Search

For $0

Modeling and synthesizing real sRGB noise is crucial for various low-level vision tasks, such as building datasets for training image denoising systems. The distribution of real sRGB noise is highly complex and affected by a multitude of factors, making its accurate modeling extremely challenging. Therefore, recent studies have proposed methods that employ data-driven generative models, such as generative adversarial networks (GAN) and Normalizing Flows. These studies achieve more accurate modeling of sRGB noise compared to traditional noise modeling methods. However, there are performance limitations due to the inherent characteristics of each generative model. To address this issue, we propose NM-FlowGAN, a hybrid approach that exploits the strengths of both GAN and Normalizing Flows. We simultaneously employ a pixel-wise noise modeling network based on Normalizing Flows, and spatial correlation modeling networks based on GAN. In our experiments, our NM-FlowGAN outperforms other baselines on the sRGB noise synthesis task. Moreover, the denoising neural network, trained with synthesized image pairs from our model, also shows superior performance compared to other baselines. Our code is available at: \url{https://github.com/YoungJooHan/NM-FlowGAN}., Comment: 25 pages, 11 figures, 7 tables

This paper proposes a simple and robust zero-shot voice conversion system with a cycle structure and mel-spectrogram pre-processing. Previous works suffer from information loss and poor synthesis quality due to their reliance on a carefully designed bottleneck structure. Moreover, models relying solely on self-reconstruction loss struggled with reproducing different speakers' voices. To address these issues, we suggested a cycle-consistency loss that considers conversion back and forth between target and source speakers. Additionally, stacked random-shuffled mel-spectrograms and a label smoothing method are utilized during speaker encoder training to extract a time-independent global speaker representation from speech, which is the key to a zero-shot conversion. Our model outperforms existing state-of-the-art results in both subjective and objective evaluations. Furthermore, it facilitates cross-lingual voice conversions and enhances the quality of synthesized speech.

Recently, numerous studies have been conducted on supervised learning-based image denoising methods. However, these methods rely on large-scale noisy-clean image pairs, which are difficult to obtain in practice. Denoising methods with self-supervised training that can be trained with only noisy images have been proposed to address the limitation. These methods are based on the convolutional neural network (CNN) and have shown promising performance. However, CNN-based methods do not consider using nonlocal self-similarities essential in the traditional method, which can cause performance limitations. This paper presents self-similarity attention (SS-Attention), a novel self-attention module that can capture nonlocal self-similarities to solve the problem. We focus on designing a lightweight self-attention module in a pixel-wise manner, which is nearly impossible to implement using the classic self-attention module due to the quadratically increasing complexity with spatial resolution. Furthermore, we integrate SS-Attention into the blind-spot network called self-similarity-based blind-spot network (SS-BSN). We conduct the experiments on real-world image denoising tasks. The proposed method quantitatively and qualitatively outperforms state-of-the-art methods in self-supervised denoising on the Smartphone Image Denoising Dataset (SIDD) and Darmstadt Noise Dataset (DND) benchmark datasets., Comment: Accepted to IJCAI 2023

Estimating position bias is a well-known challenge in Learning to Rank (L2R). Click data in e-commerce applications, such as targeted advertisements and search engines, provides implicit but abundant feedback to improve personalized rankings. However, click data inherently includes various biases like position bias. Based on the position-based click model, Result Randomization and Regression Expectation-Maximization algorithm (REM) have been proposed to estimate position bias, but they require various paired observations of (item, position). In real-world scenarios of advertising, marketers frequently display advertisements in a fixed pre-determined order, which creates difficulties in estimation due to the limited availability of various pairs in the training data, resulting in a sparse dataset. We propose a variant of the REM that utilizes item embeddings to alleviate the sparsity of (item, position). Using a public dataset and internal carousel advertisement click dataset, we empirically show that item embedding with Latent Semantic Indexing (LSI) and Variational Auto-Encoder (VAE) improves the accuracy of position bias estimation and the estimated position bias enhances Learning to Rank performance. We also show that LSI is more effective as an embedding creation method for position bias estimation.

