arXiv subjects

Boqiang Huang

Publications and source records attributed to Boqiang Huang.

6 recordsLinked to original sources

arXiv2026

Optical vortex classification via machine learning

Optical vortices carry quantized phase information (topological charge) and are considered candidates for information processing in all-optical circuits. Accurately identifying the quantized vortex charge in a way that is most efficient is essential for data processing. Here, we demonstrate that using only intensity information machine learning algorithms are able to classify vortices into distinct phase categories using a properly trained model. Preprocessing of the original intensity data leads to an improved prediction accuracy of the trained classifier and a much shorter training time (~3 orders of magnitude faster). The learning mechanism of the algorithms is revealed by the statistical analyses based on calculation of Cohen's d. We also find that the training efficiency is associated with nonlinearity, non-Hermiticity, and the vortex shape. In some cases, the trained classifier performs well across distinct physical models. Our findings will benefit and accelerate vorticity-based binary information processing and can also be extended to other physical systems.

physics.optics↗

arXiv2024

Continual Learning in Medical Image Analysis: A Comprehensive Review of Recent Advancements and Future Prospects

Medical imaging analysis has witnessed remarkable advancements even surpassing human-level performance in recent years, driven by the rapid development of advanced deep-learning algorithms. However, when the inference dataset slightly differs from what the model has seen during one-time training, the model performance is greatly compromised. The situation requires restarting the training process using both the old and the new data which is computationally costly, does not align with the human learning process, and imposes storage constraints and privacy concerns. Alternatively, continual learning has emerged as a crucial approach for developing unified and sustainable deep models to deal with new classes, tasks, and the drifting nature of data in non-stationary environments for various application areas. Continual learning techniques enable models to adapt and accumulate knowledge over time, which is essential for maintaining performance on evolving datasets and novel tasks. This systematic review paper provides a comprehensive overview of the state-of-the-art in continual learning techniques applied to medical imaging analysis. We present an extensive survey of existing research, covering topics including catastrophic forgetting, data drifts, stability, and plasticity requirements. Further, an in-depth discussion of key components of a continual learning framework such as continual learning scenarios, techniques, evaluation schemes, and metrics is provided. Continual learning techniques encompass various categories, including rehearsal, regularization, architectural, and hybrid strategies. We assess the popularity and applicability of continual learning categories in various medical sub-fields like radiology and histopathology...

eess.IV↗

arXiv2020

Deep Learning for Feynman's Path Integral in Strong-Field Time-Dependent Dynamics

Feynman's path integral approach is to sum over all possible spatio-temporal paths to reproduce the quantum wave function and the corresponding time evolution, which has enormous potential to reveal quantum processes in classical view. However, the complete characterization of quantum wave function with infinite paths is a formidable challenge, which greatly limits the application potential, especially in the strong-field physics and attosecond science. Instead of brute-force tracking every path one by one, here we propose deep-learning-performed strong-field Feynman's formulation with pre-classification scheme which can predict directly the final results only with data of initial conditions, so as to attack unsurmountable tasks by existing strong-field methods and explore new physics. Our results build up a bridge between deep learning and strong-field physics through the Feynman's path integral, which would boost applications of deep learning to study the ultrafast time-dependent dynamics in strong-field physics and attosecond science, and shed a new light on the quantum-classical correspondence.

physics.atom-ph↗

arXiv2015

A different view on the vector-valued empirical mode decomposition (VEMD)

The empirical mode decomposition (EMD) has achieved its reputation by providing a multi-scale time-frequency representation of nonlinear and/or nonstationary signals. To extend this method to vector-valued signals (VvS) in multidimensional (multi-D) space, a multivariate EMD (MEMD) has been designed recently, which employs an ensemble projection to extract local extremum locations (LELs) of the given VvS with respect to different projection directions. This idea successfully overcomes the problems of locally defining extrema of VvS. Different from the MEMD, where vector-valued envelopes (VvEs) are interpolated based on LELs extracted from the 1-D projected signal, the vector-valued EMD (VEMD) proposed in this paper employs a novel back projection method to interpolate the VvEs from 1-D envelopes in the projected space. Considering typical 4-D coordinates (3-D location and time), we show by numerical simulations that the VEMD outperforms state-of-art methods.

math.NA↗

arXiv2014

A unique polar representation of the hyperanalytic signal

The hyperanalytic signal is the straight forward generalization of the classical analytic signal. It is defined by a complexification of two canonical complex signals, which can be considered as an inverse operation of the Cayley-Dickson form of the quaternion. Inspired by the polar form of an analytic signal where the real instantaneous envelope and phase can be determined, this paper presents a novel method to generate a polar representation of the hyperanalytic signal, in which the continuously complex envelope and phase can be uniquely defined. Comparing to other existing methods, the proposed polar representation does not have sign ambiguity between the envelope and the phase, which makes the definition of the instantaneous complex frequency possible.

math.NA↗

arXiv2013

Nonlocal imaging by conditional averaging of random reference measurements

We report the nonlocal imaging of an object by conditional averaging of the random exposure frames of a reference detector, which only sees the freely propagating field from a thermal light source. A bucket detector, synchronized with the reference detector, records the intensity fluctuations of an identical beam passing through the object mask. These fluctuations are sorted according to their values relative to the mean, then the reference data in the corresponding time-bins for a given fluctuation range are averaged, to produce either positive or negative images. Since no correlation calculations are involved, this correspondence imaging technique challenges our former interpretations of "ghost" imaging. Compared with conventional correlation imaging or compressed sensing schemes, both the number of exposures and computation time are greatly reduced, while the visibility is much improved. A simple statistical model is presented to explain the phenomenon.

physics.optics↗