arXiv subjects

Chen Zhang

Publications and source records attributed to Chen Zhang.

At least 469 records · Page 26Linked to original sources

arXiv2022

Analyzing and Evaluating Faithfulness in Dialogue Summarization

Dialogue summarization is abstractive in nature, making it suffer from factual errors. The factual correctness of summaries has the highest priority before practical applications. Many efforts have been made to improve faithfulness in text summarization. However, there is a lack of systematic study on dialogue summarization systems. In this work, we first perform the fine-grained human analysis on the faithfulness of dialogue summaries and observe that over 35% of generated summaries are faithfully inconsistent respective the source dialogues. Furthermore, we present a new model-level faithfulness evaluation method. It examines generation models with multi-choice questions created by rule-based transformations. Experimental results show that our evaluation schema is a strong proxy for the factual correctness of summarization models. The human-annotated faithfulness samples and the evaluation toolkit are released to facilitate future research toward faithful dialogue summarization.

cs.CL↗

arXiv2022

Sparse Teachers Can Be Dense with Knowledge

Recent advances in distilling pretrained language models have discovered that, besides the expressiveness of knowledge, the student-friendliness should be taken into consideration to realize a truly knowledgable teacher. Based on a pilot study, we find that over-parameterized teachers can produce expressive yet student-unfriendly knowledge and are thus limited in overall knowledgableness. To remove the parameters that result in student-unfriendliness, we propose a sparse teacher trick under the guidance of an overall knowledgable score for each teacher parameter. The knowledgable score is essentially an interpolation of the expressiveness and student-friendliness scores. The aim is to ensure that the expressive parameters are retained while the student-unfriendly ones are removed. Extensive experiments on the GLUE benchmark show that the proposed sparse teachers can be dense with knowledge and lead to students with compelling performance in comparison with a series of competitive baselines.

cs.CL↗

arXiv2022

SongDriver: Real-time Music Accompaniment Generation without Logical Latency nor Exposure Bias

Real-time music accompaniment generation has a wide range of applications in the music industry, such as music education and live performances. However, automatic real-time music accompaniment generation is still understudied and often faces a trade-off between logical latency and exposure bias. In this paper, we propose SongDriver, a real-time music accompaniment generation system without logical latency nor exposure bias. Specifically, SongDriver divides one accompaniment generation task into two phases: 1) The arrangement phase, where a Transformer model first arranges chords for input melodies in real-time, and caches the chords for the next phase instead of playing them out. 2) The prediction phase, where a CRF model generates playable multi-track accompaniments for the coming melodies based on previously cached chords. With this two-phase strategy, SongDriver directly generates the accompaniment for the upcoming melody, achieving zero logical latency. Furthermore, when predicting chords for a timestep, SongDriver refers to the cached chords from the first phase rather than its previous predictions, which avoids the exposure bias problem. Since the input length is often constrained under real-time conditions, another potential problem is the loss of long-term sequential information. To make up for this disadvantage, we extract four musical features from a long-term music piece before the current time step as global information. In the experiment, we train SongDriver on some open-source datasets and an original àiSong Dataset built from Chinese-style modern pop music scores. The results show that SongDriver outperforms existing SOTA (state-of-the-art) models on both objective and subjective metrics, meanwhile significantly reducing the physical latency.

cs.SD↗

arXiv2022

Making Pretrained Language Models Good Long-tailed Learners

Prompt-tuning has shown appealing performance in few-shot classification by virtue of its capability in effectively exploiting pre-trained knowledge. This motivates us to check the hypothesis that prompt-tuning is also a promising choice for long-tailed classification, since the tail classes are intuitively few-shot ones. To achieve this aim, we conduct empirical studies to examine the hypothesis. The results demonstrate that prompt-tuning makes pretrained language models at least good long-tailed learners. For intuitions on why prompt-tuning can achieve good performance in long-tailed classification, we carry out in-depth analyses by progressively bridging the gap between prompt-tuning and commonly used finetuning. The summary is that the classifier structure and parameterization form the key to making good long-tailed learners, in comparison with the less important input structure. Finally, we verify the applicability of our finding to few-shot classification. Good long-tailed learners can be abbreviated as Glee.

cs.CL↗

arXiv2022

XPrompt: Exploring the Extreme of Prompt Tuning

Prompt tuning learns soft prompts to condition frozen Pre-trained Language Models (PLMs) for performing downstream tasks in a parameter-efficient manner. While prompt tuning has gradually reached the performance level of fine-tuning as the model scale increases, there is still a large performance gap between prompt tuning and fine-tuning for models of moderate and small scales (typically less than 11B parameters). In this paper, we empirically show that the trained prompt tokens can have a negative impact on a downstream task and thus degrade its performance. To bridge the gap, we propose a novel Prompt tuning model with an eXtremely small scale (XPrompt) under the regime of lottery tickets hypothesis. Specifically, XPrompt eliminates the negative prompt tokens at different granularity levels through a hierarchical structured pruning, yielding a more parameter-efficient prompt yet with a competitive performance. Comprehensive experiments are carried out on SuperGLUE tasks, and the extensive results indicate that XPrompt is able to close the performance gap at smaller model scales.

cs.CL↗

arXiv2022

Does Thermal Really Always Matter for RGB-T Salient Object Detection?

