Three tetra-armed cyclens with two kinds of side-arms, 3',5'-difluorobenzyl/4'-methylbenzyl, 3',5'-difluorobenzyl/1'-naphthylmethyl, and 3',5'-difluorobenzyl/9'-anthrylmethyl groups, were prepared by reductive amination of 1,7-bis(3',5'-difluorobenzyl)-1,4,7,10-tetraazacyclododecane and the corresponding aromatic aldehydes in the presence of NaBH(OAc)3. The X-ray structures of the Ag(+) complexes and Ag(+)-ion-induced (1)H NMR spectral changes suggest that (i) the chemical shift changes of the protons at the 2'- and 6'-positions in the 3',5'-difluorobenzyl/4'-methylbenzyl side-arms are dependent on the electron density on the adjacent substituted benzenes, and (ii) in the tetra-armed cyclens with 3',5'-difluorobenzyl/1'-naphthylmethyl and 3',5'-difluorobenzyl/9'-anthrylmethyl groups as side-arms, electron-rich aromatic rings preferentially cover the Ag(+) ions incorporated into the ligand cavities, and 3',5'-difluorobenzyl groups do not participate in the Ag(+) interactions. The log K values were estimated using Ag(+)-ion-induced UV-vis spectral changes.
NVIDIA cuDNN is a low-level library that provides GPU kernels frequently used in deep learning. Specifically, cuDNN implements several equivalent convolution algorithms, whose performance and memory footprint may vary considerably, depending on the layer dimensions. When an algorithm is automatically selected by cuDNN, the decision is performed on a per-layer basis, and thus it often resorts to slower algorithms that fit the workspace size constraints. We present µ-cuDNN, a transparent wrapper library for cuDNN, which divides layers' mini-batch computation into several micro-batches. Based on Dynamic Programming and Integer Linear Programming, µ-cuDNN enables faster algorithms by decreasing the workspace requirements. At the same time, µ-cuDNN keeps the computational semantics unchanged, so that it decouples statistical efficiency from the hardware efficiency safely. We demonstrate the effectiveness of µ-cuDNN over two frameworks, Caffe and TensorFlow, achieving speedups of 1.63x for AlexNet and 1.21x for ResNet-18 on P100-SXM2 GPU. These results indicate that using micro-batches can seamlessly increase the performance of deep learning, while maintaining the same memory footprint.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.