Searcharxiv⌕ Search

arXiv subjects

Hao Chen

Publications and source records attributed to Hao Chen.

At least 1,027 records · Page 57Linked to original sources
arXiv2023

Understanding Programs by Exploiting (Fuzzing) Test Cases

Semantic understanding of programs has attracted great attention in the community. Inspired by recent successes of large language models (LLMs) in natural language understanding, tremendous progress has been made by treating programming language as another sort of natural language and training LLMs on corpora of program code. However, programs are essentially different from texts after all, in a sense that they are normally heavily structured and syntax-strict. In particular, programs and their basic units (i.e., functions and subroutines) are designed to demonstrate a variety of behaviors and/or provide possible outputs, given different inputs. The relationship between inputs and possible outputs/behaviors represents the functions/subroutines and profiles the program as a whole. Therefore, we propose to incorporate such a relationship into learning, for achieving a deeper semantic understanding of programs. To obtain inputs that are representative enough to trigger the execution of most part of the code, we resort to fuzz testing and propose fuzz tuning to boost the performance of program understanding and code representation learning, given a pre-trained LLM. The effectiveness of the proposed method is verified on two program understanding tasks including code clone detection and code classification, and it outperforms current state-of-the-arts by large margins. Code is available at https://github.com/rabbitjy/FuzzTuning.

cs.LG↗
arXiv2023

A Dynamic Feature Interaction Framework for Multi-task Visual Perception

Multi-task visual perception has a wide range of applications in scene understanding such as autonomous driving. In this work, we devise an efficient unified framework to solve multiple common perception tasks, including instance segmentation, semantic segmentation, monocular 3D detection, and depth estimation. Simply sharing the same visual feature representations for these tasks impairs the performance of tasks, while independent task-specific feature extractors lead to parameter redundancy and latency. Thus, we design two feature-merge branches to learn feature basis, which can be useful to, and thus shared by, multiple perception tasks. Then, each task takes the corresponding feature basis as the input of the prediction task head to fulfill a specific task. In particular, one feature merge branch is designed for instance-level recognition the other for dense predictions. To enhance inter-branch communication, the instance branch passes pixel-wise spatial information of each instance to the dense branch using efficient dynamic convolution weighting. Moreover, a simple but effective dynamic routing mechanism is proposed to isolate task-specific features and leverage common properties among tasks. Our proposed framework, termed D2BNet, demonstrates a unique approach to parameter-efficient predictions for multi-task perception. In addition, as tasks benefit from co-training with each other, our solution achieves on par results on partially labeled settings on nuScenes and outperforms previous works for 3D detection and depth estimation on the Cityscapes dataset with full supervision.

cs.CV↗
arXiv2023

Dense gas and star formation in the Outer Milky Way

We present maps and spectra of the HCN(1-0) and HCO$^+$(1-0) lines in the extreme outer Galaxy, at galactocentric radii between 14 and 22 kpc, with the 13.7 meter Delingha telescope. The 9 molecular clouds were selected from a CO/$^{13}$CO survey of the outer quadrants. The goal is to better understand the structure of molecular clouds in these poorly studied subsolar metallicity regions and the relation with star formation. The lines are all narrow, less than 2km/s at half power, enabling detection of the HCN hyperfine structure in the stronger sources and allowing us to observationally test hyperfine collision rates. The hyperfine line ratios show that the HCN emission is optically thin with column densities estimated at N(HCN)~$3x10^{12}$\scm. The HCO$^+$ emission is approximately twice as strong as the HCN (taken as the sum of all components), in contrast with the inner Galaxy and nearby galaxies where they are similarly strong. For an abundance ratio $χ_{HCN}/χ_{HCO^+} = 3$, this requires a relatively low density solution for the dense gas, with n(H2) $\sim 10^3 - 10^4$\ccm. The $^{12}$CO/$^{13}$CO line ratios are similar to solar neighborhood values, roughly 7.5, despite the low $^{13}$CO abundance expected at such large radii. The HCO$^+$/CO and HCO$^+$/$^{13}$CO integrated intensity ratios are also standard at about 1/35 and 1/5 respectively. HCN is weak compared to the CO emission, with HCN/CO $\sim 1/70$ even after summing all hyperfine components. At the parsec scales observed here, the correlation between star formation, as traced by 24~$μ$m emission as is standard in extragalactic work, and dense gas via the HCN or HCO$^+$ emission, is poor, perhaps due to the lack of dynamic range. We find that the lowest dense gas fractions are in the sources at high galactic latitude (b>2, h>300pc above the plane), possibly due to lower pressure.

