasor. Due to the conflicting policy goals of encouraging innovation whilst preserving personal accountability, we observe that this regime is unsatisfactory. Delving into the modern history of copyright law, we seek to demonstrate that directors of technological companies operate in an environment of high legal uncertainty …

Go to article

Spandeck: A Relational View Of The Duty Of Care, Kian Peng Soh Jan 2021

Spandeck: A Relational View Of The Duty Of Care, Kian Peng Soh

Research Collection Yong Pung How School Of Law

The use of a general framework in the determination of a duty of care has seemingly fallen out of favour following the UK Supreme Court’s decision in Robinson. Relying on the example of the Spandeck framework in Singaporean jurisprudence, this piece presents the argument that such frameworks, being consistent with a relational conception of tort law, can provide a useful means of determining whether a duty of care exists. In so doing, this piece addresses some criticisms of the relational view and re-emphasises the important role the duty of care plays in the tort of negligence.

Go to article

Facial Emotion Recognition With Noisy Multi-Task Annotations, S. Zhang, Zhiwu Huang, D.P. Paudel, Gool L. Van Jan 2021

Facial Emotion Recognition With Noisy Multi-Task Annotations, S. Zhang, Zhiwu Huang, D.P. Paudel, Gool L. Van

Research Collection School Of Computing and Information Systems

Human emotions can be inferred from facial expressions. However, the annotations of facial expressions are often highly noisy in common emotion coding models, including categorical and dimensional ones. To reduce human labelling effort on multi-task labels, we introduce a new problem of facial emotion recognition with noisy multitask annotations. For this new problem, we suggest a formulation from the point of joint distribution match view, which aims at learning more reliable correlations among raw facial images and multi-task labels, resulting in the reduction of noise influence. In our formulation, we exploit a new method to enable the emotion prediction and …

Go to article

Proxy-Free Privacy-Preserving Task Matching With Efficient Revocation In Crowdsourcing, Jiangang Shu, Kan Yang, Xiaohua Jia, Ximeng Liu, Cong Wang, Robert H. Deng Jan 2021

Proxy-Free Privacy-Preserving Task Matching With Efficient Revocation In Crowdsourcing, Jiangang Shu, Kan Yang, Xiaohua Jia, Ximeng Liu, Cong Wang, Robert H. Deng

Research Collection School Of Computing and Information Systems

Task matching in crowdsourcing has been extensively explored with the increasing popularity of crowdsourcing. However, privacy of tasks and workers is usually ignored in most of exiting solutions. In this paper, we study the problem of privacy-preserving task matching for crowdsourcing with multiple requesters and multiple workers. Instead of utilizing proxy re-encryption, we propose a proxy-free task matching scheme for multi-requester/multi-worker crowdsourcing, which achieves task-worker matching over encrypted data with scalability and non-interaction. We further design two different mechanisms for worker revocation including ServerLocal Revocation (SLR) and Global Revocation (GR), which realize efficient worker revocation with minimal overhead on the …

Go to article

An Efficient Privacy Preserving Message Authentication Scheme For Internet-Of-Things, Jiannan Wei, Tran Viet Xuan Phuong, Guomin Yang Jan 2021

An Efficient Privacy Preserving Message Authentication Scheme For Internet-Of-Things, Jiannan Wei, Tran Viet Xuan Phuong, Guomin Yang

Research Collection School Of Computing and Information Systems

As an essential element of the next generation Internet, Internet of Things (IoT) has been undergoing an extensive development in recent years. In addition to the enhancement of peoples daily lives, IoT devices also generate/gather a massive amount of data that could be utilized by machine learning and big data analytics for different applications. Due to the machine-to-machine communication nature of IoT, data security and privacy are crucial issues that must be addressed to prevent different cyber attacks (e.g., impersonation and data pollution/poisoning attacks). Nevertheless, due to the constrained computation power and the diversity of IoT devices, it is a …