In recent years, RGB-T salient object detection (SOD) has attracted continuous attention, which makes it possible to identify salient objects in environments such as low light by introducing thermal image. However, most of the existing RGB-T SOD models focus on how to perform cross-modality feature fusion, ignoring whether thermal image is really always matter in SOD task. Starting from the definition and nature of this task, this paper rethinks the connotation of thermal modality, and proposes a network named TNet to solve the RGB-T SOD task. In this paper, we introduce a global illumination estimation module to predict the global illuminance score of the image, so as to regulate the role played by the two modalities. In addition, considering the role of thermal modality, we set up different cross-modality interaction mechanisms in the encoding phase and the decoding phase. On the one hand, we introduce a semantic constraint provider to enrich the semantics of thermal images in the encoding phase, which makes thermal modality more suitable for the SOD task. On the other hand, we introduce a two-stage localization and complementation module in the decoding phase to transfer object localization cue and internal integrity cue in thermal features to the RGB modality. Extensive experiments on three datasets show that the proposed TNet achieves competitive performance compared with 20 state-of-the-art methods.

cs.CV↗

arXiv2022

CIR-Net: Cross-modality Interaction and Refinement for RGB-D Salient Object Detection

Focusing on the issue of how to effectively capture and utilize cross-modality information in RGB-D salient object detection (SOD) task, we present a convolutional neural network (CNN) model, named CIR-Net, based on the novel cross-modality interaction and refinement. For the cross-modality interaction, 1) a progressive attention guided integration unit is proposed to sufficiently integrate RGB-D feature representations in the encoder stage, and 2) a convergence aggregation structure is proposed, which flows the RGB and depth decoding features into the corresponding RGB-D decoding streams via an importance gated fusion unit in the decoder stage. For the cross-modality refinement, we insert a refinement middleware structure between the encoder and the decoder, in which the RGB, depth, and RGB-D encoder features are further refined by successively using a self-modality attention refinement unit and a cross-modality weighting refinement unit. At last, with the gradually refined features, we predict the saliency map in the decoder stage. Extensive experiments on six popular RGB-D SOD benchmarks demonstrate that our network outperforms the state-of-the-art saliency detectors both qualitatively and quantitatively.

cs.CV↗

arXiv2022

SoccerNet 2022 Challenges Results

The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.

cs.CV↗

arXiv2022

Data-mining of In-Situ TEM Experiments: on the Dynamics of Dislocations in CoCrFeMnNi Alloys

High entropy alloys are a class of materials with many significant improvements in terms of mechanical properties as compared to ``classical'' alloys. The corresponding structure-property relations are not yet entirely clear, but it is commonly believed that the good mechanical performance is strongly related to dislocation interactions with the complex energy landscape formed due to alloying. Although in-situ Transmission Electron Microscopy (TEM) allows high-resolution studies of the structure and dynamics of moving dislocations and makes the local obstacle/energy ``landscape'' directly visible in the geometry of dislocations; such observation, however, are merely qualitative, and detailed three-dimensional analyses of the interaction between dislocations and the energy landscape is still missing. In this work, we utilized dislocations as ``probes'' for the local energy maxima which play the role of pinning points for the dislocation movement. To this end, we developed a unique data-mining approach that can perform coarse-grained spatio-temporal analysis, making ensemble averaging of a considerable number of snapshots possible. We investigate the effect of pinning points on the dislocation gliding behavior of CoCrFeMnNi alloy during in-situ TEM straining and find that (i) the pinning point strength changes when dislocations glide through and (ii) the pinning point moves along the direction close to the Burgers vector direction. Our data-mining method can be applied to dislocation motion in general, making it a useful tool for dislocation research.

