91 Hits in 0.95 sec

Online Linear Optimization via Smoothing [article]

Jacob Abernethy, Chansoo Lee, Abhinav Sinha, Ambuj Tewari
2014 arXiv   pre-print
We present a new optimization-theoretic approach to analyzing Follow-the-Leader style algorithms, particularly in the setting where perturbations are used as a tool for regularization. We show that adding a strongly convex penalty function to the decision rule and adding stochastic perturbations to data correspond to deterministic and stochastic smoothing operations, respectively. We establish an equivalence between "Follow the Regularized Leader" and "Follow the Perturbed Leader" up to the
more » ... thness properties. This intuition leads to a new generic analysis framework that recovers and improves the previous known regret bounds of the class of algorithms commonly known as Follow the Perturbed Leader.
arXiv:1405.6076v1 fatcat:joqzo3hkuferdpdon6ex6togba

Fighting Bandits with a New Kind of Smoothness [article]

Jacob Abernethy, Chansoo Lee, Ambuj Tewari
2015 arXiv   pre-print
We define a novel family of algorithms for the adversarial multi-armed bandit problem, and provide a simple analysis technique based on convex smoothing. We prove two main results. First, we show that regularization via the Tsallis entropy, which includes EXP3 as a special case, achieves the Θ(√(TN)) minimax regret. Second, we show that a wide class of perturbation methods achieve a near-optimal regret as low as O(√(TN N)) if the perturbation distribution has a bounded hazard rate. For example,
more » ... the Gumbel, Weibull, Frechet, Pareto, and Gamma distributions all satisfy this key property.
arXiv:1512.04152v1 fatcat:7xda44hse5hzdnm4v7ozfotzoy

Spectral Smoothing via Random Matrix Perturbations [article]

Jacob Abernethy, Chansoo Lee, Ambuj Tewari
2015 arXiv   pre-print
We consider stochastic smoothing of spectral functions of matrices using perturbations commonly studied in random matrix theory. We show that a spectral function remains spectral when smoothed using a unitarily invariant perturbation distribution. We then derive state-of-the-art smoothing bounds for the maximum eigenvalue function using the Gaussian Orthogonal Ensemble (GOE). Smoothing the maximum eigenvalue function is important for applications in semidefinite optimization and online
more » ... As a direct consequence of our GOE smoothing results, we obtain an O((N N)^1/4√(T)) expected regret bound for the online variance minimization problem using an algorithm that performs only a single maximum eigenvector computation per time step. Here T is the number of rounds and N is the matrix dimension. Our algorithm and its analysis also extend to the more general online PCA problem where the learner has to output a rank k subspace. The algorithm just requires computing k maximum eigenvectors per step and enjoys an O(k (N N)^1/4√(T)) expected regret bound.
arXiv:1507.03032v2 fatcat:yofihysieba5fivbfpbr3rvjuu

Pre-training helps Bayesian optimization too [article]

Zi Wang, George E. Dahl, Kevin Swersky, Chansoo Lee, Zelda Mariet, Zachary Nado, Justin Gilmer, Jasper Snoek, Zoubin Ghahramani
2022 arXiv   pre-print
Bayesian optimization (BO) has become a popular strategy for global optimization of many expensive real-world functions. Contrary to a common belief that BO is suited to optimizing black-box functions, it actually requires domain knowledge on characteristics of those functions to deploy BO successfully. Such domain knowledge often manifests in Gaussian process priors that specify initial beliefs on functions. However, even with expert knowledge, it is not an easy task to select a prior. This is
more » ... especially true for hyperparameter tuning problems on complex machine learning models, where landscapes of tuning objectives are often difficult to comprehend. We seek an alternative practice for setting these functional priors. In particular, we consider the scenario where we have data from similar functions that allow us to pre-train a tighter distribution a priori. To verify our approach in realistic model training setups, we collected a large multi-task hyperparameter tuning dataset by training tens of thousands of configurations of near-state-of-the-art models on popular image and text datasets, as well as a protein sequence dataset. Our results show that on average, our method is able to locate good hyperparameters at least 3 times more efficiently than the best competing methods.
arXiv:2207.03084v1 fatcat:hwa43bqjljcypenzubbjkx7yj4