astro-ph.GA↗
arXiv2023

Lattice-Aided Extraction of Spread-Spectrum Hidden Data

This paper discusses the problem of extracting spread spectrum hidden data from the perspective of lattice decoding. Since the conventional blind extraction scheme multi-carrier iterative generalize least-squares (M-IGLS) and non-blind extraction scheme minimum mean square error (MMSE) suffer from performance degradation when the carriers lack sufficient orthogonality, we present two novel schemes from the viewpoint of lattice decoding, namely multi-carrier iterative successive interference cancellation (M-ISIC) and sphere decoding (SD). The better performance of M-ISIC and SD are confirmed by both theoretical justification and numerical simulations.

cs.CR↗
arXiv2023

Few-Shot Speaker Identification Using Lightweight Prototypical Network with Feature Grouping and Interaction

Existing methods for few-shot speaker identification (FSSI) obtain high accuracy, but their computational complexities and model sizes need to be reduced for lightweight applications. In this work, we propose a FSSI method using a lightweight prototypical network with the final goal to implement the FSSI on intelligent terminals with limited resources, such as smart watches and smart speakers. In the proposed prototypical network, an embedding module is designed to perform feature grouping for reducing the memory requirement and computational complexity, and feature interaction for enhancing the representational ability of the learned speaker embedding. In the proposed embedding module, audio feature of each speech sample is split into several low-dimensional feature subsets that are transformed by a recurrent convolutional block in parallel. Then, the operations of averaging, addition, concatenation, element-wise summation and statistics pooling are sequentially executed to learn a speaker embedding for each speech sample. The recurrent convolutional block consists of a block of bidirectional long short-term memory, and a block of de-redundancy convolution in which feature grouping and interaction are conducted too. Our method is compared to baseline methods on three datasets that are selected from three public speech corpora (VoxCeleb1, VoxCeleb2, and LibriSpeech). The results show that our method obtains higher accuracy under several conditions, and has advantages over all baseline methods in computational complexity and model size.

eess.AS↗
arXiv2023

Scale-aware Super-resolution Network with Dual Affinity Learning for Lesion Segmentation from Medical Images

Convolutional Neural Networks (CNNs) have shown remarkable progress in medical image segmentation. However, lesion segmentation remains a challenge to state-of-the-art CNN-based algorithms due to the variance in scales and shapes. On the one hand, tiny lesions are hard to be delineated precisely from the medical images which are often of low resolutions. On the other hand, segmenting large-size lesions requires large receptive fields, which exacerbates the first challenge. In this paper, we present a scale-aware super-resolution network to adaptively segment lesions of various sizes from the low-resolution medical images. Our proposed network contains dual branches to simultaneously conduct lesion mask super-resolution and lesion image super-resolution. The image super-resolution branch will provide more detailed features for the segmentation branch, i.e., the mask super-resolution branch, for fine-grained segmentation. Meanwhile, we introduce scale-aware dilated convolution blocks into the multi-task decoders to adaptively adjust the receptive fields of the convolutional kernels according to the lesion sizes. To guide the segmentation branch to learn from richer high-resolution features, we propose a feature affinity module and a scale affinity module to enhance the multi-task learning of the dual branches. On multiple challenging lesion segmentation datasets, our proposed network achieved consistent improvements compared to other state-of-the-art methods.

eess.IV↗
arXiv2023

Lattice Codes for Lattice-Based PKE

Existing error correction mechanisms in lattice-based public key encryption (PKE) rely on either naive modulation or its concatenation with error correction codes (ECC). This paper shows that lattice coding, as a joint ECC and modulation technique, can substitute the naive modulation in existing lattice-based PKEs to enjoy better correction performance. We begin by modeling the FrodoPKE protocol as a noisy point-to-point communication system, where the communication channel is similar to the additive white Gaussian noise (AWGN) channel. To employ lattice codes for this special channel that hinges on hypercube shaping, we propose an efficient labeling function that converts between binary information bits and lattice codewords. The parameter sets of FrodoPKE are improved towards either higher security levels or smaller ciphertext sizes. For example, the proposed Frodo-1344-E$_\text{8}$ has a 10-bit classical security gain over Frodo-1344.

