arXiv subjects

Liangliang Shi

Publications and source records attributed to Liangliang Shi.

4 recordsLinked to original sources

arXiv2024

Double-Bounded Optimal Transport for Advanced Clustering and Classification

Optimal transport (OT) is attracting increasing attention in machine learning. It aims to transport a source distribution to a target one at minimal cost. In its vanilla form, the source and target distributions are predetermined, which contracts to the real-world case involving undetermined targets. In this paper, we propose Doubly Bounded Optimal Transport (DB-OT), which assumes that the target distribution is restricted within two boundaries instead of a fixed one, thus giving more freedom for the transport to find solutions. Based on the entropic regularization of DB-OT, three scaling-based algorithms are devised for calculating the optimal solution. We also show that our DB-OT is helpful for barycenter-based clustering, which can avoid the excessive concentration of samples in a single cluster. Then we further develop DB-OT techniques for long-tailed classification which is an emerging and open problem. We first propose a connection between OT and classification, that is, in the classification task, training involves optimizing the Inverse OT to learn the representations, while testing involves optimizing the OT for predictions. With this OT perspective, we first apply DB-OT to improve the loss, and the Balanced Softmax is shown as a special case. Then we apply DB-OT for inference in the testing process. Even with vanilla Softmax trained features, our extensive experimental results show that our method can achieve good results with our improved inference scheme in the testing stage.

cs.LG↗

arXiv2023

IID-GAN: an IID Sampling Perspective for Regularizing Mode Collapse

Despite its success, generative adversarial networks (GANs) still suffer from mode collapse, i.e., the generator can only map latent variables to a partial set of modes in the target distribution. In this paper, we analyze and seek to regularize this issue with an independent and identically distributed (IID) sampling perspective and emphasize that holding the IID property referring to the target distribution for generation can naturally avoid mode collapse. This is based on the basic IID assumption for real data in machine learning. However, though the source samples {z} obey IID, the generations {G(z)} may not necessarily be IID sampling from the target distribution. Based on this observation, considering a necessary condition of IID generation that the inverse samples from target data should also be IID in the source distribution, we propose a new loss to encourage the closeness between inverse samples of real data and the Gaussian source in latent space to regularize the generation to be IID from the target distribution. Experiments on both synthetic and real-world data show the effectiveness of our model.

cs.LG↗

arXiv2019

Acoustic funnel and buncher for nanoparticle injection

Acoustics-based techniques are investigated to focus and bunch nanoparticle beams. This allows to overcome the prominent problem of the longitudinal and transverse mismatch of particle stream and x-ray beam in single-particle/single molecule imaging at x-ray free-electron lasers (XFEL). It will also enable synchronised injection of particle streams at kHz repetition rates. Transverse focusing concentrates the particle flux to the (sub)micrometer x-ray focus. In the longitudinal direction, focused acoustic waves can be used to bunch the particle to the same repetition rate as the x-ray pulses. The acoustic manipulation is based on simple mechanical recoil effects and could be advantageous over light-pressure-based methods, which rely on absorption. The acoustic equipment is easy to implement and can be conveniently inserted into current XFEL endstations. With the proposed method, data collection times could be reduced by a factor of $10^4$. This work does not just provide an efficient method for acoustic manipulation of streams of arbitrary gas phase particles, but also opens up wide avenues for acoustics-based particle optics.

physics.app-ph↗

arXiv2018

Higher-order mode-based cavity misalignment measurements at the free-electron laser FLASH

At the Free-Electron Laser in Hamburg (FLASH) and the European X-Ray Free-Electron Laser, superconducting TeV-energy superconducting linear accelerator (TESLA)-type cavities are used for the acceleration of electron bunches, generating intense free-electron laser (FEL) beams. A long rf pulse structure allows one to accelerate long bunch trains, which considerably increases the efficiency of the machine. However, intrabunch-train variations of rf parameters and misalignments of rf structures induce significant trajectory variations that may decrease the FEL performance. The accelerating cavities are housed inside cryomodules, which restricts the ability for direct alignment measurements. In order to determine the transverse cavity position, we use a method based on beam-excited dipole modes in the cavities. We have developed an efficient measurement and signal processing routine and present its application to multiple accelerating modules at FLASH. The measured rms cavity offset agrees with the specification of the TESLA modules. For the first time, the tilt of a TESLA cavity inside a cryomodule is measured. The preliminary result agrees well with the ratio between the offset and angle dependence of the dipole mode which we calculated with eigenmode simulations.

physics.acc-ph↗