Learning and Selecting Features Jointly with Point-wise Gated Boltzmann Machines

Kihyuk Sohn, Guanyu Zhou, Chansoo Lee, Honglak Lee
2013 International Conference on Machine Learning  
The PGBM can be extended to a convolutional setting (Lee et al., 2011) , where we share the filter weights over different locations in large images.  ...  This construction makes sense because the first layer features are mostly generic, and the class-specific features emerge in higher layers (Lee et al., 2011) .  ... 
dblp:conf/icml/SohnZLL13 fatcat:4xcsch3op5ahjndycwqr5bsvxu

A Generalized Family of Multidimensional Continued Fractions: TRIP Maps [article]

Krishna Dasaratha, Laure Flapan, Thomas Garrity, Chansoo Lee, Cornelia Mihaila, Nicholas Neumann-Chun, Sarah Peluse, Matthew Stroffregen
2012 arXiv   pre-print
Most well-known multidimensional continued fractions, including the Mönkemeyer map and the triangle map, are generated by repeatedly subdividing triangles. This paper constructs a family of multidimensional continued fractions by permuting the vertices of these triangles before and after each subdivision. We obtain an even larger class of multidimensional continued fractions by composing the maps in the family. These include the algorithms of Brun, Parry-Daniels and Güting. We give criteria for
more » ... when multidimensional continued fractions associate sequences to unique points, which allows us to determine when periodicity of the corresponding multidimensional continued fraction corresponds to pairs of real numbers being cubic irrationals in the same number field.
arXiv:1206.7077v1 fatcat:ertjvbc6xzeype3vw5cqvc2tfy

Open Source Vizier: Distributed Infrastructure and API for Reliable and Flexible Blackbox Optimization [article]

Xingyou Song, Sagi Perel, Chansoo Lee, Greg Kochanski, Daniel Golovin
2022 arXiv   pre-print
Acknowledgements The Vizier team consists of: Xingyou Song, Sagi Perel, Chansoo Lee, Greg Kochanski, Richard Zhang, Tzu-Kuo Huang, Setareh Ariafar, Lior Belenki, Daniel Golovin, and Adrian Reyes.  ...  We further thank Emily Fertig, Srinivas Vasudevan, Jacob Burnim, Brian Patton, Ben Lee, Christopher Suter for Tensor ow Probability integrations, Daiyi Peng for PyGlove integrations, Yingjie Miao for AutoRL  ...  ., 2002) , Fire y (Yang, 2010) , and Harmony Search (Lee and Geem, 2005) to name a few (For a survey on meta-heuristics, see Beheshti and Shamsuddin (2013) ).  ... 
arXiv:2207.13676v1 fatcat:hahuqsg4wvetjhflmgsvpbbiym

Efficient Shot Detector: Lightweight Network Based on Deep Learning Using Feature Pyramid

Chansoo Park, Sanghun Lee, Hyunho Han
2021 Applied Sciences  
Convolutional-neural-network (CNN)-based methods are continuously used in various industries with the rapid development of deep learning technologies. However, an inference efficiency problem was reported in applications that require real-time performance, such as a mobile device. It is important to design a lightweight network that can be used in general-purpose environments such as mobile environments and GPU environments. In this study, we propose a lightweight network efficient shot
more » ... (ESDet) based on deep training with small parameters. The feature extraction process was performed using depthwise and pointwise convolution to minimize the computational complexity of the proposed network. The subsequent layer was formed in a feature pyramid structure to ensure that the extracted features were robust to multiscale objects. The network was trained by defining a prior box optimized for the data set of each feature scale. We defined an ESDet-baseline with optimal parameters through experiments and expanded it by gradually increasing the input resolution for detection accuracy. ESDet training and evaluation was performed using the PASCAL VOC and MS COCO2017 Dataset. Moreover, the average precision (AP) evaluation index was used for quantitative evaluation of detection performance. Finally, superior detection efficiency was demonstrated through the experiment compared to the conventional detection method.
doi:10.3390/app11188692 fatcat:zs6s3npnjvavjdtjxjfmvelj2a

Real-time device-scale imaging of conducting filament dynamics in resistive switching materials

