arXiv subjects

Han Yu

Publications and source records attributed to Han Yu.

At least 253 records · Page 14Linked to original sources

arXiv2019

Privacy-preserving Crowd-guided AI Decision-making in Ethical Dilemmas

With the rapid development of artificial intelligence (AI), ethical issues surrounding AI have attracted increasing attention. In particular, autonomous vehicles may face moral dilemmas in accident scenarios, such as staying the course resulting in hurting pedestrians or swerving leading to hurting passengers. To investigate such ethical dilemmas, recent studies have adopted preference aggregation, in which each voter expresses her/his preferences over decisions for the possible ethical dilemma scenarios, and a centralized system aggregates these preferences to obtain the winning decision. Although a useful methodology for building ethical AI systems, such an approach can potentially violate the privacy of voters since moral preferences are sensitive information and their disclosure can be exploited by malicious parties. In this paper, we report a first-of-its-kind privacy-preserving crowd-guided AI decision-making approach in ethical dilemmas. We adopt the notion of differential privacy to quantify privacy and consider four granularities of privacy protection by taking voter-/record-level privacy protection and centralized/distributed perturbation into account, resulting in four approaches VLCP, RLCP, VLDP, and RLDP. Moreover, we propose different algorithms to achieve these privacy protection granularities, while retaining the accuracy of the learned moral preference model. Specifically, VLCP and RLCP are implemented with the data aggregator setting a universal privacy parameter and perturbing the averaged moral preference to protect the privacy of voters' data. VLDP and RLDP are implemented in such a way that each voter perturbs her/his local moral preference with a personalized privacy parameter. Extensive experiments on both synthetic and real data demonstrate that the proposed approach can achieve high accuracy of preference aggregation while protecting individual voter's privacy.

cs.CR↗

arXiv2019

Transfer Learning with Dynamic Distribution Adaptation

Transfer learning aims to learn robust classifiers for the target domain by leveraging knowledge from a source domain. Since the source and the target domains are usually from different distributions, existing methods mainly focus on adapting the cross-domain marginal or conditional distributions. However, in real applications, the marginal and conditional distributions usually have different contributions to the domain discrepancy. Existing methods fail to quantitatively evaluate the different importance of these two distributions, which will result in unsatisfactory transfer performance. In this paper, we propose a novel concept called Dynamic Distribution Adaptation (DDA), which is capable of quantitatively evaluating the relative importance of each distribution. DDA can be easily incorporated into the framework of structural risk minimization to solve transfer learning problems. On the basis of DDA, we propose two novel learning algorithms: (1) Manifold Dynamic Distribution Adaptation (MDDA) for traditional transfer learning, and (2) Dynamic Distribution Adaptation Network (DDAN) for deep transfer learning. Extensive experiments demonstrate that MDDA and DDAN significantly improve the transfer learning performance and setup a strong baseline over the latest deep and adversarial methods on digits recognition, sentiment analysis, and image classification. More importantly, it is shown that marginal and conditional distributions have different contributions to the domain divergence, and our DDA is able to provide good quantitative evaluation of their relative importance which leads to better performance. We believe this observation can be helpful for future research in transfer learning.

cs.LG↗

arXiv2019

Generating Persuasive Visual Storylines for Promotional Videos

Video contents have become a critical tool for promoting products in E-commerce. However, the lack of automatic promotional video generation solutions makes large-scale video-based promotion campaigns infeasible. The first step of automatically producing promotional videos is to generate visual storylines, which is to select the building block footage and place them in an appropriate order. This task is related to the subjective viewing experience. It is hitherto performed by human experts and thus, hard to scale. To address this problem, we propose WundtBackpack, an algorithmic approach to generate storylines based on available visual materials, which can be video clips or images. It consists of two main parts, 1) the Learnable Wundt Curve to evaluate the perceived persuasiveness based on the stimulus intensity of a sequence of visual materials, which only requires a small volume of data to train; and 2) a clustering-based backpacking algorithm to generate persuasive sequences of visual materials while considering video length constraints. In this way, the proposed approach provides a dynamic structure to empower artificial intelligence (AI) to organize video footage in order to construct a sequence of visual stimuli with persuasive power. Extensive real-world experiments show that our approach achieves close to 10% higher perceived persuasiveness scores by human testers, and 12.5% higher expected revenue compared to the best performing state-of-the-art approach.

cs.MM↗

arXiv2019

Approximate arithmetic structure in large sets of integers

We prove that if a set is `large' in the sense of Erdős, then it approximates arbitrarily long arithmetic progressions in a strong quantitative sense. More specifically, expressing the error in the approximation in terms of the gap length $Δ$ of the progression, we improve a previous result of $o(Δ)$ to $O(Δ^α)$ for any $α\in (0,1)$.