In this paper, we study item advertisements for small businesses. This application recommends prospective customers to specific items requested by businesses. From analysis, we found that the existing Recommender Systems (RS) were ineffective for small/new businesses with a few sales history. Training samples in RS can be highly biased toward popular businesses with sufficient sales and can decrease advertising performance for small businesses. We propose a meta-learning-based RS to improve advertising performance for small/new businesses and shops: Meta-Shop. Meta-Shop leverages an advanced meta-learning optimization framework and builds a model for a shop-level recommendation. It also integrates and transfers knowledge between large and small shops, consequently learning better features in small shops. We conducted experiments on a real-world E-commerce dataset and a public benchmark dataset. Meta-Shop outperformed a production baseline and the state-of-the-art RS models. Specifically, it achieved up to 16.6% relative improvement of Recall@1M and 40.4% relative improvement of nDCG@3 for user recommendations to new shops compared to the other RS models.

In one-class recommendation systems, the goal is to learn a model from a small set of interacted users and items and then identify the positively-related user-item pairs among a large number of pairs with unknown interactions. Most previous loss functions rely on dissimilar pairs of users and items, which are selected from the ones with unknown interactions, to obtain better prediction performance. This strategy introduces several challenges such as increasing training time and hurting the performance by picking "similar pairs with the unknown interactions" as dissimilar pairs. In this paper, the goal is to only use the similar set to train the models. We point out three trivial solutions that the models converge to when they are trained only on similar pairs: collapsed, partially collapsed, and shrinking solutions. We propose two terms that can be added to the objective functions in the literature to avoid these solutions. The first one is a hinge pairwise distance loss that avoids the shrinking and collapsed solutions by keeping the average pairwise distance of all the representations greater than a margin. The second one is an orthogonality term that minimizes the correlation between the dimensions of the representations and avoids the partially collapsed solution. We conduct experiments on a variety of tasks on public and real-world datasets. The results show that our approach using only similar pairs outperforms state-of-the-art methods using similar pairs and a large number of dissimilar pairs., Comment: 16 pages, 7 figures

Recently online advertisers utilize Recommender systems (RSs) for display advertising to improve users' engagement. The contextual bandit model is a widely used RS to exploit and explore users' engagement and maximize the long-term rewards such as clicks or conversions. However, the current models aim to optimize a set of ads only in a specific domain and do not share information with other models in multiple domains. In this paper, we propose dynamic collaborative filtering Thompson Sampling (DCTS), the novel yet simple model to transfer knowledge among multiple bandit models. DCTS exploits similarities between users and between ads to estimate a prior distribution of Thompson sampling. Such similarities are obtained based on contextual features of users and ads. Similarities enable models in a domain that didn't have much data to converge more quickly by transferring knowledge. Moreover, DCTS incorporates temporal dynamics of users to track the user's recent change of preference. We first show transferring knowledge and incorporating temporal dynamics improve the performance of the baseline models on a synthetic dataset. Then we conduct an empirical analysis on a real-world dataset and the result showed that DCTS improves click-through rate by 9.7% than the state-of-the-art models. We also analyze hyper-parameters that adjust temporal dynamics and similarities and show the best parameter which maximizes CTR., Comment: Published at ADKDD 2022

Hashing-based Recommender Systems (RSs) are widely studied to provide scalable services. The existing methods for the systems combine three modules to achieve efficiency: feature extraction, interaction modeling, and binarization. In this paper, we study an unexplored module combination for the hashing-based recommender systems, namely Compact Cross-Similarity Recommender (CCSR). Inspired by cross-modal retrieval, CCSR utilizes Maximum a Posteriori similarity instead of matrix factorization and rating reconstruction to model interactions between users and items. We conducted experiments on MovieLens1M, Amazon product review, Ichiba purchase dataset and confirmed CCSR outperformed the existing matrix factorization-based methods. On the Movielens1M dataset, the absolute performance improvements are up to 15.69% in NDCG and 4.29% in Recall. In addition, we extensively studied three binarization modules: $sign$, scaled tanh, and sign-scaled tanh. The result demonstrated that although differentiable scaled tanh is popular in recent discrete feature learning literature, a huge performance drop occurs when outputs of scaled $tanh$ are forced to be binary.