cs.IT↗
arXiv2023

Zero-shot Multi-level Feature Transmission Policy Powered by Semantic Knowledge Base

Remote zero-shot object recognition, i.e., offloading zero-shot object recognition task from one mobile device to remote mobile edge computing (MEC) server or another mobile device, has become a common and important task to solve for 6G. In order to tackle this problem, this paper first establishes a zero-shot multi-level feature extractor, which projects the image into visual, semantic, as well as intermediate feature space in a lightweight way. Then, this paper proposes a novel multi-level feature transmission framework powered by semantic knowledge base (SKB), and characterizes the semantic loss and required transmission latency at each level. Under this setup, this paper formulates the multi-level feature transmission optimization problem to minimize the semantic loss under the end-to-end latency constraint. The optimization problem, however, is a multi-choice knapsack problem, and thus very difficult to be optimized. To resolve this issue, this paper proposes an efficient algorithm based on convex concave procedure to find a high-quality solution. Numerical results show that the proposed design outperforms the benchmarks, and illustrate the tradeoff between the transmission latency and zero-shot classification accuracy, as well as the effects of the SKBs at both the transmitter and receiver on classification accuracy.

cs.IT↗
arXiv2023

Multi-channel Integrated Recommendation with Exposure Constraints

Integrated recommendation, which aims at jointly recommending heterogeneous items from different channels in a main feed, has been widely applied to various online platforms. Though attractive, integrated recommendation requires the ranking methods to migrate from conventional user-item models to the new user-channel-item paradigm in order to better capture users' preferences on both item and channel levels. Moreover, practical feed recommendation systems usually impose exposure constraints on different channels to ensure user experience. This leads to greater difficulty in the joint ranking of heterogeneous items. In this paper, we investigate the integrated recommendation task with exposure constraints in practical recommender systems. Our contribution is forth-fold. First, we formulate this task as a binary online linear programming problem and propose a two-layer framework named Multi-channel Integrated Recommendation with Exposure Constraints (MIREC) to obtain the optimal solution. Second, we propose an efficient online allocation algorithm to determine the optimal exposure assignment of different channels from a global view of all user requests over the entire time horizon. We prove that this algorithm reaches the optimal point under a regret bound of $ \mathcal{O}(\sqrt{T}) $ with linear complexity. Third, we propose a series of collaborative models to determine the optimal layout of heterogeneous items at each user request. The joint modeling of user interests, cross-channel correlation, and page context in our models aligns more with the browsing nature of feed products than existing models. Finally, we conduct extensive experiments on both offline datasets and online A/B tests to verify the effectiveness of MIREC. The proposed framework has now been implemented on the homepage of Taobao to serve the main traffic.

cs.IR↗
arXiv2023

Multi-factor Sequential Re-ranking with Perception-Aware Diversification

Feed recommendation systems, which recommend a sequence of items for users to browse and interact with, have gained significant popularity in practical applications. In feed products, users tend to browse a large number of items in succession, so the previously viewed items have a significant impact on users' behavior towards the following items. Therefore, traditional methods that mainly focus on improving the accuracy of recommended items are suboptimal for feed recommendations because they may recommend highly similar items. For feed recommendation, it is crucial to consider both the accuracy and diversity of the recommended item sequences in order to satisfy users' evolving interest when consecutively viewing items. To this end, this work proposes a general re-ranking framework named Multi-factor Sequential Re-ranking with Perception-Aware Diversification (MPAD) to jointly optimize accuracy and diversity for feed recommendation in a sequential manner. Specifically, MPAD first extracts users' different scales of interests from their behavior sequences through graph clustering-based aggregations. Then, MPAD proposes two sub-models to respectively evaluate the accuracy and diversity of a given item by capturing users' evolving interest due to the ever-changing context and users' personal perception of diversity from an item sequence perspective. This is consistent with the browsing nature of the feed scenario. Finally, MPAD generates the return list by sequentially selecting optimal items from the candidate set to maximize the joint benefits of accuracy and diversity of the entire list. MPAD has been implemented in Taobao's homepage feed to serve the main traffic and provide services to recommend billions of items to hundreds of millions of users every day.

cs.IR↗
arXiv2023

Ultrawideband solid-state terahertz phase shifter electrically modulated by tunable conductive interface in total internal reflection geometry

