Searcharxiv⌕ Search

arXiv subjects

Jing Lu

Publications and source records attributed to Jing Lu.

At least 109 records · Page 6Linked to original sources
arXiv2023

Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework

With the proliferation of short video applications, the significance of short video recommendations has vastly increased. Unlike other recommendation scenarios, short video recommendation systems heavily rely on feedback from watch time. Existing approaches simply treat watch time as a direct label, failing to effectively harness its extensive semantics and introduce bias, thereby limiting the potential for modeling user interests based on watch time. To overcome this challenge, we propose a framework named Debiased Multiple-semantics-extracting Labeling(DML). DML constructs labels that encompass various semantics by utilizing quantiles derived from the distribution of watch time, prioritizing relative order rather than absolute label values. This approach facilitates easier model learning while aligning with the ranking objective of recommendations. Furthermore, we introduce a method inspired by causal adjustment to refine label definitions, thereby directly mitigating bias at the label level. We substantiate the effectiveness of our DML framework through both online and offline experiments. Extensive results demonstrate that our DML could effectively leverage watch time to discover users' real interests, enhancing their engagement in our application.

cs.IR↗
arXiv2023

Ground-based and JWST Observations of SN 2022pul: I. Unusual Signatures of Carbon, Oxygen, and Circumstellar Interaction in a Peculiar Type Ia Supernova

Nebular-phase observations of peculiar Type Ia supernovae (SNe Ia) provide important constraints on progenitor scenarios and explosion dynamics for both these rare SNe and the more common, cosmologically useful SNe Ia. We present observations from an extensive ground-based and space-based follow-up campaign to characterize SN 2022pul, a "super-Chandrasekhar" mass SN Ia (alternatively "03fg-like" SN), from before peak brightness to well into the nebular phase across optical to mid-infrared (MIR) wavelengths. The early rise of the light curve is atypical, exhibiting two distinct components, consistent with SN Ia ejecta interacting with dense carbon-oxygen rich circumstellar material (CSM). In the optical, SN 2022pul is most similar to SN 2012dn, having a low estimated peak luminosity ($M_{B}=-18.9$ mag) and high photospheric velocity relative to other 03fg-like SNe. In the nebular phase, SN 2022pul adds to the increasing diversity of the 03fg-like subclass. From 168 to 336 days after peak $B$-band brightness, SN 2022pul exhibits asymmetric and narrow emission from [O I] $λλ6300,\ 6364$ (${\rm FWHM} \approx 2{,}000$ km s$^{-1}$), strong, broad emission from [Ca II] $λλ7291,\ 7323$ (${\rm FWHM} \approx 7{,}300$ km s$^{-1}$), and a rapid Fe III to Fe II ionization change. Finally, we present the first-ever optical-to-mid-infrared (MIR) nebular spectrum of an 03fg-like SN Ia using data from JWST. In the MIR, strong lines of neon and argon, weak emission from stable nickel, and strong thermal dust emission (with $T \approx 500$ K), combined with prominent [O I] in the optical, suggest that SN 2022pul was produced by a white dwarf merger within carbon/oxygen-rich CSM.

astro-ph.HE↗
arXiv2023

Al-Based Few-Hydrogen Metal-Bonded Perovskite High-$T_c$ Superconductor Al$_4$H up to 54 K under Atmospheric Pressure

Multi-hydrogen lanthanum hydrides have shown the highest critical temperature $T_c$ at 250-260 K under 170-200 GPa. However, such high pressure is a great challenge for sample preparation and practical application. To address this challenge, we propose a novel design strategy for ambient-pressure high-$T_c$ superconductors by constructing new few-hydrogen metal-bonded perovskite hydrides, such as Al-based superconductor $\rm {Al_4H}$, with better ductility than the well-known multi-hydrogen, cuprate and Fe-based superconductors. Based on the Migdal-Eliashberg theory, we predict that the structurally stable $\rm {Al_4H}$ has a favorable high $T_c$ up to 54 K under atmospheric pressure, similar to SmOFeAs.