math.MG↗

arXiv2019

Easy Transfer Learning By Exploiting Intra-domain Structures

Transfer learning aims at transferring knowledge from a well-labeled domain to a similar but different domain with limited or no labels. Unfortunately, existing learning-based methods often involve intensive model selection and hyperparameter tuning to obtain good results. Moreover, cross-validation is not possible for tuning hyperparameters since there are often no labels in the target domain. This would restrict wide applicability of transfer learning especially in computationally-constraint devices such as wearables. In this paper, we propose a practically Easy Transfer Learning (EasyTL) approach which requires no model selection and hyperparameter tuning, while achieving competitive performance. By exploiting intra-domain structures, EasyTL is able to learn both non-parametric transfer features and classifiers. Extensive experiments demonstrate that, compared to state-of-the-art traditional and deep methods, EasyTL satisfies the Occam's Razor principle: it is extremely easy to implement and use while achieving comparable or better performance in classification accuracy and much better computational efficiency. Additionally, it is shown that EasyTL can increase the performance of existing transfer feature learning methods.

cs.LG↗

arXiv2019

Ethically Aligned Opportunistic Scheduling for Productive Laziness

In artificial intelligence (AI) mediated workforce management systems (e.g., crowdsourcing), long-term success depends on workers accomplishing tasks productively and resting well. This dual objective can be summarized by the concept of productive laziness. Existing scheduling approaches mostly focus on efficiency but overlook worker wellbeing through proper rest. In order to enable workforce management systems to follow the IEEE Ethically Aligned Design guidelines to prioritize worker wellbeing, we propose a distributed Computational Productive Laziness (CPL) approach in this paper. It intelligently recommends personalized work-rest schedules based on local data concerning a worker's capabilities and situational factors to incorporate opportunistic resting and achieve superlinear collective productivity without the need for explicit coordination messages. Extensive experiments based on a real-world dataset of over 5,000 workers demonstrate that CPL enables workers to spend 70% of the effort to complete 90% of the tasks on average, providing more ethically aligned scheduling than existing approaches.

cs.AI↗

arXiv2018

Building Ethics into Artificial Intelligence

As artificial intelligence (AI) systems become increasingly ubiquitous, the topic of AI governance for ethical decision-making by AI has captured public imagination. Within the AI research community, this topic remains less familiar to many researchers. In this paper, we complement existing surveys, which largely focused on the psychological, social and legal discussions of the topic, with an analysis of recent advances in technical solutions for AI governance. By reviewing publications in leading AI conferences including AAAI, AAMAS, ECAI and IJCAI, we propose a taxonomy which divides the field into four areas: 1) exploring ethical dilemmas; 2) individual ethical decision frameworks; 3) collective ethical decision frameworks; and 4) ethics in human-AI interactions. We highlight the intuitions and key techniques used in each approach, and discuss promising future research directions towards successful integration of ethical AI systems into human societies.

cs.AI↗

arXiv2018

Dimension growth for iterated sumsets

We study dimensions of sumsets and iterated sumsets and provide natural conditions which guarantee that a set $F \subseteq \mathbb{R}$ satisfies $\overline{\dim}_\text{B} F+F > \overline{\dim}_\text{B} F$ or even $\dim_\text{H} n F \to 1$. Our results apply to, for example, all uniformly perfect sets, which include Ahlfors-David regular sets. Our proofs rely on Hochman's inverse theorem for entropy and the Assouad and lower dimensions play a critical role. We give several applications of our results including an Erdős-Volkmann type theorem for semigroups and new lower bounds for the box dimensions of distance sets for sets with small dimension.

math.MG↗

arXiv2018

Cross-position Activity Recognition with Stratified Transfer Learning

Human activity recognition aims to recognize the activities of daily living by utilizing the sensors on different body parts. However, when the labeled data from a certain body position (i.e. target domain) is missing, how to leverage the data from other positions (i.e. source domain) to help learn the activity labels of this position? When there are several source domains available, it is often difficult to select the most similar source domain to the target domain. With the selected source domain, we need to perform accurate knowledge transfer between domains. Existing methods only learn the global distance between domains while ignoring the local property. In this paper, we propose a \textit{Stratified Transfer Learning} (STL) framework to perform both source domain selection and knowledge transfer. STL is based on our proposed \textit{Stratified} distance to capture the local property of domains. STL consists of two components: Stratified Domain Selection (STL-SDS) can select the most similar source domain to the target domain; Stratified Activity Transfer (STL-SAT) is able to perform accurate knowledge transfer. Extensive experiments on three public activity recognition datasets demonstrate the superiority of STL. Furthermore, we extensively investigate the performance of transfer learning across different degrees of similarities and activity levels between domains. We also discuss the potential applications of STL in other fields of pervasive computing for future research.

