Tae-Lim Choi

Bio: Tae-Lim Choi is an academic researcher from Seoul National University. The author has contributed to research in topics: Polymerization & Copolymer. The author has an hindex of 40, co-authored 171 publications receiving 6801 citations. Previous affiliations of Tae-Lim Choi include Samsung & UPRRP College of Natural Sciences.

A General Model for Selectivity in Olefin Cross Metathesis

Abstract: In recent years, olefin cross metathesis (CM) has emerged as a powerful and convenient synthetic technique in organic chemistry; however, as a general synthetic method, CM has been limited by the lack of predictability in product selectivity and stereoselectivity. Investigations into olefin cross metathesis with several classes of olefins, including substituted and functionalized styrenes, secondary allylic alcohols, tertiary allylic alcohols, and olefins with α-quaternary centers, have led to a general model useful for the prediction of product selectivity and stereoselectivity in cross metathesis. As a general ranking of olefin reactivity in CM, olefins can be categorized by their relative abilities to undergo homodimerization via cross metathesis and the susceptibility of their homodimers toward secondary metathesis reactions. When an olefin of high reactivity is reacted with an olefin of lower reactivity (sterically bulky, electron-deficient, etc.), selective cross metathesis can be achieved using fee...

Compression of Deep Convolutional Neural Networks for Fast and Low Power Mobile Applications

Abstract: Although the latest high-end smartphone has powerful CPU and GPU, running deeper convolutional neural networks (CNNs) for complex tasks such as ImageNet classification on mobile devices is challenging. To deploy deep CNNs on mobile devices, we present a simple and effective scheme to compress the entire CNN, which we call one-shot whole network compression. The proposed scheme consists of three steps: (1) rank selection with variational Bayesian matrix factorization, (2) Tucker decomposition on kernel tensor, and (3) fine-tuning to recover accumulated loss of accuracy, and each step can be easily implemented using publicly available tools. We demonstrate the effectiveness of the proposed scheme by testing the performance of various compressed CNNs (AlexNet, VGGS, GoogLeNet, and VGG-16) on the smartphone. Significant reductions in model size, runtime, and energy consumption are obtained, at the cost of small loss in accuracy. In addition, we address the important implementation level issue on 1?1 convolution, which is a key operation of inception module of GoogLeNet as well as CNNs compressed by our proposed scheme.

Synthesis and Activity of Ruthenium Alkylidene Complexes Coordinated with Phosphine and N-Heterocyclic Carbene Ligands

Abstract: This paper reports the synthesis and characterization of a variety of ruthenium complexes coordinated with phosphine and N-heterocyclic carbene (NHC) ligands. These complexes include several alkylidene derivatives of the general formula (NHC)(PR3)(Cl)2RuCHR‘, which are highly active olefin metathesis catalysts. Although these catalysts can be prepared adequately by the reaction of bis(phosphine) ruthenium alkylidene precursors with free NHCs, we have developed an alternative route that employs NHC-alcohol or -chloroform adducts as “protected” forms of the NHC ligands. This route is advantageous because NHC adducts are easier to handle than their free carbene counterparts. We also demonstrate that sterically bulky bis(NHC) complexes can be made by reaction of the pyridine-coordinated precursor (NHC)(py)2(Cl)2RuCHPh with free NHCs or NHC adducts. Two crystal structures are presented, one of the mixed bis(NHC) derivative (H2IMes)(IMes)(Cl)2RuCHPh, and the other of (PCy3)(Cl)(CO)Ru[η2-(CH2-C6H2Me2)(N2C3H4)(C6...

