Search

Despite the tremendous advances achieved over the past years by deep learning techniques, the latest risk prediction models for industrial applications still rely on highly handtuned stage-wised statistical learning tools, such as gradient boosting and random forest methods. Different from images or languages, real-world financial data are high-dimensional, sparse, noisy and extremely imbalanced, which makes deep neural network models particularly challenging to train and fragile in practice. In this work, we propose DeRisk, an effective deep learning risk prediction framework for credit risk prediction on real-world financial data. DeRisk is the first deep risk prediction model that outperforms statistical learning approaches deployed in our company's production system. We also perform extensive ablation studies on our method to present the most critical factors for the empirical success of DeRisk.

Despite having data for over 10^9 stars from Gaia, only less than 10^4 star clusters and candidates have been discovered. Particularly, distant star clusters are rarely identified, due to the challenges posed by heavy extinction and great distance. However, Gaia data has continued to improve, enabling even fainter cluster members to be distinguished from field stars. In this work, we will introduce a star cluster search method based on the DBSCAN algorithm; we have made improvements to make it better suited for identifying clusters on dimmer and more distant stars. After removing member stars of known Gaia-based clusters, we have identified 2086 objects with |b|<10 deg, of which 1488 are highly reliable open star clusters, along with 569 candidates, 28 globular cluster candidates and 1 irregular galaxy IC 10 at low Galactic latitudes. We found that the proper motion of IC10 is similar yet slightly different from the water maser observations, which is an important result for the comparison with Gaia and VLBA. Besides, when compared with the star clusters appearing in Gaia DR2/EDR3, we have found nearly three times as many new objects above a distance of 5 kpc, including hundreds of them above Av > 5 mag. And it has enabled us to detect a higher number of old clusters, over a billion years old, that are difficult to detect due to observational limitations. Our findings significantly expand the remote cluster sample and enhance our understanding of the limits of Gaia DR3 data in stellar aggregates research. The full figure set for 2085 clusters can be seen in \url{https://nadc.china-vo.org/res/r101258/}, Comment: 18 pages, a url link with a full figure set (.pdf file with 88 pages) for 2085 distant clusters in abstract, accepted by ApJS

We report 1,656 new star clusters found in the Galactic disk (|b|<20 degrees) beyond 1.2 kpc, using Gaia EDR3 data. Based on an unsupervised machine learning algorithm, DBSCAN, and followed our previous studies, we utilized a unique method to do the data preparation and obtained the clustering coefficients, which proved to be an effective way to search blindly for star clusters. We tabulated the physical parameters and member stars of the new clusters, and presented some interesting examples, including a globular cluster candidate. The cluster parameters and member stars are available at CDS via anonymous ftp to https://cdsarc.cds.unistra.fr/ftp/vizier.submit//he22c. We examined the new discoveries and discussed their statistical properties. The proper motion dispersions and radii of the new clusters were the same as the previously reported ones. The new star clusters beyond 1.2 kpc were older than those in the solar neighborhood, and the new objects found in the third Galactic quadrant presented the lowest line-of-sight extinctions. Combined with our previous results, the total population of new clusters detected through our method was 2,541, corresponding to 55% of all newly published clusters in the Gaia era. The number of cataloged Gaia star clusters was also increased to nearly six thousand. In the near future, it is necessary to make a unified confirmation and member star determination for all reported clusters., Comment: 16 pages, 11 figures, 3 tables with full clusters/members data link in CDS, accepted for publication in ApJS

Although previous searches for star clusters have been very successful, many clusters are likely still omitted, especially at high Galactic latitude regions. In this work, based on the astrometry of Gaia EDR3, we searched nearby (parallax > 0.8 mas) all-sky regions, obtaining 886 star clusters, of which 270 candidates have not been cataloged before. At the same time, we have presented the physical parameters of the clusters by fitting theoretical isochrones to their optical magnitudes. More halo members and expanding structures in many star clusters were also found. Most of the new objects are young clusters that are less than 100 million years old. Our work greatly increased the sample size and physical parameters of star clusters in the solar neighborhood, in particular, 46 clusters are newly found with |b| > 20 deg, which represents an increase of nearly three fold of cluster numbers at high Galactic latitude regions. The cluster parameters and member stars are available at CDS via https://cdsarc.u-strasbg.fr/ftp/vizier.submit//hezh22b/, and the cluster figure sets are available via https://doi.org/10.12149/101133., Comment: 16 figures, 2 tables (full version available at CDS), 886 figures in figure set, accepted for publication on ApJs

Few-shot abstractive summarization has become a challenging task in natural language generation. To support it, we designed a novel soft prompts architecture coupled with a prompt pre-training plus fine-tuning paradigm that is effective and tunes only extremely light parameters. The soft prompts include continuous input embeddings across an encoder and a decoder to fit the structure of the generation models. Importantly, a novel inner-prompt placed in the text is introduced to capture document-level information. The aim is to devote attention to understanding the document that better prompts the model to generate document-related content. The first step in the summarization procedure is to conduct prompt pre-training with self-supervised pseudo-data. This teaches the model basic summarizing capabilities. The model is then fine-tuned with few-shot examples. Experimental results on the CNN/DailyMail and XSum datasets show that our method, with only 0.1% of the parameters, outperforms full-model tuning where all model parameters are tuned. It also surpasses Prompt Tuning by a large margin and delivers competitive results against Prefix-Tuning with 3% of the parameters., Comment: Accepted as a long paper in COLING 2022

Autonomous vehicles use 3D sensors for perception. Cooperative perception enables vehicles to share sensor readings with each other to improve safety. Prior work in cooperative perception scales poorly even with infrastructure support. AutoCast enables scalable infrastructure-less cooperative perception using direct vehicle-to-vehicle communication. It carefully determines which objects to share based on positional relationships between traffic participants, and the time evolution of their trajectories. It coordinates vehicles and optimally schedules transmissions in a distributed fashion. Extensive evaluation results under different scenarios show that, unlike competing approaches, AutoCast can avoid crashes and near-misses which occur frequently without cooperative perception, its performance scales gracefully in dense traffic scenarios providing 2-4x visibility into safety critical objects compared to existing cooperative perception schemes, its transmission schedules can be completed on the real radio testbed, and its scheduling algorithm is near-optimal with negligible computation overhead.