Neural network-based collaborative filtering systems focus on designing network architectures to learn better representations while fixing the input to the user/item interaction vectors and/or ID. In this paper, we first show that the non-zero elements of the inputs are learnable parameters that determine the weights in combining the user/item embeddings, and fixing them limits the power of the models in learning the representations. Then, we propose to learn the value of the non-zero elements of the inputs jointly with the neural network parameters. We analyze the model complexity and the empirical risk of our approach and prove that learning the input leads to a better generalization bound. Our experiments on several real-world datasets show that our method outperforms the state-of-the-art methods, even using shallow network structures with a smaller number of layers and parameters.

This paper studies a parametric family of algorithmic fairness metrics, called generalized entropy, which originally has been used in public welfare and recently introduced to machine learning community. As a meaningful metric to evaluate algorithmic fairness, it requires that generalized entropy specify fairness requirements of a classification problem and the fairness requirements should be realized with small deviation by an algorithm. We investigate the role of generalized entropy as a design parameter for fair classification algorithm through a fair empirical risk minimization with a constraint specified in terms of generalized entropy. We theoretically and experimentally study learnability of the problem., Comment: 56pages and 92 figures Revised for adding experimental results

Cross-modal retrieval aims to search for data with similar semantic meanings across different content modalities. However, cross-modal retrieval requires huge amounts of storage and retrieval time since it needs to process data in multiple modalities. Existing works focused on learning single-source compact features such as binary hash codes that preserve similarities between different modalities. In this work, we propose a jointly learned deep hashing and quantization network (HQ) for cross-modal retrieval. We simultaneously learn binary hash codes and quantization codes to preserve semantic information in multiple modalities by an end-to-end deep learning architecture. At the retrieval step, binary hashing is used to retrieve a subset of items from the search space, then quantization is used to re-rank the retrieved items. We theoretically and empirically show that this two-stage retrieval approach provides faster retrieval results while preserving accuracy. Experimental results on the NUS-WIDE, MIR-Flickr, and Amazon datasets demonstrate that HQ achieves boosts of more than 7% in precision compared to supervised neural network-based compact coding models., Comment: Accepted at BMVC 2021

We introduce a new class of conjugations and characterize complex symmetric Toeplitz operators on the Hardy space with respect to those conjugations. Also, we prove that complex symmetricity and \uet~ property are the same for a certain class of Toeplitz operators. We also discuss the analytic symmetricity for Toeplitz operators. Our results extend several known results by providing unified ways of treating them., Comment: 24 pages

We introduce an approach to recommending short-lived dynamic packages for golf booking services. Two challenges are addressed in this work. The first is the short life of the items, which puts the system in a state of a permanent cold start. The second is the uninformative nature of the package attributes, which makes clustering or figuring latent packages challenging. Although such settings are fairly pervasive, they have not been studied in traditional recommendation research, and there is thus a call for original approaches for recommender systems. In this paper, we introduce a hybrid method that leverages user analysis and its relation to the packages, as well as package pricing and environmental analysis, and traditional collaborative filtering. The proposed approach achieved appreciable improvement in precision compared with baselines.

Autoencoder-based hybrid recommender systems have become popular recently because of their ability to learn user and item representations by reconstructing various information sources, including users' feedback on items (e.g., ratings) and side information of users and items (e.g., users' occupation and items' title). However, existing systems still use representations learned by matrix factorization (MF) to predict the rating, while using representations learned by neural networks as the regularizer. In this paper, we define the neural representation for prediction (NRP) framework and apply it to the autoencoder-based recommendation systems. We theoretically analyze how our objective function is related to the previous MF and autoencoder-based methods and explain what it means to use neural representations as the regularizer. We also apply the NRP framework to a direct neural network structure which predicts the ratings without reconstructing the user and item information. We conduct extensive experiments on two MovieLens datasets and two real-world e-commerce datasets. The results confirm that neural representations are better for prediction than regularization and show that the NRP framework, combined with the direct neural network structure, outperforms the state-of-the-art methods in the prediction task, with less training time and memory.