Go to article

Discovering Hidden Topical Hubs And Authorities Across Multiple Online Social Networks, Ka Wei, Roy Lee, Tuan-Anh Hoang, Ee-Peng Lim Jan 2021

Discovering Hidden Topical Hubs And Authorities Across Multiple Online Social Networks, Ka Wei, Roy Lee, Tuan-Anh Hoang, Ee-Peng Lim

Research Collection School Of Computing and Information Systems

Finding influential users in online social networks (OSNs) is an important problem with many possible useful applications. Many methods have been proposed to identify influential users in OSNs. PageRank and HITs are two well known examples that determine influential users through link analysis. In recent years, new models that consider both content and social network links have been developed. The Hub and Authority Topic (HAT) model is one that extends HITS to identify topic-specific hubs and authorities by jointly learning hubs, authorities, and topical interests from users’ relationship and textual content. However, many of the previous works are confined to …

Go to article

Chronic Customers Or Increased Awareness? The Dynamics Of Social Media Customer Service, Shujing Sun, Yang Gao, Huaxia Rui Jan 2021

Chronic Customers Or Increased Awareness? The Dynamics Of Social Media Customer Service, Shujing Sun, Yang Gao, Huaxia Rui

Research Collection School Of Computing and Information Systems

Despite that social media has become a promising alternative to traditional call centers, managers hesitate to fully harness its power because they worry that active service intervention may encourage excessive use of the channel by disgruntled customers. This paper sheds light on such a concern by examining the dynamics between brand-level customer complaints and service interventions on social media. Using details of customer-brand interactions of 40 airlines on Twitter, we find that more service interventions indeed cause more customer complaints, accounting for the online customer population and service quality. However, the increased complaints are primarily driven by the awareness enhancement …

Go to article

Hategan: Adversarial Generative-Based Data Augmentation For Hate Speech Detection, Rui Cao, Roy Ka-Wei Lee Dec 2020

Hategan: Adversarial Generative-Based Data Augmentation For Hate Speech Detection, Rui Cao, Roy Ka-Wei Lee

SCIS Student Publications

Academia and industry have developed machine learning and natural language processing models to detect online hate speech automatically. However, most of these existing methods adopt a supervised approach that heavily depends on labeled datasets for training. This results in the methods’ poor detection performance of the hate speech class as the training datasets are highly imbalanced. In this paper, we propose HateGAN, a deep generative reinforcement learning model, which addresses the challenge of imbalance class by augmenting the dataset with hateful tweets. We conduct extensive experiments to augment two commonly-used hate speech detection datasets with the HateGAN generated tweets. Our …

Go to article

Point Optimal Testing With Roots That Are Functionally Local To Unity, Anna Bykhovskaya, Peter C. B. Phillips Dec 2020

Point Optimal Testing With Roots That Are Functionally Local To Unity, Anna Bykhovskaya, Peter C. B. Phillips

Research Collection School Of Economics

Limit theory for regressions involving local to unit roots (LURs) is now used extensively in time series econometric work, establishing power properties for unit root and cointegration tests, assisting the construction of uniform confidence intervals for autoregressive coefficients, and enabling the development of methods robust to departures from unit roots. The present paper shows how to generalize LUR asymptotics to cases where the localized departure from unity is a time varying function rather than a constant. Such a functional local unit root (FLUR) model has much greater generality and encompasses many cases of additional interest that appear in practical work, …

Go to article

Decomposing Duration Dependence: Skill Depreciation Vs. Statistical Discrimination, Ismail Baydur, Jianhuan Xu Dec 2020

Decomposing Duration Dependence: Skill Depreciation Vs. Statistical Discrimination, Ismail Baydur, Jianhuan Xu

Research Collection School Of Economics

This paper develops a random matching model with unobserved worker heterogeneity and learning about worker types from unemployment duration. The model features negative duration dependence that stems from unobserved heterogeneity as well as statistical discrimination and skill depreciation. We estimate our model using micro-level data from Current Population Survey (CPS) and we decompose the contribution of each channel to job finding rates by duration. We find that shutting down statistical discrimination substantially increases the job finding rates of the long-term unemployed while skill depreciation mainly affects the medium-term unemployed.