cond-mat.mtrl-sci↗

arXiv2022

X-ray performance of a customized large-format scientifc CMOS detector

In recent years, the performance of Scientifc Complementary Metal Oxide Semiconductor (sCMOS) sensors has been improved signifcantly. Compared with CCD sensors, sCMOS sensors have various advantages, making them potentially better devices for optical and X-ray detection, especially in time-domain astronomy. After a series of tests of sCMOS sensors, we proposed a new dedicated high-speed, large-format X-ray detector in 2016 cooperating with Gpixel Inc. This new sCMOS sensor has a physical size of 6 cm by 6 cm, with an array of 4096 by 4096 pixels and a pixel size of 15 um. The frame rate is 20.1 fps under current condition and can be boosted to a maximum value around 100 fps. The epitaxial thickness is increased to 10 um compared to the previous sCMOS product. We show the results of its frst taped-out product in this work. The dark current of this sCMOS is lower than 10 e/pixel/s at 20C, and lower than 0.02 e/pixel/s at -30C. The Fixed Pattern Noise (FPN) and the readout noise are lower than 5 e in high-gain situation and show a small increase at low temperature. The energy resolution reaches 180.1 eV (3.1%) at 5.90 keV for single-pixel events and 212.3 eV (3.6%) for all split events. The continuous X-ray spectrum measurement shows that this sensor is able to response to X-ray photons from 500 eV to 37 keV. The excellent performance, as demonstrated from these test results, makes sCMOS sensor an ideal detector for X-ray imaging and spectroscopic application.

astro-ph.IM↗

arXiv2022

Design and test results of scientific X-ray CMOS cameras

In recent years, scientific CMOS (sCMOS) sensors have found increasing applications to X-ray detection, including X-ray astronomical observations. In order to examine the performance of sCMOS sensors, we have developed X-ray cameras based on sCMOS sensors. Two cameras, CNX22 and CNX 66, have been developed using sCMOS sensors with a photosensitive area of 2 cm * 2 cm and 6 cm * 6 cm, respectively. The designs of the cameras are presented in this paper. The CNX22 camera has a frame rate of 48 fps, whereas CNX66 has a frame rate of currently 20 fps, that can be boosted to 100 fps in the future. The operating temperature of the sCMOS sensor can reach to -20C for CNX22 and -30C for CNX66 with a peltier cooler device. In addition to the commonly used mode of saving original images, the cameras provide a mode of real-time extraction of X-ray events and storage their information, which significantly reduces the requirement for data storage and offline analysis work. For both cameras, the energy resolutions can reach less than 200 eV at 5.9 keV using single-pixel events. These cameras are suitable for X-ray spectroscopy applications in laboratories and calibration for the space X-ray telescopes.

astro-ph.IM↗

arXiv2022

Asset Pricing and Deep Learning

Traditional machine learning methods have been widely studied in financial innovation. My study focuses on the application of deep learning methods on asset pricing. I investigate various deep learning methods for asset pricing, especially for risk premia measurement. All models take the same set of predictive signals (firm characteristics, systematic risks and macroeconomics). I demonstrate high performance of all kinds of state-of-the-art (SOTA) deep learning methods, and figure out that RNNs with memory mechanism and attention have the best performance in terms of predictivity. Furthermore, I demonstrate large economic gains to investors using deep learning forecasts. The results of my comparative experiments highlight the importance of domain knowledge and financial theory when designing deep learning models. I also show return prediction tasks bring new challenges to deep learning. The time varying distribution causes distribution shift problem, which is essential for financial time series prediction. I demonstrate that deep learning methods can improve asset risk premium measurement. Due to the booming deep learning studies, they can constantly promote the study of underlying financial mechanisms behind asset pricing. I also propose a promising research method that learning from data and figuring out the underlying economic mechanisms through explainable artificial intelligence (AI) methods. My findings not only justify the value of deep learning in blooming fintech development, but also highlight their prospects and advantages over traditional machine learning methods.

q-fin.ST↗

arXiv2022

Nesting Forward Automatic Differentiation for Memory-Efficient Deep Neural Network Training

An activation function is an element-wise mathematical function and plays a crucial role in deep neural networks (DNN). Many novel and sophisticated activation functions have been proposed to improve the DNN accuracy but also consume massive memory in the training process with back-propagation. In this study, we propose the nested forward automatic differentiation (Forward-AD), specifically for the element-wise activation function for memory-efficient DNN training. We deploy nested Forward-AD in two widely-used deep learning frameworks, TensorFlow and PyTorch, which support the static and dynamic computation graph, respectively. Our evaluation shows that nested Forward-AD reduces the memory footprint by up to 1.97x than the baseline model and outperforms the recomputation by 20% under the same memory reduction ratio.

cs.LG↗

arXiv2022

The Einstein Probe Mission

The Einstein Probe (EP) is a mission designed to monitor the sky in the soft X-ray band. It will perform systematic surveys and characterisation of high-energy transients and monitoring of variable objects at unprecedented sensitivity and monitoring cadences. It has a large instantaneous field-of-view (3,600 sq. deg.), that is realised via the lobster-eye micro-pore X-ray focusing optics. EP also carries a conventional X-ray focusing telescope with a larger effective area to perform followup observations and precise positioning of newly-discovered transients. Alerts for transient objects will be issued publicly and timely. The scientific goals of EP are concerned with discovering faint, distant or rare types of high-energy transients and variable sources. Within the confines of a modest-sized mission, EP will cover a wide range of scientific topics, from the nearby to high-redshift Universe. The Einstein Probe is a mission of the Chinese Academy of Sciences, and also an international collaborative project. This paper presents the background, scientific objectives, and the mission design including the micro-pore optics and CMOS technologies adopted, the instruments and their expected performance, and the mission profile. The development status of the project is also presented.