cond-mat.supr-con↗
arXiv2023

Fast and Not-so-Furious: Case Study of the Fast and Faint Type IIb SN 2021bxu

We present photometric and spectroscopic observations and analysis of SN 2021bxu (ATLAS21dov), a low-luminosity, fast-evolving Type IIb supernova (SN). SN 2021bxu is unique, showing a large initial decline in brightness followed by a short plateau phase. With $M_r = -15.93 \pm 0.16\, \mathrm{mag}$ during the plateau, it is at the lower end of the luminosity distribution of stripped-envelope supernovae (SE-SNe) and shows a distinct $\sim$10 day plateau not caused by H- or He-recombination. SN 2021bxu shows line velocities which are at least $\sim1500\,\mathrm{km\,s^{-1}}$ slower than typical SE-SNe. It is photometrically and spectroscopically similar to Type IIb SNe during the photospheric phases of evolution, with similarities to Ca-rich IIb SNe. We find that the bolometric light curve is best described by a composite model of shock interaction between the ejecta and an envelope of extended material, combined with a typical SN IIb powered by the radioactive decay of $^{56}$Ni. The best-fit parameters for SN 2021bxu include a $^{56}$Ni mass of $M_{\mathrm{Ni}} = 0.029^{+0.004}_{-0.005}\,\mathrm{M_{\odot}}$, an ejecta mass of $M_{\mathrm{ej}} = 0.61^{+0.06}_{-0.05}\,\mathrm{M_{\odot}}$, and an ejecta kinetic energy of $K_{\mathrm{ej}} = 8.8^{+1.1}_{-1.0} \times 10^{49}\, \mathrm{erg}$. From the fits to the properties of the extended material of Ca-rich IIb SNe we find a trend of decreasing envelope radius with increasing envelope mass. SN 2021bxu has $M_{\mathrm{Ni}}$ on the low end compared to SE-SNe and Ca-rich SNe in the literature, demonstrating that SN 2021bxu-like events are rare explosions in extreme areas of parameter space. The progenitor of SN 2021bxu is likely a low mass He star with an extended envelope.

astro-ph.HE↗
arXiv2023

TWIN: TWo-stage Interest Network for Lifelong User Behavior Modeling in CTR Prediction at Kuaishou

Life-long user behavior modeling, i.e., extracting a user's hidden interests from rich historical behaviors in months or even years, plays a central role in modern CTR prediction systems. Conventional algorithms mostly follow two cascading stages: a simple General Search Unit (GSU) for fast and coarse search over tens of thousands of long-term behaviors and an Exact Search Unit (ESU) for effective Target Attention (TA) over the small number of finalists from GSU. Although efficient, existing algorithms mostly suffer from a crucial limitation: the \textit{inconsistent} target-behavior relevance metrics between GSU and ESU. As a result, their GSU usually misses highly relevant behaviors but retrieves ones considered irrelevant by ESU. In such case, the TA in ESU, no matter how attention is allocated, mostly deviates from the real user interests and thus degrades the overall CTR prediction accuracy. To address such inconsistency, we propose \textbf{TWo-stage Interest Network (TWIN)}, where our Consistency-Preserved GSU (CP-GSU) adopts the identical target-behavior relevance metric as the TA in ESU, making the two stages twins. Specifically, to break TA's computational bottleneck and extend it from ESU to GSU, or namely from behavior length $10^2$ to length $10^4-10^5$, we build a novel attention mechanism by behavior feature splitting. For the video inherent features of a behavior, we calculate their linear projection by efficient pre-computing \& caching strategies. And for the user-item cross features, we compress each into a one-dimentional bias term in the attention score calculation to save the computational cost. The consistency between two stages, together with the effective TA-based relevance metric in CP-GSU, contributes to significant performance gain in CTR prediction.

cs.IR↗
arXiv2023

Harmonic enhancement using learnable comb filter for light-weight full-band speech enhancement model

With fewer feature dimensions, filter banks are often used in light-weight full-band speech enhancement models. In order to further enhance the coarse speech in the sub-band domain, it is necessary to apply a post-filtering for harmonic retrieval. The signal processing-based comb filters used in RNNoise and PercepNet have limited performance and may cause speech quality degradation due to inaccurate fundamental frequency estimation. To tackle this problem, we propose a learnable comb filter to enhance harmonics. Based on the sub-band model, we design a DNN-based fundamental frequency estimator to estimate the discrete fundamental frequencies and a comb filter for harmonic enhancement, which are trained via an end-to-end pattern. The experiments show the advantages of our proposed method over PecepNet and DeepFilterNet.

eess.AS↗
arXiv2023

Carnegie Supernova Project-II: Near-infrared spectral diversity and template of Type Ia Supernovae