We present an approach to overcoming challenges in dynamical dexterity for robots through tunable origami structures. Our work leverages a one-parameter family of flat sheet crease patterns that folds into origami bellows, whose axial compliance can be tuned to select desired stiffness. Concentrically arranged cylinder pairs reliably manifest additive stiffness, extending the tunable range by nearly an order of magnitude and achieving bulk axial stiffness spanning 200-1500 N/m using 8 mil thick polyester-coated paper. Accordingly, we design origami energy-storing springs with a stiffness of 1035 N/m each and incorporate them into a three degree-of-freedom (DOF) tendon-driven spatial pointing mechanism that exhibits trajectory tracking accuracy less than 15% rms error within a ~2 cm^3 volume. The origami springs can sustain high power throughput, enabling the robot to achieve asymptotically stable juggling for both highly elastic (1~kg resilient shot put ball) and highly damped ("medicine ball") collisions in the vertical direction with apex heights approaching 10 cm. The results demonstrate that "soft" robotic mechanisms are able to perform a controlled, dynamically actuated task., Comment: This paper is submitted to the IEEE Robotics and Automation Letters, in review

The Rockefeller Institute of Government is conducting a multistate examination of potential supply and demand issues for teachers. In other words, are states facing teaching shortages around the country? This report, the second state-specific review, is an initial overview of the P-12 teacher labor market in New York State (NYS). It examines the question of whether a shortage of elementary and secondary school teachers exists, or is soon to occur, from three perspectives: the balance of current teachers and students; the supply of potential new teachers and how their qualifications fit, or fail to fit, the near-term needs of school districts; and the distribution of challenges in staffing across districts of different economic, racial, and ethnic compositions.

The link between education funding levels and student outcomes is a matter of continuing debate, including in New York State. State education aid in New York follows a progressive formula that targets more aid to lower-income, higher-need school districts, yet these districts continue to dominate lists of the state's lowest performing schools. The Center for Education Pipeline Systems Change at the Rockefeller Institute of Government examines various aspects of the education system to pinpoint problems and to identify and grow programs that are improving student outcomes and success. In this brief, education spending and academic performance in New York are examined. The state's more than 4,000 public schools are ranked by academic performance, and measures of those schools' fiscal capacity and wealth and poverty measures are presented. District per pupil spending is highlighted.

Working with the South Dakota Department of Education, the authors have begun collecting data to pinpoint where potential shortages and turnover of teachers are as a way to lay the foundation of targeted solutions in specific areas. Given the data collection process, they have broken the study into two phases: (1) Phase one: is a district-by-district analysis of teacher shortages and turnover in the South Dakota public school system; and (2) Phase two: will add teacher-level data to enable them to map a more finely grained analysis to better understand the attributes of teachers who flow into positions, move among schools, and exit from South Dakota's public schools. In this initial analysis, the focus is on the indicators of potential shortages as conveyed in a recent, one-year snapshot. This report lays out some of the challenges and opportunities districts face filling positions. It does not offer policy prescriptions but suggests some directions for consideration, based on the preliminary findings.

We develop a Stata command xthenreg to implement the first-differenced GMM estimation of the dynamic panel threshold model, which Seo and Shin (2016, Journal of Econometrics 195: 169-186) have proposed. Furthermore, We derive the asymptotic variance formula for a kink constrained GMM estimator of the dynamic threshold model and include an estimation algorithm. We also propose a fast bootstrap algorithm to implement the bootstrap for the linearity test. The use of the command is illustrated through a Monte Carlo simulation and an economic application.

This study starts from the counter-intuitive question of how we can render a conventional stiff, non-stretchable and even brittle material conformable so that it can fully wrap around a curved surface, such as a sphere, without failure. Here, we answer this conundrum by extending geometrical design in computational kirigami (paper cutting and folding) to paper wrapping. Our computational paper wrapping-based approach provides the more robust and reliable fabrication of conformal devices than paper folding approaches. This in turn leads to a significant increase in the applicability of computational kirigami to real-world fabrication. This new computer-aided design transforms 2D-based conventional materials, such as Si and copper, into a variety of targeted conformal structures that can fully wrap the desired 3D structure without plastic deformation or fracture. We further demonstrated that our novel approach enables a pluripotent design platform to transform conventional non-stretchable 2D-based devices, such as electroluminescent lighting and a paper battery, into wearable and conformable 3D curved devices.