Keundong Lee, Youngbin Tchoe, Hosang Yoon, Hyeonjun Baek, Kunook Chung, Sangik Lee, Chansoo Yoon, Bae Ho Park, Gyu-Chul Yi
2016 Scientific Reports  
ReRAM is a compelling candidate for next-generation non-volatile memory owing to its various advantages. However, fluctuation of operation parameters are critical weakness occurring failures in 'reading' and 'writing' operations. To enhance the stability, it is important to understand the mechanism of the devices. Although numerous studies have been conducted using AFM or TEM, the understanding of the device operation is still limited due to the destructive nature and/or limited imaging range
more » ... the previous methods. Here, we propose a new hybrid device composed of ReRAM and LED enabling us to monitor the conducting filament (CF) configuration on the device scale during resistive switching. We directly observe the change in CF configuration across the whole device area through light emission from our hybrid device. In contrast to former studies, we found that minor CFs were formed earlier than major CF contributing to the resistive switching. Moreover, we investigated the substitution of a stressed major CF with a fresh minor CF when large fluctuation of operation voltage appeared after more than 50 times of resistive switching in atmospheric condition. Our results present an advancement in the understanding of ReRAM operation mechanism, and a step toward stabilizing the fluctuations in ReRAM switching parameters.
doi:10.1038/srep27451 pmid:27271792 pmcid:PMC4895219 fatcat:stiix3oozfcfdcvidknga4c744

Towards Learning Universal Hyperparameter Optimizers with Transformers [article]

Yutian Chen, Xingyou Song, Chansoo Lee, Zi Wang, Qiuyi Zhang, David Dohan, Kazuya Kawakami, Greg Kochanski, Arnaud Doucet, Marc'aurelio Ranzato, Sagi Perel, Nando de Freitas
2022 arXiv   pre-print
Meta-learning hyperparameter optimization (HPO) algorithms from prior experiments is a promising approach to improve optimization efficiency over objective functions from a similar distribution. However, existing methods are restricted to learning from experiments sharing the same set of hyperparameters. In this paper, we introduce the OptFormer, the first text-based Transformer HPO framework that provides a universal end-to-end interface for jointly learning policy and function prediction when
more » ... trained on vast tuning data from the wild. Our extensive experiments demonstrate that the OptFormer can imitate at least 7 different HPO algorithms, which can be further improved via its function uncertainty estimates. Compared to a Gaussian Process, the OptFormer also learns a robust prior distribution for hyperparameter response functions, and can thereby provide more accurate and better calibrated predictions. This work paves the path to future extensions for training a Transformer-based model as a general HPO optimizer.
arXiv:2205.13320v1 fatcat:a2jpjiaftndp7jknav23oryjwq

Hardness of Online Sleeping Combinatorial Optimization Problems [article]

Satyen Kale and Chansoo Lee and Dávid Pál
2016 arXiv   pre-print
We show that several online combinatorial optimization problems that admit efficient no-regret algorithms become computationally hard in the sleeping setting where a subset of actions becomes unavailable in each round. Specifically, we show that the sleeping versions of these problems are at least as hard as PAC learning DNF expressions, a long standing open problem. We show hardness for the sleeping versions of Online Shortest Paths, Online Minimum Spanning Tree, Online k-Subsets, Online
more » ... cated Permutations, Online Minimum Cut, and Online Bipartite Matching. The hardness result for the sleeping version of the Online Shortest Paths problem resolves an open problem presented at COLT 2015 (Koolen et al., 2015).
arXiv:1509.03600v3 fatcat:o3w62rvyg5a6xiv4xlh3eboq4a

KIST-NOMAD - a Repository to Manage Large Amounts of Computational Materials Science Data

Samuel Boateng, Kwang Ryeol Lee, Deepika, Haneol Cho, Kyu Hwan Lee, Chansoo Kim
2020 Korean Journal of Metals and Materials  
Samuel Boateng, Kwang Ryeol Lee, Deepika, Haneol Cho, Kyu Hwan Lee, and Chansoo Kim 737 button on the search GUI after log in.  ...  Conclusion - Samuel Boateng: 학생, 이광열·이규환·김찬수: 연구원, Deepika·조한얼: 박사 후 연구원 *Corresponding Author: Chansoo Kim [Tel: +82-2-958-6448, E-mail:] *Corresponding Author: Kyu Hwan Lee [Tel: +82-2  ... 
doi:10.3365/kjmm.2020.58.10.728 fatcat:jij6xgfkinguritltaa2o4pqry