We present the largest and most homogeneous collection of near-infrared (NIR) spectra of Type Ia supernovae (SNe Ia): 339 spectra of 98 individual SNe obtained as part of the Carnegie Supernova Project-II. These spectra, obtained with the FIRE spectrograph on the 6.5 m Magellan Baade telescope, have a spectral range of 0.8--2.5 $μ$m. Using this sample, we explore the NIR spectral diversity of SNe Ia and construct a template of spectral time series as a function of the light-curve-shape parameter, color stretch $s_{BV}$. Principal component analysis is applied to characterize the diversity of the spectral features and reduce data dimensionality to a smaller subspace. Gaussian process regression is then used to model the subspace dependence on phase and light-curve shape and the associated uncertainty. Our template is able to predict spectral variations that are correlated with $s_{BV}$, such as the hallmark NIR features: Mg II at early times and the $H$-band break after peak. Using this template reduces the systematic uncertainties in K-corrections by ~90% compared to those from the Hsiao template. These uncertainties, defined as the mean K-correction differences computed with the color-matched template and observed spectra, are on the level of $4\times10^{-4}$ mag on average. This template can serve as the baseline spectral energy distribution for light-curve fitters and can identify peculiar spectral features that might point to compelling physics. The results presented here will substantially improve future SN~Ia cosmological experiments, for both nearby and distant samples.

astro-ph.HE↗
arXiv2023

Personalized speech enhancement combining band-split RNN and speaker attentive module

Target speaker information can be utilized in speech enhancement (SE) models to more effectively extract the desired speech. Previous works introduce the speaker embedding into speech enhancement models by means of concatenation or affine transformation. In this paper, we propose a speaker attentive module to calculate the attention scores between the speaker embedding and the intermediate features, which are used to rescale the features. By merging this module in the state-of-the-art SE model, we construct the personalized SE model for ICASSP Signal Processing Grand Challenge: DNS Challenge 5 (2023). Our system achieves a final score of 0.529 on the blind test set of track1 and 0.549 on track2.

eess.AS↗
arXiv2023

Near-infrared and Optical Nebular-phase Spectra of Type Ia Supernovae SN 2013aa and SN 2017cbv in NGC 5643

We present multi-wavelength time-series spectroscopy of SN 2013aa and SN 2017cbv, two Type Ia supernovae (SNe Ia) on the outskirts of the same host galaxy, NGC 5643. This work utilizes new nebular-phase near-infrared (NIR) spectra obtained by the Carnegie Supernova Project-II, in addition to previously published optical and NIR spectra. By measuring nebular-phase [Fe II] lines in both the optical and NIR, we examine the explosion kinematics and test the efficacy of several emission line fitting techniques commonly used in the literature. The NIR [Fe II] 1.644 $μ$m line provides the most robust velocity measurements against variations due to the choice of the fit method and line blending. The resulting effects on velocity measurements due to choosing different fit methods, initial fit parameters, continuum and line profile functions, and fit region boundaries were also investigated. The NIR [Fe II] velocities yield the same radial shift direction as velocities measured using the optical [Fe II] 7155 A line, but the sizes of the shifts are consistently and substantially lower, pointing to a potential issue in optical studies. The NIR [Fe II] 1.644 $μ$m emission profile shows a lack of significant asymmetry in both SNe Ia, and the observed low velocities elevate the importance for correcting for any radial velocity contribution from the host galaxy's rotation. The low [Fe II] velocities measured in the NIR at nebular phases disfavors most progenitor scenarios in close double-degenerate systems for both SN 2013aa and SN 2017cbv. The time evolution of the NIR [Fe II] 1.644 $μ$m line also indicates moderately high progenitor white dwarf central density and potentially high magnetic fields. These sibling SNe Ia were well observed at both early and late times, providing an excellent opportunity to study the intrinsic diversity of SNe Ia.

astro-ph.HE↗
arXiv2023

Attention does not guarantee best performance in speech enhancement

Attention mechanism has been widely utilized in speech enhancement (SE) because theoretically it can effectively model the long-term inherent connection of signal both in time domain and spectrum domain. However, the generally used global attention mechanism might not be the best choice since the adjacent information naturally imposes more influence than the far-apart information in speech enhancement. In this paper, we validate this conjecture by replacing attention with RNN in two typical state-of-the-art (SOTA) models, multi-scale temporal frequency convolutional network (MTFAA) with axial attention and conformer-based metric-GAN network (CMGAN).