Origami structures enabled by folding and unfolding can create complex 3D shapes. However, even a small 3D shape can have large 2D unfoldings. The huge initial dimension of the 2D flattened structure makes fabrication difficult, and defeats the main purpose, namely compactness, of many origami-inspired engineering. In this work, we propose a novel algorithmic kirigami method that provides super compaction of an arbitrary 3D shape with non-negligible surface thickness called "algorithmic stacking". Our approach computationally finds a way of cutting the thick surface of the shape into a strip. This strip forms a Hamiltonian cycle that covers the entire surface and can realize transformation between two target shapes: from a super compact stacked shape to the input 3D shape. Depending on the surface thickness, the stacked structure takes merely 0.001% to 6% of the original volume. This super compacted structure not only can be manufactured in a workspace that is significantly smaller than the provided 3D shape, but also makes packing and transportation easier for a deployable application. We further demonstrate that, the proposed stackable structure also provides high pluripotency and can transform into multiple 3D target shapes if these 3D shapes can be dissected in specific ways and form a common stacked structure. In contrast to many designs of origami structure that usually target at a particular shape, our results provide a universal platform for pluripotent 3D transformable structures.

This paper presents a novel pre-shared key (PSK) agreement scheme to establish a secure connection between a Wi-Fi client and access point (AP) without prior knowledge of a password. The standard IEEE 802.11 security method, Robust Security Network Association, widely known as Wi-Fi Protected Access (WPA) and WPA2, derives a shared cryptographic key if and only if a user provides an identical password which an AP possesses, causing ofinconvenience of obtaining and entering the password. In this paper, a proposed scheme, Secure Open AP (SOAP), adopts two public key algorithms, the elliptic curve Diffie-Hellman key exchange algorithm (ECDH) and digital signature algorithm (ECDSA) to establish a secure connection between a client and an AP without having prior knowledge of a password. Implementation and experiment results demonstrate the viability of the proposed scheme.

The Korean very-long-baseline interferometry (VLBI) network (KVN) and VLBI Exploration of Radio Astrometry (VERA) Array (KaVA) is the first international VLBI array dedicated to high-frequency (23 and 43 GHz bands) observations in East Asia. Here, we report the first imaging observations of three bright active galactic nuclei (AGNs) known for their complex morphologies: 4C 39.25, 3C 273, and M 87. This is one of the initial result of KaVA early science. Our KaVA images reveal extended outflows with complex substructure such as knots and limb brightening, in agreement with previous Very Long Baseline Array (VLBA) observations. Angular resolutions are better than 1.4 and 0.8 milliarcsecond at 23 GHz and 43 GHz, respectively. KaVA achieves a high dynamic range of ~1000, more than three times the value achieved by VERA. We conclude that KaVA is a powerful array with a great potential for the study of AGN outflows, at least comparable to the best existing radio interferometric arrays., Comment: 15 pages, 10 figures, 6 tables, accepted for publication in PASJ

We have carried out the first very long baseline interferometry (VLBI) imaging of 44 GHz class I methanol maser (7_{0}-6_{1}A^{+}) associated with a millimeter core MM2 in a massive star-forming region IRAS 18151-1208 with KaVA (KVN and VERA Array), which is a newly combined array of KVN (Korean VLBI Network) and VERA (VLBI Exploration of Radio Astrometry). We have succeeded in imaging compact maser features with a synthesized beam size of 2.7 milliarcseconds x 1.5 milliarcseconds (mas). These features are detected at a limited number of baselines within the length of shorter than approximately 650 km corresponding to 100 Mlambda in the uv-coverage. The central velocity and the velocity width of the 44 GHz methanol maser are consistent with those of the quiescent gas rather than the outflow traced by the SiO thermal line. The minimum component size among the maser features is ~ 5 mas x 2 mas, which corresponds to the linear size of ~ 15 AU x 6 AU assuming a distance of 3 kpc. The brightness temperatures of these features range from ~ 3.5 x 10^{8} to 1.0 x 10^{10} K, which are higher than estimated lower limit from a previous Very Large Array observation with the highest spatial resolution of ~ 50 mas. The 44 GHz class I methanol maser in IRAS 18151-1208 is found to be associated with the MM2 core, which is thought to be less evolved than another millimeter core MM1 associated with the 6.7 GHz class II methanol maser., Comment: 19 pages, 3 figures

