arXiv subjects

I-Hsuan Li

Publications and source records attributed to I-Hsuan Li.

4 recordsLinked to original sources

arXiv2026

An azimuthally resolved study of sloshing cold fronts in three nearby galaxy clusters

We present a detailed analysis of sloshing cold fronts in a sample of three nearby galaxy clusters (Abell 496, Abell 2029, and Abell 1644) observed with the Chandra X-ray Observatory. Cold fronts manifest as sharp edges in the X-ray surface brightness of the intracluster medium (ICM) in galaxy clusters. In the residual X-ray surface brightness maps, where the global ICM distribution has been subtracted, cold fronts generated by gas sloshing are observed at the boundaries of the spiral excesses. We perform a systematic and comprehensive study of the surface brightness edges along the spiral excesses. We find the deficit of the thermal pressure radially inward of the brightness edges, in contrast to stripping cold fronts that typically exhibit higher thermal pressure in brightness edges. Assuming that the sharp edges in the X-ray surface brightness distributions are sustained entirely by the gas bulk motions, we estimate the velocity gradients across the edges that are required to compensate for the deficit of the thermal pressure. We do not find statistically significant velocity gradients along the azimuthal direction. Our results suggest that alternative mechanisms such as magnetic fields and viscosity are necessary to maintain the sharpness of sloshing cold fronts.

astro-ph.CO↗

arXiv2025

Dynamic Gradient Sparse Update for Edge Training

Training on edge devices enables personalized model fine-tuning to enhance real-world performance and maintain data privacy. However, the gradient computation for backpropagation in the training requires significant memory buffers to store intermediate features and compute losses. This is unacceptable for memory-constrained edge devices such as microcontrollers. To tackle this issue, we propose a training acceleration method using dynamic gradient sparse updates. This method updates the important channels and layers only and skips gradient computation for the less important channels and layers to reduce memory usage for each update iteration. In addition, the channel selection is dynamic for different iterations to traverse most of the parameters in the update layers along the time dimension for better performance. The experimental result shows that the proposed method enables an ImageNet pre-trained MobileNetV2 trained on CIFAR-10 to achieve an accuracy of 85.77\% while updating only 2\% of convolution weights within 256KB on-chip memory. This results in a remarkable 98\% reduction in feature memory usage compared to dense model training.

cs.AR↗

arXiv2021

Technical Report for Valence-Arousal Estimation in ABAW2 Challenge

In this work, we describe our method for tackling the valence-arousal estimation challenge from ABAW2 ICCV-2021 Competition. The competition organizers provide an in-the-wild Aff-Wild2 dataset for participants to analyze affective behavior in real-life settings. We use a two stream model to learn emotion features from appearance and action respectively. To solve data imbalanced problem, we apply label distribution smoothing (LDS) to re-weight labels. Our proposed method achieves Concordance Correlation Coefficient (CCC) of 0.591 and 0.617 for valence and arousal on the validation set of Aff-wild2 dataset.

cs.CV↗

arXiv2021

Technical Report for Valence-Arousal Estimation on Affwild2 Dataset

In this work, we describe our method for tackling the valence-arousal estimation challenge from ABAW FG-2020 Competition. The competition organizers provide an in-the-wild Aff-Wild2 dataset for participants to analyze affective behavior in real-life settings. We use MIMAMO Net \cite{deng2020mimamo} model to achieve information about micro-motion and macro-motion for improving video emotion recognition and achieve Concordance Correlation Coefficient (CCC) of 0.415 and 0.511 for valence and arousal on the reselected validation set.

cs.CV↗