Phase modulation plays a crucial role in various terahertz applications, including biomedical imaging, high-rate communication, and radar detection. Existing terahertz phase shifters typically rely on tuning the resonant effect of metamaterial structures to achieve a narrow bandwidth phase shift. However, the terahertz band offers a wide bandwidth resource, which has great advantages in high longitudinal resolution detection, high-capacity communication, spectral imaging and so on. Here, we propose and demonstrate an ultrawideband terahertz phase shifting mechanism that utilizes an optical conductivity tuneable interface combined with a non-resonant metasurface operating in the total internal reflection geometry. This approach effectively modulates the phase of the reflected terahertz signal in an ultrawideband. To implement this mechanism, we designed a structure consisting of graphene-loaded non-resonant periodic metal microslits arranged in the total internal reflection geometry. By controlling the gate voltage of the graphene within a range of 5 V, an averaged ~120° continuous phase shift in the frequency range of 0.4 to 1.2 THz was achieved. Notably, in the frequency range of 1 to 1.2 THz, the phase modulation exhibited a linear relationship with the driving voltage. Our device demonstrated minimal fluctuations in the reflected amplitude, with a deviation of less than 1 dB and an insertion loss of less than 10 dB. Additionally, the modulation speed of this solid-state device reached the kHz level. Remarkably, the phase modulation bandwidth (Δf/f) achieved approximately 100% of the arithmetic centre frequency at 0.8 THz, surpassing the definition of ultrawideband, which typically encompasses 20% of the centre frequency. To the best of our knowledge, this is the first and most wideband phase shifter developed for the terahertz regime to date.

physics.optics↗
arXiv2023

Maybe Only 0.5% Data is Needed: A Preliminary Exploration of Low Training Data Instruction Tuning

Instruction tuning for large language models (LLMs) has gained attention from researchers due to its ability to unlock the potential of LLMs in following instructions. While instruction tuning offers advantages for facilitating the adaptation of large language models (LLMs) to downstream tasks as a fine-tuning approach, training models with tens of millions or even billions of parameters on large amounts of data results in unaffordable computational costs. To address this, we focus on reducing the data used in LLM instruction tuning to decrease training costs and improve data efficiency, dubbed as Low Training Data Instruction Tuning (LTD Instruction Tuning). Specifically, this paper conducts a preliminary exploration into reducing the data used in LLM training and identifies several observations regarding task specialization for LLM training, such as the optimization of performance for a specific task, the number of instruction types required for instruction tuning, and the amount of data required for task-specific models. The results suggest that task-specific models can be trained using less than 0.5% of the original dataset, with a 2% improvement in performance over those trained on full task-related data.

cs.AI↗
arXiv2023

PromptRank: Unsupervised Keyphrase Extraction Using Prompt

The keyphrase extraction task refers to the automatic selection of phrases from a given document to summarize its core content. State-of-the-art (SOTA) performance has recently been achieved by embedding-based algorithms, which rank candidates according to how similar their embeddings are to document embeddings. However, such solutions either struggle with the document and candidate length discrepancies or fail to fully utilize the pre-trained language model (PLM) without further fine-tuning. To this end, in this paper, we propose a simple yet effective unsupervised approach, PromptRank, based on the PLM with an encoder-decoder architecture. Specifically, PromptRank feeds the document into the encoder and calculates the probability of generating the candidate with a designed prompt by the decoder. We extensively evaluate the proposed PromptRank on six widely used benchmarks. PromptRank outperforms the SOTA approach MDERank, improving the F1 score relatively by 34.18%, 24.87%, and 17.57% for 5, 10, and 15 returned results, respectively. This demonstrates the great potential of using prompt for unsupervised keyphrase extraction. We release our code at https://github.com/HLT-NLP/PromptRank.

cs.IR↗
arXiv2023

Semantic Framework based Query Generation for Temporal Question Answering over Knowledge Graphs

Answering factual questions with temporal intent over knowledge graphs (temporal KGQA) attracts rising attention in recent years. In the generation of temporal queries, existing KGQA methods ignore the fact that some intrinsic connections between events can make them temporally related, which may limit their capability. We systematically analyze the possible interpretation of temporal constraints and conclude the interpretation structures as the Semantic Framework of Temporal Constraints, SF-TCons. Based on the semantic framework, we propose a temporal question answering method, SF-TQA, which generates query graphs by exploring the relevant facts of mentioned entities, where the exploring process is restricted by SF-TCons. Our evaluations show that SF-TQA significantly outperforms existing methods on two benchmarks over different knowledge graphs.