Go to article

Cooperative Approaches To Managing Social Responsibility In A Market With Externalities, Xin Fang, Soo-Haeng Cho Dec 2020

Cooperative Approaches To Managing Social Responsibility In A Market With Externalities, Xin Fang, Soo-Haeng Cho

Research Collection Lee Kong Chian School Of Business

Problem definition: This paper studies two cooperative approaches of firms in managing social responsibility violations of their supplier: auditing a common supplier jointly (joint auditing) and sharing independent audit results with other firms (audit sharing). We study this problem in a market with externalities and a large number of firms. Academic/practical relevance: With numerous firms procuring their materials and parts worldwide, there are many cases in which overseas suppliers violate safety, labor, or environmental standards. Those violations have externalities in the sense that one firm’s violation affects other firms in the same market. …

Go to article

Predicting Counterproductive Work Behavior: Do Implicit Motives Have Incremental Validity Beyond Explicit Traits?, J. Malte Runge, Jonas W. B. Lang, Ingo Zettler, Filip Lievens Dec 2020

Predicting Counterproductive Work Behavior: Do Implicit Motives Have Incremental Validity Beyond Explicit Traits?, J. Malte Runge, Jonas W. B. Lang, Ingo Zettler, Filip Lievens

Research Collection Lee Kong Chian School Of Business

This study extends research on the link between personality and Counterproductive Work Behavior (CWB) by investigating whether the implicit Affiliation, Achievement, and Power motives contribute to the prediction of CWB beyond basic personality traits. Employees high in Affiliation, Achievement, and Power motives may disengage from CWB because it is not rewarding and thwarts goal attainment. In Study 1 (N = 263), we found that Affiliation predicted self-rated CWB beyond traits. In Study 2 (N = 121), we found that Affiliation and Power predicted supervisor-rated CWB. Our findings thus suggest to also consider implicit motives as personality determinants of CWB.

Go to article

Enhancing Students’ Global Competence Through International Business Study Missions, Mark Chong, Benjamin Gan, Thomas Menkhoff Dec 2020

Enhancing Students’ Global Competence Through International Business Study Missions, Mark Chong, Benjamin Gan, Thomas Menkhoff

Research Collection Lee Kong Chian School Of Business

This article shares how an Asian University enhanced students’ global competence through international business study missions (BSMs). More specifically, it focuses on (i) how the design of these BSMs enabled “deep” learning beyond industry tourism and (ii) how 21st century competencies such as ‘global competence’ can be acquired through participation in short-term, faculty-led study missions.Using the case study approach, it critically analyses the learning goals and objectives, design decisions, implementation details and learning outcomes underlying three business study missions led by three instructors from the same university to the USA (New York), Germany (Berlin and Stuttgart), and South Korea (Seoul).The …

Go to article

Towards Theoretically Understanding Why Sgd Generalizes Better Than Adam In Deep Learning, Pan Zhou, Jiashi Feng, Chao Ma, Caiming Xiong, Steven C. H. Hoi, Weinan E Dec 2020

Towards Theoretically Understanding Why Sgd Generalizes Better Than Adam In Deep Learning, Pan Zhou, Jiashi Feng, Chao Ma, Caiming Xiong, Steven C. H. Hoi, Weinan E

Research Collection School Of Computing and Information Systems

It is not clear yet why ADAM-alike adaptive gradient algorithms suffer from worse generalization performance than SGD despite their faster training speed. This work aims to provide understandings on this generalization gap by analyzing their local convergence behaviors. Specifically, we observe the heavy tails of gradient noise in these algorithms. This motivates us to analyze these algorithms through their Lévy-driven stochastic differential equations (SDEs) because of the similar convergence behaviors of an algorithm and its SDE. Then we establish the escaping time of these SDEs from a local basin. The result shows that (1) the escaping time of both SGD …

Go to article

First Prev Next Last