User profiles for Kenji Kawaguchi
Kenji KawaguchiPresidential Young Professor, National University of Singapore Verified email at nus.edu.sg Cited by 6315 |
Deep learning without poor local minima
K Kawaguchi - Advances in neural information processing …, 2016 - proceedings.neurips.cc
In this paper, we prove a conjecture published in 1989 and also partially address an open
problem announced at the Conference on Learning Theory (COLT) 2015. For an expected …
problem announced at the Conference on Learning Theory (COLT) 2015. For an expected …
How does mixup help with robustness and generalization?
Mixup is a popular data augmentation technique based on taking convex combinations of
pairs of examples and their labels. This simple technique has been shown to substantially …
pairs of examples and their labels. This simple technique has been shown to substantially …
Lymphoplasmacytic sclerosing pancreatitis with cholangitis: a variant of primary sclerosing cholangitis extensively involving pancreas
K Kawaguchi, M Koike, K Tsuruta, A Okamoto, I Tabata… - Human pathology, 1991 - Elsevier
Pancreatic involvement in primary sclerosing cholangitis (PSC) is an extremely rare condition,
and its pathologic features are poorly documented. We report two cases of an unusual …
and its pathologic features are poorly documented. We report two cases of an unusual …
Early detection of DNA strand breaks in the brain after transient focal ischemia: implications for the role of DNA damage in apoptosis and neuronal cell death
…, K Jin, M Chen, W Pei, K Kawaguchi… - Journal of …, 1997 - Wiley Online Library
Using in situ DNA polymerase I‐mediated biotin‐dATP nick‐translation (PANT) and terminal
deoxynucleotidyl‐transferase‐mediated dUTP nick end‐labeling (TUNEL), we investigated …
deoxynucleotidyl‐transferase‐mediated dUTP nick end‐labeling (TUNEL), we investigated …
Elimination of all bad local minima in deep learning
K Kawaguchi, L Kaelbling - International Conference on …, 2020 - proceedings.mlr.press
In this paper, we theoretically prove that adding one special neuron per output unit eliminates
all suboptimal local minima of any deep neural network, for multi-class classification, …
all suboptimal local minima of any deep neural network, for multi-class classification, …
[HTML][HTML] Interpolation consistency training for semi-supervised learning
We introduce Interpolation Consistency Training (ICT), a simple and computation efficient
algorithm for training Deep Neural Networks in the semi-supervised learning paradigm. ICT …
algorithm for training Deep Neural Networks in the semi-supervised learning paradigm. ICT …
Adaptive activation functions accelerate convergence in deep and physics-informed neural networks
We employ adaptive activation functions for regression in deep and physics-informed neural
networks (PINNs) to approximate smooth and discontinuous functions as well as solutions …
networks (PINNs) to approximate smooth and discontinuous functions as well as solutions …
Nucleotide-dependent single-to double-headed binding of kinesin
K Kawaguchi, S Ishiwata - Science, 2001 - science.org
The motility of kinesin motors is explained by a “hand-over-hand” model in which two heads
of kinesin alternately repeat single-headed and double-headed binding with a microtubule. …
of kinesin alternately repeat single-headed and double-headed binding with a microtubule. …
Preparation of carbon quantum dots with tunable photoluminescence by rapid laser passivation in ordinary organic solvents
…, Y Shimizu, A Pyatenko, K Kawaguchi… - Chemical …, 2011 - pubs.rsc.org
A simple approach to prepare carbon quantum dots is presented in this communication by
laser rapid passivation of nano carbon particles in ordinary organic solvent. The as-prepared …
laser rapid passivation of nano carbon particles in ordinary organic solvent. The as-prepared …
Generalization in deep learning
This chapter provides theoretical insights into why and how deep learning can generalize
well, despite its large capacity, complexity, possible algorithmic instability, non-robustness, …
well, despite its large capacity, complexity, possible algorithmic instability, non-robustness, …