cs.CV↗

arXiv2018

$L^q$-spectra of self-affine measures: closed forms, counterexamples, and split binomial sums

We study $L^q$-spectra of planar self-affine measures generated by diagonal systems with an emphasis on providing closed form expressions. We answer a question posed by Fraser in 2016 in the negative by proving that a certain natural closed form expression does not generally give the $L^q$-spectrum and, using a similar approach, find counterexamples to a statement of Falconer-Miao from 2007 and a conjecture of Miao from 2008 concerning a closed form expression for the generalised dimensions of generic self-affine measures. In the positive direction we provide new non-trivial closed form bounds in both of the above settings, which in certain cases yield sharp results. We also provide examples of self-affine measures whose $L^q$-spectra exhibit new types of phase transitions. Our examples depend on a combinatorial estimate for the exponential growth of certain split binomial sums.

math.MG↗

arXiv2018

On GILP's group theoretic approach to Falconer's distance problem

In this paper, we follow and extend a group-theoretic method introduced by Greenleaf-Iosevich-Liu-Palsson (GILP) to study finite points configurations spanned by Borel sets in $\mathbb{R}^n,n\geq 2,n\in\mathbb{N}.$ We remove a technical continuity condition in a GILP's theorem in [GILP15]. This allows us to extend the Wolff-Erdogan dimension bound for distance sets to finite points configurations with $k$ points for $k\in\{2,\dots,n+1\}.$ At the end of this paper, we extend this group-theoretic method and illustrate a `Fourier free' approach to Falconer's distance set problem for the Lebesgue measure. We explain how to use tubular incidence estimates in distance set problems. Curiously, tubular incidence estimates are also related to the Kakeya problem.

math.MG↗

arXiv2018

Dimensions of triangle sets

In this paper, we discuss some dimension results for triangle sets of compact sets in $\mathbb{R}^2$. In particular, we prove that for any compact set $F$ in $\mathbb{R}^2$, the triangle set $Δ(F)$ satisfies \[ \dim_{\mathrm{A}} Δ(F)\geq \frac{3}{2}\dim_{\mathrm{A}} F. \] If $\dim_{\mathrm{A}} F>1$ then we have \[ \dim_{\mathrm{A}} Δ(F)\geq 1+\dim_{\mathrm{A}} F. \] If $\dim_{\mathrm{A}} F>4/3$ then we have the following better bound, \[ \dim_{\mathrm{A}} Δ(F)\geq \min\left\{\frac{5}{2}\dim_{\mathrm{A}} F-1,3\right\}. \] Moreover, if $F$ satisfies a mild separation condition then the above result holds also for the box dimensions, namely, \[ \underline{\dim_{\mathrm{B}}} F\geq \frac{3}{2}\underline{\dim_{\mathrm{B}}} Δ(F) \text{ and }\overline{\dim_{\mathrm{B}}} F\geq \frac{3}{2}\overline{\dim_{\mathrm{B}}} Δ(F). \]

math.MG↗

arXiv2018

A Fourier analytic approach to inhomogeneous Diophantine approximation

In this paper, we study inhomogeneous Diophantine approximation with rational numbers of reduced form. The central object to study is the set $W(f,θ)$ as follows, \begin{eqnarray*} \left\{x\in [0,1]:\left |x-\frac{m+θ(n)}{n}\right|<\frac{f(n)}{n}\text{ for infinitely many coprime pairs of numbers } m,n\right\}, \end{eqnarray*} where $\{f(n)\}_{n\in\mathbb{N}}$ and $\{θ(n)\}_{n\in\mathbb{N}}$ are sequences of real numbers in $[0,1/2]$. We will completely determine the Hausdorff dimension of $W(f,θ)$ in terms of $f$ and $θ$. As a by-product, we also obtain a new sufficient condition for $W(f,θ)$ to have full Lebesgue measure and this result is closely related to the study of \ds with extra conditions.

math.NT↗

arXiv2018

Multi-rotations on the unit circle

In this paper, we study multi-rotation orbits on the unit circle. We obtain a natural generalization of a classical result which says that orbits of irrational rotations on the unit circle are dense. It is possible to show that this result holds true if instead of iterating a single irrational rotation, one takes a multi-rotation orbit along a finitely recurrent sequence over finitely many different irrational rotations. We also discuss some connections between the box dimensions of multi-rotation orbits and Diophantine approximations. In particular, we improve a result by Feng and Xiong in the case when the rotation parameters are algebraic numbers.