Ultra-thin resistive switching oxide layers self-assembled by field-induced oxygen migration (FIOM) technique

Sangik Lee, Inrok Hwang, Sungtaek Oh, Sahwan Hong, Yeonsoo Kim, Yoonseung Nam, Keundong Lee, Chansoo Yoon, Wondong Kim, Bae Ho Park
2014 Scientific Reports  
High-performance ultra-thin oxide layers are required for various next-generation electronic and optical devices. In particular, ultra-thin resistive switching (RS) oxide layers are expected to become fundamental building blocks of three-dimensional high-density non-volatile memory devices. Until now, special deposition techniques have been introduced for realization of high-quality ultra-thin oxide layers. Here, we report that ultra-thin oxide layers with reliable RS behavior can be
more » ... led by field-induced oxygen migration (FIOM) at the interface of an oxide-conductor/oxide-insulator or oxide-conductor/metal. The formation via FIOM of an ultra-thin oxide layer with a thickness of approximately 2-5 nm and 2.5% excess oxygen content is demonstrated using cross-sectional transmission electron microscopy and secondary ion mass spectroscopy depth profile. The observed RS behavior, such as the polarity dependent forming process, can be attributed to the formation of an ultra-thin oxide layer. In general, as oxygen ions are mobile in many oxide-conductors, FIOM can be used for the formation of ultra-thin oxide layers with desired properties at the interfaces or surfaces of oxide-conductors in high-performance oxide-based devices. O wing to its large effect on structural and electromagnetic properties, migration of cations and oxygen ions in oxide materials has been extensively investigated. In particular, the concentration and distribution of oxygen ions play an important role in determining the performance of electrical and optical devices 1-5 . Recently, in order to develop high-performance electronic devices based on emerging oxide materials, such as thin film transistors, oxide diodes, photovoltaic devices, memristors, resistive random access memories (ReRAMs), ferroelectric random access memories (FeRAMs) 5-10 , many researchers have focused their efforts on precisely controlling the concentration and distribution of oxygen ions through various deposition methods, including electric field modulation, post-annealing process, or buffered layer structure 11-17 . Among several electronic devices, ReRAMs using resistive switching (RS) behavior induced by an external electric stress show excellent advantages. For instance, they have simple and highly scalable two terminal structures, high resistive switching speed, low power consumption, and material diversity that includes oxides, polymers, and chalcogenides [12] [13] [14] [18] [19] [20] . In RS oxide materials, innovative performance and scaling-down require that RS phenomena are confined to nanostructures (such as nanoparticles, nanowires, and ultra-thin films) 21-26 . These phenomena are closely related to changes in concentration or distribution of oxygen ions under an external electric field. Specifically, unipolar RS is caused by a change in stoichiometry between transition metal and oxygen ions 9,12-13,27 , while bipolar RS is induced by oxygen migration into the active interfacial layer 3,5,22-24 . As a result, the ability to manipulate oxygen ions using an external electric field to fabricate self-assembled oxide nanostructures with RS behavior is very interesting and sought after, although major results are yet to be achieved. In this paper, we report on the fabrication of self-assembled ultra-thin oxide layers by the field-induced oxygen migration (FIOM) technique at oxide-conductor/oxide-insulator or oxide-conductor/metal junctions. Al-doped (4 wt %) ZnO (AZO) is a well-known transparent conductive oxide (TCO). Combining Al with ZnO leads to high conductivity, enhancing oxygen migration under external electric field and Joule heating, which reduce the activation energy for the migration of oxygen atoms [28] [29] [30] . Therefore, it is expected that an external electric field can easily induce migration of oxygen ions in AZO depending on the bias polarity. In AZO/NiO/Pt structures, an ultra-thin oxygen-rich (O-rich) AZO layer is formed at the AZO/NiO interface under negative bias polarity. Here, we report for the first time, to the best of our knowledge, the formation of a self-assembled ultra-thin O-rich AZO OPEN SUBJECT AREAS: ELECTRONIC AND SPINTRONIC DEVICES ELECTRONIC PROPERTIES AND MATERIALS
doi:10.1038/srep06871 pmid:25362933 pmcid:PMC4217097 fatcat:i2rz4jkvy5etpk7owbealwad3y