Using a revised version of the catalog of AGB stars by Suh & Kwon (2009), we present various infrared two-colour diagrams (2CDs) for 3003 O-rich, 1168 C-rich, 362 S-type and 35 silicate carbon stars in our Galaxy. For each object in the new catalog, we cross-identify the AKARI, MSX and 2MASS counterparts by finding the nearest one from the position information in the IRAS PSC. For the large sample of AGB stars, we present infrared two-colour diagrams using IRAS (PSC), AKARI (PSC and BSC), MSX (PSC) and near infrared (K and L bands; including 2MASS data at KS band) data for different classes of AGB stars based on the chemistry of the dust shell and/or the central star. The infrared 2CDs of AGB stars can provide useful information about the structure and evolution of the dust envelopes as well as the central stars. On the 2CDs, we plot tracks of the theoretical radiative transfer model results with increasing dust shell optical depths. Comparing the observations with the theoretical models on the new 2CDs, we find that the basic model tracks roughly coincide with the densely populated observed points. Generally, we can explain the observations of O-rich and C-rich AGB stars on the various 2CDs with the theoretical models using dust opacity functions of amorphous silicate, amorphous carbon, SiC and corundum. For O-rich AGB stars, we find that the models using corundum as well as silicate can improve the fit with the observations.

We report room temperature ferromagnetism in Ni doped CeO2 nanoparticles using X-ray diffraction (XRD), high resolution transmission electron microscopy (HR-TEM), and dc magnetization measurements. Nanoparticles of Ce1-xNixO2 (0.0 \leq x \leq 0.10) were prepared by using a co-precipitation method. XRD measurements indicate that all samples exhibit single phase nature with cubic structure and ruled out the presence of any secondary phase. Lattice parameters, strain and particle size calculated from XRD data have been found to decrease with increase in Ni doping. Inter-planner distance measured from HR-TEM images for different Ni doped samples indicate that Ni ions are substituting Ce ions in CeO2 matrix. Magnetization measurements performed at room temperature display weak ferromagnetic behavior of Ce1-xNixO2 (0.0 \leq x \leq 0.10) nanoparticles. Magnetic moment calculated from magnetic hysteresis loop was found to increases with Ni doping up to 7% and then start decreasing with further doping., Comment: 8 pages 4 figures

In this paper we present a compositional semantics for the channel-based coordination language Reo which enables the analysis of quality of service (QoS) properties of service compositions. For this purpose, we annotate Reo channels with stochastic delay rates and explicitly model data-arrival rates at the boundary of a connector, to capture its interaction with the services that comprise its environment. We propose Stochastic Reo automata as an extension of Reo automata, in order to compositionally derive a QoS-aware semantics for Reo. We further present a translation of Stochastic Reo automata to Continuous-Time Markov Chains (CTMCs). This translation enables us to use third-party CTMC verification tools to do an end-to-end performance analysis of service compositions., Comment: In Proceedings FOCLASA 2010, arXiv:1007.4993

This article provides guidance for music teachers who want to introduce Korean folk music to their students. It introduces "Arirang"--both a specific folk song and a category--and describes three different versions: the "Miryang Arirang," the "Jindo Arirang," and the "Gyeonggi Arirang." The "Arirang" tune appears in many social contexts that can be shared with students; these examples show how well known the form has become worldwide. In addition, "Arirang" songs have rhythms and lyrics that lend themselves to being taught through a rich variety of listening and performing activities; they can also provide opportunities for student composition and improvisation. The article includes suggestions for presenting "Arirang" in the music classroom.