cs.CL↗
arXiv2023

On the nature of X(3960)

A near-threshold enhancement in the $D_s^+ D_s^-$ system, dubbed as $X(3960)$, is observed by the LHCb collaboration recently. A combined analysis on $χ_{c0}(3930)~(\to D^+ D^-)$, $X(3960)~(\to D_s^+ D_s^-)$, and $X(3915)~(\to J/ψω)$ is performed using both a $K$-matrix approach of $D_{(s)}\bar{D}_{(s)}$ four-point contact interactions and a model of Flatté-like parameterizations. The use of the pole counting rule and spectral density function sum rule indicate, under current statistics, that this $D_s^+ D_s^-$ near-threshold state has probably the mixed nature of a $c\bar{c}$ confining state and $D_s^+ D_s^-$ continuum.

hep-ph↗
arXiv2023

Reference-based OCT Angiogram Super-resolution with Learnable Texture Generation

Optical coherence tomography angiography (OCTA) is a new imaging modality to visualize retinal microvasculature and has been readily adopted in clinics. High-resolution OCT angiograms are important to qualitatively and quantitatively identify potential biomarkers for different retinal diseases accurately. However, one significant problem of OCTA is the inevitable decrease in resolution when increasing the field-of-view given a fixed acquisition time. To address this issue, we propose a novel reference-based super-resolution (RefSR) framework to preserve the resolution of the OCT angiograms while increasing the scanning area. Specifically, textures from the normal RefSR pipeline are used to train a learnable texture generator (LTG), which is designed to generate textures according to the input. The key difference between the proposed method and traditional RefSR models is that the textures used during inference are generated by the LTG instead of being searched from a single reference image. Since the LTG is optimized throughout the whole training process, the available texture space is significantly enlarged and no longer limited to a single reference image, but extends to all textures contained in the training samples. Moreover, our proposed LTGNet does not require a reference image at the inference phase, thereby becoming invulnerable to the selection of the reference image. Both experimental and visual results show that LTGNet has superior performance and robustness over state-of-the-art methods, indicating good reliability and promise in real-life deployment. The source code will be made available upon acceptance.

eess.IV↗
arXiv2023

Understanding the Tricks of Deep Learning in Medical Image Segmentation: Challenges and Future Directions

Over the past few years, the rapid development of deep learning technologies for computer vision has significantly improved the performance of medical image segmentation (MedISeg). However, the diverse implementation strategies of various models have led to an extremely complex MedISeg system, resulting in a potential problem of unfair result comparisons. In this paper, we collect a series of MedISeg tricks for different model implementation phases (i.e., pre-training model, data pre-processing, data augmentation, model implementation, model inference, and result post-processing), and experimentally explore the effectiveness of these tricks on consistent baselines. With the extensive experimental results on both the representative 2D and 3D medical image datasets, we explicitly clarify the effect of these tricks. Moreover, based on the surveyed tricks, we also open-sourced a strong MedISeg repository, where each component has the advantage of plug-and-play. We believe that this milestone work not only completes a comprehensive and complementary survey of the state-of-the-art MedISeg approaches, but also offers a practical guide for addressing the future medical image processing challenges including but not limited to small dataset, class imbalance learning, multi-modality learning, and domain adaptation. The code and training weights have been released at: https://github.com/hust-linyi/seg_trick.

cs.CV↗
arXiv2023

Multi-Content Interaction Network for Few-Shot Segmentation

Few-Shot Segmentation (FSS) is challenging for limited support images and large intra-class appearance discrepancies. Most existing approaches focus on extracting high-level representations of the same layers for support-query correlations, neglecting the shift issue between different layers and scales, due to the huge difference between support and query samples. In this paper, we propose a Multi-Content Interaction Network (MCINet) to remedy this issue by fully exploiting and interacting with the multi-scale contextual information contained in the support-query pairs to supplement the same-layer correlations. Specifically, MCINet improves FSS from the perspectives of boosting the query representations by incorporating the low-level structural information from another query branch into the high-level semantic features, enhancing the support-query correlations by exploiting both the same-layer and adjacent-layer features, and refining the predicted results by a multi-scale mask prediction strategy, with which the different scale contents have bidirectionally interacted. Experiments on two benchmarks demonstrate that our approach reaches SOTA performances and outperforms the best competitors with many desirable advantages, especially on the challenging COCO dataset.

cs.CV↗