Understanding filamentary growth and rupture by Ag ion migration through single-crystalline 2D layered CrPS4

Mi Jung Lee, Sung-Hoon Kim, Sangik Lee, Chansoo Yoon, Kyung-Ah Min, Hyunsoo Choi, Suklyun Hong, Sungmin Lee, Je-Geun Park, Jae-Pyoung Ahn, Bae Ho Park
2020 NPG Asia Materials  
AbstractMemristive electrochemical metallization (ECM) devices based on cation migration and electrochemical metallization in solid electrolytes are considered promising for neuromorphic computing systems. Two-dimensional (2D) layered materials are emerging as potential candidates for electrolytes in reliable ECM devices due to their two-dimensionally confined material properties. However, electrochemical metallization within a single-crystalline 2D layered material has not yet been verified.
more » ... re, we use transmission electron microscopy and energy-dispersive X-ray spectroscopy to investigate the resistive switching mechanism of an ECM device containing a single-crystalline 2D layered CrPS4 electrolyte. We observe the various conductive filament (CF) configurations induced by an applied voltage in an Ag/CrPS4/Au device in the initial/low-resistance/high-resistance/breakdown states. These observations provide concrete experimental evidence that CFs consisting of Ag metal can be formed inside single-crystalline 2D layered CrPS4 and that their configuration can be changed by an applied voltage. Density functional theory calculations confirm that the sulfur vacancies in single-crystalline CrPS4 can facilitate Ag ion migration from the active electrode layer. The electrically induced changes in Ag CFs inside single-crystalline 2D layered CrPS4 raise the possibility of a reliable ECM device that exploits the properties of two-dimensionally confined materials.
doi:10.1038/s41427-020-00272-x fatcat:qomjxudurvcrxe3z3co4f2cevm

Enhancement of resistive switching under confined current path distribution enabled by insertion of atomically thin defective monolayer graphene

Keundong Lee, Inrok Hwang, Sangik Lee, Sungtaek Oh, Dukhyun Lee, Cheol Kyeom Kim, Yoonseung Nam, Sahwan Hong, Chansoo Yoon, Robert B. Morgan, Hakseong Kim, Sunae Seo (+3 others)
2015 Scientific Reports  
Resistive random access memory (ReRAM) devices have been extensively investigated resulting in significant enhancement of switching properties. However fluctuations in switching parameters are still critical weak points which cause serious failures during 'reading' and 'writing' operations of ReRAM devices. It is believed that such fluctuations may be originated by random creation and rupture of conducting filaments inside ReRAM oxides. Here, we introduce defective monolayer graphene between an
more » ... oxide film and an electrode to induce confined current path distribution inside the oxide film, and thus control the creation and rupture of conducting filaments. The ReRAM device with an atomically thin interlayer of defective monolayer graphene reveals much reduced fluctuations in switching parameters compared to a conventional one. Our results demonstrate that defective monolayer graphene paves the way to reliable ReRAM devices operating under confined current path distribution. Nowadays one of the most widely used nonvolatile memories is flash memory. It has been employed to numerous mobile devices and becomes a representative product supplied by the Si based semiconductor industries. The flash memory is expected to reach limitations in operating speed, power consumption, and density of memory in near future because it is a charge-storage type memory based on a Si transistor. To overcome the limitations, many researchers have tried to develop next generation nonvolatile memories (NG-NVM) with high performances, which do not rely on stored charges and Si transistors 1-4 . ReRAM whose resistive change is induced by applied external electrical stress is considered as one of these NG-NVMs. In addition to the advantageous properties of oxide-based ReRAM such as simple composition, facile fabrication process, and compatibility with conventional semiconductor processes, this resistance-change memory has presented promising nonvolatile memory effects including fast operation speed, low power consumption, and high scalability 5-9 . Resistive switching can be classified into two categories: Uni-polar and bi-polar resistive switching caused by conducting filament formation in bulk and oxygen migration at interface, respectively. Especially, uni-polar resistive switching, which is usually observed in binary oxide, has been explained
doi:10.1038/srep11279 pmid:26161992 pmcid:PMC4498384 fatcat:c6avn7dnqfd2bhnuhefvtzuyna
« Previous Showing results 1 — 15 out of 91 results