cs.SD↗
arXiv2023

Local spectral attention for full-band speech enhancement

Attention mechanism has been widely utilized in speech enhancement (SE) because theoretically it can effectively model the inherent connection of signal both in time domain and spectrum domain. Usually, the span of attention is limited in time domain while the attention in frequency domain spans the whole frequency range. In this paper, we notice that the attention over the whole frequency range hampers the inference for full-band SE and possibly leads to excessive residual noise. To alleviate this problem, we introduce local spectral attention (LSA) into full-band SE model by limiting the span of attention. The ablation test on the state-of-the-art (SOTA) full-band SE model reveals that the local frequency attention can effectively improve overall performance. The improved model achieves the best objective score on the full-band VoiceBank+DEMAND set.

cs.SD↗
arXiv2023

A JWST Near- and Mid-Infrared Nebular Spectrum of the Type Ia Supernova 2021aefx

We present JWST near- and mid-infrared spectroscopic observations of the nearby normal Type Ia supernova SN 2021aefx in the nebular phase at $+255$ days past maximum light. Our Near Infrared Spectrograph (NIRSpec) and Mid Infrared Instrument (MIRI) observations, combined with ground-based optical data from the South African Large Telescope (SALT), constitute the first complete optical $+$ NIR $+$ MIR nebular SN Ia spectrum covering 0.3$-$14 $μ$m. This spectrum unveils the previously unobserved 2.5$-$5 $μ$m region, revealing strong nebular iron and stable nickel emission, indicative of high-density burning that can constrain the progenitor mass. The data show a significant improvement in sensitivity and resolution compared to previous Spitzer MIR data. We identify numerous NIR and MIR nebular emission lines from iron-group elements and as well as lines from the intermediate-mass element argon. The argon lines extend to higher velocities than the iron-group elements, suggesting stratified ejecta that are a hallmark of delayed-detonation or double-detonation SN Ia models. We present fits to simple geometric line profiles to features beyond 1.2 $μ$m and find that most lines are consistent with Gaussian or spherical emission distributions, while the [Ar III] 8.99 $μ$m line has a distinctively flat-topped profile indicating a thick spherical shell of emission. Using our line profile fits, we investigate the emissivity structure of SN 2021aefx and measure kinematic properties. Continued observations of SN 2021aefx and other SNe Ia with JWST will be transformative to the study of SN Ia composition, ionization structure, density, and temperature, and will provide important constraints on SN Ia progenitor and explosion models.

astro-ph.HE↗
arXiv2023

Serendipitous Nebular-phase JWST Imaging of SN Ia 2021aefx: Testing the Confinement of 56-Co Decay Energy

We present new 0.3-21 micron photometry of SN 2021aefx in the spiral galaxy NGC 1566 at +357 days after B-band maximum, including the first detection of any SN Ia at >15 micron. These observations follow earlier JWST observations of SN 2021aefx at +255 days after the time of maximum brightness, allowing us to probe the temporal evolution of the emission properties. We measure the fraction of flux emerging at different wavelengths and its temporal evolution. Additionally, the integrated 0.3-14 micron decay rate of $Δm_{0.3-14} = 1.35 \pm 0.05$ mag/100 days is higher than the decline rate from the radioactive decay of $^{56}$Co of $\sim 1.2$mag/100 days. The most plausible explanation for this discrepancy is that flux is shifting to >14 micron, and future JWST observations of SNe Ia will be able to directly test this hypothesis. However, models predicting non-radiative energy loss cannot be excluded with the present data.

astro-ph.GA↗
arXiv2022

Distributed Active Noise Control System Based on a Block Diffusion FxLMS Algorithm with Bidirectional Communication

Recently, distributed active noise control systems based on diffusion adaptation have attracted significant research interest due to their balance between computational complexity and stability compared to conventional centralized and decentralized adaptation schemes. However, the existing diffusion FxLMS algorithm employs node-specific adaptation and neighborhood-wide combination, and assumes that the control filters of neighbor nodes are similar to each other. This assumption is not true in practical applications, and it leads to inferior performance to the centralized controller approach. In contrast, this paper proposes a Block Diffusion FxLMS algorithm with bidirectional communication, which uses neighborhood-wide adaptation and node-specific combination to update the control filters. Simulation results validate that the proposed algorithm converges to the solution of the centralized controller with reduced computational burden.

