Triplet Distillation for Deep Face Recognition
Authors:
Yushu Feng,
Huan Wang,
Daniel T. Yi,
Roland Hu
Abstract:
Convolutional neural networks (CNNs) have achieved a great success in face recognition, which unfortunately comes at the cost of massive computation and storage consumption. Many compact face recognition networks are thus proposed to resolve this problem. Triplet loss is effective to further improve the performance of those compact models. However, it normally employs a fixed margin to all the sam…
▽ More
Convolutional neural networks (CNNs) have achieved a great success in face recognition, which unfortunately comes at the cost of massive computation and storage consumption. Many compact face recognition networks are thus proposed to resolve this problem. Triplet loss is effective to further improve the performance of those compact models. However, it normally employs a fixed margin to all the samples, which neglects the informative similarity structures between different identities. In this paper, we propose an enhanced version of triplet loss, named triplet distillation, which exploits the capability of a teacher model to transfer the similarity information to a small model by adaptively varying the margin between positive and negative pairs. Experiments on LFW, AgeDB, and CPLFW datasets show the merits of our method compared to the original triplet loss.
△ Less
Submitted 19 May, 2019; v1 submitted 11 May, 2019;
originally announced May 2019.
Chemical Enhancements in Shock-accelerated Particles: Ab-initio Simulations
Authors:
Damiano Caprioli,
Dennis T. Yi,
Anatoly Spitkovsky
Abstract:
We study the thermalization, injection, and acceleration of ions with different mass/charge ratios, $A/Z$, in non-relativistic collisionless shocks via hybrid (kinetic ions-fluid electrons) simulations. In general, ions thermalize to a post-shock temperature proportional to $A$. When diffusive shock acceleration is efficient, ions develop a non-thermal tail whose extent scales with $Z$ and whose n…
▽ More
We study the thermalization, injection, and acceleration of ions with different mass/charge ratios, $A/Z$, in non-relativistic collisionless shocks via hybrid (kinetic ions-fluid electrons) simulations. In general, ions thermalize to a post-shock temperature proportional to $A$. When diffusive shock acceleration is efficient, ions develop a non-thermal tail whose extent scales with $Z$ and whose normalization is enhanced as $(A/Z)^2$, so that incompletely-ionized heavy ions are preferentially accelerated. We discuss how these findings can explain observed heavy-ion enhancements in Galactic cosmic rays.
△ Less
Submitted 26 April, 2017;
originally announced April 2017.