math.DS↗

arXiv2018

Visual Domain Adaptation with Manifold Embedded Distribution Alignment

Visual domain adaptation aims to learn robust classifiers for the target domain by leveraging knowledge from a source domain. Existing methods either attempt to align the cross-domain distributions, or perform manifold subspace learning. However, there are two significant challenges: (1) degenerated feature transformation, which means that distribution alignment is often performed in the original feature space, where feature distortions are hard to overcome. On the other hand, subspace learning is not sufficient to reduce the distribution divergence. (2) unevaluated distribution alignment, which means that existing distribution alignment methods only align the marginal and conditional distributions with equal importance, while they fail to evaluate the different importance of these two distributions in real applications. In this paper, we propose a Manifold Embedded Distribution Alignment (MEDA) approach to address these challenges. MEDA learns a domain-invariant classifier in Grassmann manifold with structural risk minimization, while performing dynamic distribution alignment to quantitatively account for the relative importance of marginal and conditional distributions. To the best of our knowledge, MEDA is the first attempt to perform dynamic distribution alignment for manifold domain adaptation. Extensive experiments demonstrate that MEDA shows significant improvements in classification accuracy compared to state-of-the-art traditional and deep methods.

cs.CV↗

arXiv2018

A multi-channel DAQ system based on FPGA for long-distance transmission in nuclear physics experiments

As the development of electronic science and technology, electronic data acquisition (DAQ) system is more and more widely applied to nuclear physics experiments. Workstations are often utilized for data storage, data display, data processing and data analysis by researchers. Nevertheless, the workstations are ordinarily separated from detectors in nuclear physics experiments by several kilometers or even tens of kilometers. Thus a DAQ system that can transmit data for long distance is in demand. In this paper, we designed a DAQ system suitable for high-speed and high-precision sampling for remote data transfer. An 8-channel, 24-bit simultaneous sampling analog-to-digital converter(ADC) named AD7779 was utilized for high-speed and high-precision sampling, the maximum operating speed of which runs up to 16 kilo samples per second(KSPS). ADC is responsible for collecting signals from detectors, which is sent to Field Programmable Gate Array(FPGA) for processing and long-distance transmission to the workstation through optical fiber. As the central processing unit of DAQ system, FPGA provides powerful computing capability and has enough flexibility. The most prominent feature of the system is real-time mass data transfer based on streaming transmission mode, highly reliable data transmission based on error detection and correction and high-speed high-precision data acquisition. The results of our tests show that the system is able to transmit data stably at the bandwidth of 1Gbps.

physics.ins-det↗

arXiv2018

Weak tangents and level sets of Takagi functions

In this paper we study some properties of Takagi functions and their level sets. We show that for Takagi functions $T_{a,b}$ with parameters $a,b$ such that $ab$ is a root of a Littlewood polynomial, there exist large level sets. As a consequence we show that for some parameters $a,b$, the Assouad dimension of graphs of $T_{a,b}$ is strictly larger than their upper box dimension. In particular we can find weak tangents of those graphs with large Hausdorff dimension, larger than the upper box dimension of the graphs.

math.CA↗

arXiv2018

New dimension spectra: finer information on scaling and homogeneity

We introduce a new dimension spectrum motivated by the Assouad dimension; a familiar notion of dimension which, for a given metric space, returns the minimal exponent $α\geq 0$ such that for any pair of scales $0<r<R$, any ball of radius $R$ may be covered by a constant times $(R/r)^α$ balls of radius $r$. To each $θ\in (0,1)$, we associate the appropriate analogue of the Assouad dimension with the restriction that the two scales $r$ and $R$ used in the definition satisfy $\log R/\log r = θ$. The resulting `dimension spectrum' (as a function of $θ$) thus gives finer geometric information regarding the scaling structure of the space and, in some precise sense, interpolates between the upper box dimension and the Assouad dimension. This latter point is particularly useful because the spectrum is generally better behaved than the Assouad dimension. We also consider the corresponding `lower spectrum', motivated by the lower dimension, which acts as a dual to the Assouad spectrum. We conduct a detailed study of these dimension spectra; including analytic, geometric, and measureability properties. We also compute the spectra explicitly for some common examples of fractals including decreasing sequences with decreasing gaps and spirals with sub-exponential and monotonic winding. We also give several applications of our results, including: dimension distortion estimates under bi-Hölder maps for Assouad dimension and the provision of new bi-Lipschitz invariants.

math.CA↗