astro-ph.HE↗

arXiv2022

Doge Tickets: Uncovering Domain-general Language Models by Playing Lottery Tickets

Over-parameterized models, typically pretrained language models (LMs), have shown an appealing expressive power due to their small learning bias. However, the huge learning capacity of LMs can also lead to large learning variance. In a pilot study, we find that, when faced with multiple domains, a critical portion of parameters behave unexpectedly in a domain-specific manner while others behave in a domain-general one. Motivated by this phenomenon, we for the first time posit that domain-general parameters can underpin a domain-general LM that can be derived from the original LM. To uncover the domain-general LM, we propose to identify domain-general parameters by playing lottery tickets (dubbed doge tickets). In order to intervene the lottery, we propose a domain-general score, which depicts how domain-invariant a parameter is by associating it with the variance. Comprehensive experiments are conducted on the Amazon, Mnli and OntoNotes datasets. The results show that the doge tickets obtains an improved out-of-domain generalization in comparison with a range of competitive baselines. Analysis results further hint the existence of domain-general parameters and the performance consistency of doge tickets.

cs.CL↗

arXiv2022

A Weakly Supervised Learning Framework for Salient Object Detection via Hybrid Labels

Fully-supervised salient object detection (SOD) methods have made great progress, but such methods often rely on a large number of pixel-level annotations, which are time-consuming and labour-intensive. In this paper, we focus on a new weakly-supervised SOD task under hybrid labels, where the supervision labels include a large number of coarse labels generated by the traditional unsupervised method and a small number of real labels. To address the issues of label noise and quantity imbalance in this task, we design a new pipeline framework with three sophisticated training strategies. In terms of model framework, we decouple the task into label refinement sub-task and salient object detection sub-task, which cooperate with each other and train alternately. Specifically, the R-Net is designed as a two-stream encoder-decoder model equipped with Blender with Guidance and Aggregation Mechanisms (BGA), aiming to rectify the coarse labels for more reliable pseudo-labels, while the S-Net is a replaceable SOD network supervised by the pseudo labels generated by the current R-Net. Note that, we only need to use the trained S-Net for testing. Moreover, in order to guarantee the effectiveness and efficiency of network training, we design three training strategies, including alternate iteration mechanism, group-wise incremental mechanism, and credibility verification mechanism. Experiments on five SOD benchmarks show that our method achieves competitive performance against weakly-supervised/unsupervised methods both qualitatively and quantitatively.

cs.CV↗

arXiv2022

Knowledge-enhanced Iterative Instruction Generation and Reasoning for Knowledge Base Question Answering

Multi-hop Knowledge Base Question Answering(KBQA) aims to find the answer entity in a knowledge base which is several hops from the topic entity mentioned in the question. Existing Retrieval-based approaches first generate instructions from the question and then use them to guide the multi-hop reasoning on the knowledge graph. As the instructions are fixed during the whole reasoning procedure and the knowledge graph is not considered in instruction generation, the model cannot revise its mistake once it predicts an intermediate entity incorrectly. To handle this, we propose KBIGER(Knowledge Base Iterative Instruction GEnerating and Reasoning), a novel and efficient approach to generate the instructions dynamically with the help of reasoning graph. Instead of generating all the instructions before reasoning, we take the (k-1)-th reasoning graph into consideration to build the k-th instruction. In this way, the model could check the prediction from the graph and generate new instructions to revise the incorrect prediction of intermediate entities. We do experiments on two multi-hop KBQA benchmarks and outperform the existing approaches, becoming the new-state-of-the-art. Further experiments show our method does detect the incorrect prediction of intermediate entities and has the ability to revise such errors.

cs.CL↗

arXiv2022

Multi-scale temporal-frequency attention for music source separation

In recent years, deep neural networks (DNNs) based approaches have achieved the start-of-the-art performance for music source separation (MSS). Although previous methods have addressed the large receptive field modeling using various methods, the temporal and frequency correlations of the music spectrogram with repeated patterns have not been explicitly explored for the MSS task. In this paper, a temporal-frequency attention module is proposed to model the spectrogram correlations along both temporal and frequency dimensions. Moreover, a multi-scale attention is proposed to effectively capture the correlations for music signal. The experimental results on MUSDB18 dataset show that the proposed method outperforms the existing state-of-the-art systems with 9.51 dB signal-to-distortion ratio (SDR) on separating the vocal stems, which is the primary practical application of MSS.

eess.AS↗