n-hot: Efficient bit-level sparsity for powers-of-two neural network quantization

no code implementations • 22 Mar 2021 • Yuiko Sakuma, Hiroshi Sumihiro, Jun Nishikawa, Toshiki Nakamura, Ryoji Ikegaya

Moreover, we use a two-stage fine-tuning algorithm to recover the accuracy drop that is triggered by introducing the bit-level sparsity.

object-detection Object Detection +1

First Detection of Hydroxyl Radical Emission from an Exoplanet Atmosphere: High-dispersion Characterization of WASP-33b using Subaru/IRD

no code implementations • 4 Mar 2021 • Stevanus K. Nugroho, Hajime Kawahara, Neale P. Gibson, Ernst J. W. de Mooij, Teruyuki Hirano, Takayuki Kotani, Yui Kawashima, Kento Masuda, Matteo Brogi, Jayne L. Birkby, Chris A. Watson, Motohide Tamura, Konstanze Zwintz, Hiroki Harakawa, Tomoyuki Kudo, Masayuki Kuzuhara, Klaus Hodapp, Masato Ishizuka, Shane Jacobson, Mihoko Konishi, Takashi Kurokawa, Jun Nishikawa, Masashi Omiya, Takuma Serizawa, Akitoshi Ueda, Sébastien Vievard

Nonetheless, this marginal detection is consistent with the prediction that H$_{2}$O is mostly thermally dissociated in the upper atmosphere of the ultra-hot Jupiters.

Earth and Planetary Astrophysics

Filter Pre-Pruning for Improved Fine-tuning of Quantized Deep Neural Networks

no code implementations • 13 Nov 2020 • Jun Nishikawa, Ryoji Ikegaya

Third, we propose a workflow of fine-tuning for quantized DNNs using the proposed pruning method(PfQ).

Quantization