Compression of Deep Convolutional Neural Networks for Fast and Low Power Mobile Applications

Abstract: Although the latest high-end smartphone has powerful CPU and GPU, running deeper convolutional neural networks (CNNs) for complex tasks such as ImageNet classification on mobile devices is challenging. To deploy deep CNNs on mobile devices, we present a simple and effective scheme to compress the entire CNN, which we call one-shot whole network compression. The proposed scheme consists of three steps: (1) rank selection with variational Bayesian matrix factorization, (2) Tucker decomposition on kernel tensor, and (3) fine-tuning to recover accumulated loss of accuracy, and each step can be easily implemented using publicly available tools. We demonstrate the effectiveness of the proposed scheme by testing the performance of various compressed CNNs (AlexNet, VGGS, GoogLeNet, and VGG-16) on the smartphone. Significant reductions in model size, runtime, and energy consumption are obtained, at the cost of small loss in accuracy. In addition, we address the important implementation level issue on 1?1 convolution, which is a key operation of inception module of GoogLeNet as well as CNNs compressed by our proposed scheme.

Aggregated Residual Transformations for Deep Neural Networks

Abstract: We present a simple, highly modularized network architecture for image classification. Our network is constructed by repeating a building block that aggregates a set of transformations with the same topology. Our simple design results in a homogeneous, multi-branch architecture that has only a few hyper-parameters to set. This strategy exposes a new dimension, which we call cardinality (the size of the set of transformations), as an essential factor in addition to the dimensions of depth and width. On the ImageNet-1K dataset, we empirically show that even under the restricted condition of maintaining complexity, increasing cardinality is able to improve classification accuracy. Moreover, increasing cardinality is more effective than going deeper or wider when we increase the capacity. Our models, named ResNeXt, are the foundations of our entry to the ILSVRC 2016 classification task in which we secured 2nd place. We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its ResNet counterpart. The code and models are publicly available online.

N-heterocyclic carbenes: a new concept in organometallic catalysis.

Abstract: N-Heterocyclic carbenes have become universal ligands in organometallic and inorganic coordination chemistry. They not only bind to any transition metal, be it in low or high oxidation states, but also to main group elements such as beryllium, sulfur, and iodine. Because of their specific coordination chemistry, N-heterocyclic carbenes both stabilize and activate metal centers in quite different key catalytic steps of organic syntheses, for example, C-H activation, C-C, C-H, C-O, and C-N bond formation. There is now ample evidence that in the new generation of organometallic catalysts the established ligand class of organophosphanes will be supplemented and, in part, replaced by N-heterocyclic carbenes. Over the past few years, this chemistry has been the field of vivid scientific competition, and yielded previously unexpected successes in key areas of homogeneous catalysis. From the work in numerous academic laboratories and in industry, a revolutionary turning point in oraganometallic catalysis is emerging.

Aggregated Residual Transformations for Deep Neural Networks

Abstract: We present a simple, highly modularized network architecture for image classification. Our network is constructed by repeating a building block that aggregates a set of transformations with the same topology. Our simple design results in a homogeneous, multi-branch architecture that has only a few hyper-parameters to set. This strategy exposes a new dimension, which we call "cardinality" (the size of the set of transformations), as an essential factor in addition to the dimensions of depth and width. On the ImageNet-1K dataset, we empirically show that even under the restricted condition of maintaining complexity, increasing cardinality is able to improve classification accuracy. Moreover, increasing cardinality is more effective than going deeper or wider when we increase the capacity. Our models, named ResNeXt, are the foundations of our entry to the ILSVRC 2016 classification task in which we secured 2nd place. We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its ResNet counterpart. The code and models are publicly available online.

Interparticle Coupling Effect on the Surface Plasmon Resonance of Gold Nanoparticles: From Theory to Applications

Abstract: 12.2.2. Four-Wave Mixing (FWM) 4849 12.2.3. Dye Aggregation 4850 12.2.4. Optoelectronic Nanodevices 4850 12.3. Sensor 4851 12.3.1. Chemical Sensor 4851 12.3.2. Biological Sensor 4851 12.4. Catalysis 4852 13. Conclusion and Perspectives 4852 14. Abbreviations 4853 15. Acknowledgements 4854 16. References 4854 * Corresponding author E-mail: tpal@chem.iitkgp.ernet.in. † Raidighi College. § Indian Institute of Technology. 4797 Chem. Rev. 2007, 107, 4797−4862