eess.AS↗
arXiv2022

HYRR: Hybrid Infused Reranking for Passage Retrieval

We present Hybrid Infused Reranking for Passages Retrieval (HYRR), a framework for training rerankers based on a hybrid of BM25 and neural retrieval models. Retrievers based on hybrid models have been shown to outperform both BM25 and neural models alone. Our approach exploits this improved performance when training a reranker, leading to a robust reranking model. The reranker, a cross-attention neural model, is shown to be robust to different first-stage retrieval systems, achieving better performance than rerankers simply trained upon the first-stage retrievers in the multi-stage systems. We present evaluations on a supervised passage retrieval task using MS MARCO and zero-shot retrieval tasks using BEIR. The empirical results show strong performance on both evaluations.

cs.CL↗
arXiv2022

Distilling Object Detectors With Global Knowledge

Knowledge distillation learns a lightweight student model that mimics a cumbersome teacher. Existing methods regard the knowledge as the feature of each instance or their relations, which is the instance-level knowledge only from the teacher model, i.e., the local knowledge. However, the empirical studies show that the local knowledge is much noisy in object detection tasks, especially on the blurred, occluded, or small instances. Thus, a more intrinsic approach is to measure the representations of instances w.r.t. a group of common basis vectors in the two feature spaces of the teacher and the student detectors, i.e., global knowledge. Then, the distilling algorithm can be applied as space alignment. To this end, a novel prototype generation module (PGM) is proposed to find the common basis vectors, dubbed prototypes, in the two feature spaces. Then, a robust distilling module (RDM) is applied to construct the global knowledge based on the prototypes and filtrate noisy global and local knowledge by measuring the discrepancy of the representations in two feature spaces. Experiments with Faster-RCNN and RetinaNet on PASCAL and COCO datasets show that our method achieves the best performance for distilling object detectors with various backbones, which even surpasses the performance of the teacher model. We also show that the existing methods can be easily combined with global knowledge and obtain further improvement. Code is available: https://github.com/hikvision-research/DAVAR-Lab-ML.

cs.CV↗
arXiv2022

RankT5: Fine-Tuning T5 for Text Ranking with Ranking Losses

Recently, substantial progress has been made in text ranking based on pretrained language models such as BERT. However, there are limited studies on how to leverage more powerful sequence-to-sequence models such as T5. Existing attempts usually formulate text ranking as classification and rely on postprocessing to obtain a ranked list. In this paper, we propose RankT5 and study two T5-based ranking model structures, an encoder-decoder and an encoder-only one, so that they not only can directly output ranking scores for each query-document pair, but also can be fine-tuned with "pairwise" or "listwise" ranking losses to optimize ranking performances. Our experiments show that the proposed models with ranking losses can achieve substantial ranking performance gains on different public text ranking data sets. Moreover, when fine-tuned with listwise ranking losses, the ranking model appears to have better zero-shot ranking performance on out-of-domain data sets compared to the model fine-tuned with classification losses.

cs.IR↗
arXiv2022

Promptagator: Few-shot Dense Retrieval From 8 Examples

Much recent research on information retrieval has focused on how to transfer from one task (typically with abundant supervised data) to various other tasks where supervision is limited, with the implicit assumption that it is possible to generalize from one task to all the rest. However, this overlooks the fact that there are many diverse and unique retrieval tasks, each targeting different search intents, queries, and search domains. In this paper, we suggest to work on Few-shot Dense Retrieval, a setting where each task comes with a short description and a few examples. To amplify the power of a few examples, we propose Prompt-base Query Generation for Retriever (Promptagator), which leverages large language models (LLM) as a few-shot query generator, and creates task-specific retrievers based on the generated data. Powered by LLM's generalization ability, Promptagator makes it possible to create task-specific end-to-end retrievers solely based on a few examples {without} using Natural Questions or MS MARCO to train %question generators or dual encoders. Surprisingly, LLM prompting with no more than 8 examples allows dual encoders to outperform heavily engineered models trained on MS MARCO like ColBERT v2 by more than 1.2 nDCG on average on 11 retrieval sets. Further training standard-size re-rankers using the same generated data yields another 5.0 point nDCG improvement. Our studies determine that query generation can be far more effective than previously observed, especially when a small amount of task-specific knowledge is given.

cs.CL↗