arXiv subjects

Yuchen Ding

Publications and source records attributed to Yuchen Ding.

59 records · Page 4Linked to original sources

arXiv2022

The Fornax3D project: discovery of ancient massive merger events in the Fornax cluster galaxies NGC 1380 and NGC 1427

We report the discovery of ancient massive merger events in the early-type galaxies NGC 1380 and NGC 1427, members of the Fornax galaxy cluster. Both galaxies have been observed by the MUSE IFU instrument on the VLT, as part of the Fornax3D project. By fitting recently-developed population-orbital superposition models to the observed surface brightness as well as stellar kinematic, age, and metallicity maps, we obtain the stellar orbits, age and metallicity distributions of each galaxy. We then decompose each galaxy into multiple orbital-based components, including a dynamically hot inner stellar halo component which is identified as the relic of past massive mergers. By comparing to analogues from cosmological galaxy simulations, chiefly TNG50, we find that the formation of such a hot inner stellar halo requires the merger with a now-destroyed massive satellite galaxy of $3.7_{-1.5}^{+2.7} \times 10^{10}$ Msun (about $1/5$ of its current stellar mass) in the case of NGC 1380 and of $1.5_{-0.7}^{+1.6} \times10^{10}$ Msun (about $1/4$ of its current stellar mass) in the case of NGC 1427. Moreover, we infer that the last massive merger in NGC 1380 happened $\sim10$ Gyr ago based on the stellar age distribution of the re-grown dynamically cold disk, whereas the merger in NGC 1427 ended $t\lesssim 8$ Gyr ago based on the stellar populations in its hot inner stellar halo. The major merger event in NGC 1380 is the first one with both merger mass and merger time quantitatively inferred in a galaxy beyond the Local Volume. Moreover, it is the oldest and most massive merger uncovered in nearby galaxies so far.

astro-ph.GA↗

arXiv2022

On a conjecture of Erdős

Let $\mathcal{P}$ denote the set of all primes. In 1950, P. Erdős conjectured that if $c$ is an arbitrarily given constant, $x$ is sufficiently large and $a_1,\dots , a_t$ are positive integers with $a_1 \log x$, then there exists an integer $n$ so that the number of solutions of $n=p+a_i$ $(p\in \mathcal{P}, 1\le i\le t)$ is greater than $c$. In this note, we confirm this old conjecture of Erdős.

math.NT↗

arXiv2021

RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering

In open-domain question answering, dense passage retrieval has become a new paradigm to retrieve relevant passages for finding answers. Typically, the dual-encoder architecture is adopted to learn dense representations of questions and passages for semantic matching. However, it is difficult to effectively train a dual-encoder due to the challenges including the discrepancy between training and inference, the existence of unlabeled positives and limited training data. To address these challenges, we propose an optimized training approach, called RocketQA, to improving dense passage retrieval. We make three major technical contributions in RocketQA, namely cross-batch negatives, denoised hard negatives and data augmentation. The experiment results show that RocketQA significantly outperforms previous state-of-the-art models on both MSMARCO and Natural Questions. We also conduct extensive experiments to examine the effectiveness of the three strategies in RocketQA. Besides, we demonstrate that the performance of end-to-end QA can be improved based on our RocketQA retriever.

cs.CL↗

arXiv2021

Real-time Face Mask Detection in Video Data

In response to the ongoing COVID-19 pandemic, we present a robust deep learning pipeline that is capable of identifying correct and incorrect mask-wearing from real-time video streams. To accomplish this goal, we devised two separate approaches and evaluated their performance and run-time efficiency. The first approach leverages a pre-trained face detector in combination with a mask-wearing image classifier trained on a large-scale synthetic dataset. The second approach utilizes a state-of-the-art object detection network to perform localization and classification of faces in one shot, fine-tuned on a small set of labeled real-world images. The first pipeline achieved a test accuracy of 99.97% on the synthetic dataset and maintained 6 FPS running on video data. The second pipeline achieved a mAP(0.5) of 89% on real-world images while sustaining 52 FPS on video data. We have concluded that if a larger dataset with bounding-box labels can be curated, this task is best suited using object detection architectures such as YOLO and SSD due to their superior inference speed and satisfactory performance on key evaluation metrics.

cs.CV↗

arXiv2018

A note on the sum of reciprocals

For a fixed positive integer $m$ and any partition $m = m_1 + m_2 + \cdots + m_e$ , there exists a sequence $\{n_{i}\}_{i=1}^{k}$ of positive integers such that $$m=\frac{1}{n_{1}}+\frac{1}{n_{2}}+\cdots+\frac{1}{n_{k}},$$ with the property that partial sums of the series $\{\frac{1}{n_i}\}_{i=1}^{k}$ can only represent the integers with the form $\sum_{i\in I}m_i$, where $I\subset\{1,...,e\}$.

math.NT↗