arXiv subjects

Gang Wu

Publications and source records attributed to Gang Wu.

At least 91 records · Page 5Linked to original sources

arXiv2024

FLIQS: One-Shot Mixed-Precision Floating-Point and Integer Quantization Search

Quantization has become a mainstream compression technique for reducing model size, computational requirements, and energy consumption for modern deep neural networks (DNNs). With improved numerical support in recent hardware, including multiple variants of integer and floating point, mixed-precision quantization has become necessary to achieve high-quality results with low model cost. Prior mixed-precision methods have performed either a post-training quantization search, which compromises on accuracy, or a differentiable quantization search, which leads to high memory usage from branching. Therefore, we propose the first one-shot mixed-precision quantization search that eliminates the need for retraining in both integer and low-precision floating point models. We evaluate our search (FLIQS) on multiple convolutional and vision transformer networks to discover Pareto-optimal models. Our approach improves upon uniform precision, manual mixed-precision, and recent integer quantization search methods. With integer models, we increase the accuracy of ResNet-18 on ImageNet by 1.31% and ResNet-50 by 0.90% with equivalent model cost over previous methods. Additionally, for the first time, we explore a novel mixed-precision floating-point search and improve MobileNetV2 by up to 0.98% compared to prior state-of-the-art FP8 models. Finally, we extend FLIQS to simultaneously search a joint quantization and neural architecture space and improve the ImageNet accuracy by 2.69% with similar model cost on a MobileNetV2 search space.

cs.CV↗

arXiv2024

Hybrid Bit and Semantic Communications

Semantic communication technology is regarded as a method surpassing the Shannon limit of bit transmission, capable of effectively enhancing transmission efficiency. However, current approaches that directly map content to transmission symbols are challenging to deploy in practice, imposing significant limitations on the development of semantic communication. To address this challenge, we propose a hybrid bit and semantic communication system, named HybridBSC, in which encoded semantic information is inserted into bit information for transmission via conventional digital communication systems utilizing same spectrum resources. The system can be easily deployed using existing communication architecture to achieve bit and semantic information transmission. Particularly, we design a semantic insertion and extraction scheme to implement this strategy. Furthermore, we conduct experimental validation based on the pluto-based software defined radio (SDR) platform in a real wireless channel, demonstrating that the proposed strategy can simultaneously transmit semantic and bit information.

eess.SP↗

arXiv2024

The Third Monocular Depth Estimation Challenge

This paper discusses the results of the third edition of the Monocular Depth Estimation Challenge (MDEC). The challenge focuses on zero-shot generalization to the challenging SYNS-Patches dataset, featuring complex scenes in natural and indoor settings. As with the previous edition, methods can use any form of supervision, i.e. supervised or self-supervised. The challenge received a total of 19 submissions outperforming the baseline on the test set: 10 among them submitted a report describing their approach, highlighting a diffused use of foundational models such as Depth Anything at the core of their method. The challenge winners drastically improved 3D F-Score performance, from 17.51% to 23.72%.

cs.CV↗

arXiv2024

Exploiting Self-Supervised Constraints in Image Super-Resolution

Recent advances in self-supervised learning, predominantly studied in high-level visual tasks, have been explored in low-level image processing. This paper introduces a novel self-supervised constraint for single image super-resolution, termed SSC-SR. SSC-SR uniquely addresses the divergence in image complexity by employing a dual asymmetric paradigm and a target model updated via exponential moving average to enhance stability. The proposed SSC-SR framework works as a plug-and-play paradigm and can be easily applied to existing SR models. Empirical evaluations reveal that our SSC-SR framework delivers substantial enhancements on a variety of benchmark datasets, achieving an average increase of 0.1 dB over EDSR and 0.06 dB over SwinIR. In addition, extensive ablation studies corroborate the effectiveness of each constituent in our SSC-SR framework. Codes are available at https://github.com/Aitical/SSCSR.

cs.CV↗

arXiv2024

Fully $1\times1$ Convolutional Network for Lightweight Image Super-Resolution

Deep models have achieved significant process on single image super-resolution (SISR) tasks, in particular large models with large kernel ($3\times3$ or more). However, the heavy computational footprint of such models prevents their deployment in real-time, resource-constrained environments. Conversely, $1\times1$ convolutions bring substantial computational efficiency, but struggle with aggregating local spatial representations, an essential capability to SISR models. In response to this dichotomy, we propose to harmonize the merits of both $3\times3$ and $1\times1$ kernels, and exploit a great potential for lightweight SISR tasks. Specifically, we propose a simple yet effective fully $1\times1$ convolutional network, named Shift-Conv-based Network (SCNet). By incorporating a parameter-free spatial-shift operation, it equips the fully $1\times1$ convolutional network with powerful representation capability while impressive computational efficiency. Extensive experiments demonstrate that SCNets, despite its fully $1\times1$ convolutional structure, consistently matches or even surpasses the performance of existing lightweight SR models that employ regular convolutions. The code and pre-trained models can be found at https://github.com/Aitical/SCNet.

cs.CV↗

arXiv2024

Token-Level Adversarial Prompt Detection Based on Perplexity Measures and Contextual Information

In recent years, Large Language Models (LLM) have emerged as pivotal tools in various applications. However, these models are susceptible to adversarial prompt attacks, where attackers can carefully curate input strings that mislead LLMs into generating incorrect or undesired outputs. Previous work has revealed that with relatively simple yet effective attacks based on discrete optimization, it is possible to generate adversarial prompts that bypass moderation and alignment of the models. This vulnerability to adversarial prompts underscores a significant concern regarding the robustness and reliability of LLMs. Our work aims to address this concern by introducing a novel approach to detecting adversarial prompts at a token level, leveraging the LLM's capability to predict the next token's probability. We measure the degree of the model's perplexity, where tokens predicted with high probability are considered normal, and those exhibiting high perplexity are flagged as adversarial. Additionaly, our method also integrates context understanding by incorporating neighboring token information to encourage the detection of contiguous adversarial prompt sequences. To this end, we design two algorithms for adversarial prompt detection: one based on optimization techniques and another on Probabilistic Graphical Models (PGM). Both methods are equipped with efficient solving methods, ensuring efficient adversarial prompt detection. Our token-level detection result can be visualized as heatmap overlays on the text sequence, allowing for a clearer and more intuitive representation of which part of the text may contain adversarial prompts.

cs.CL↗

arXiv2024

Learning from History: Task-agnostic Model Contrastive Learning for Image Restoration

Contrastive learning has emerged as a prevailing paradigm for high-level vision tasks, which, by introducing properly negative samples, has also been exploited for low-level vision tasks to achieve a compact optimization space to account for their ill-posed nature. However, existing methods rely on manually predefined and task-oriented negatives, which often exhibit pronounced task-specific biases. To address this challenge, our paper introduces an innovative method termed 'learning from history', which dynamically generates negative samples from the target model itself. Our approach, named Model Contrastive Learning for Image Restoration (MCLIR), rejuvenates latency models as negative models, making it compatible with diverse image restoration tasks. We propose the Self-Prior guided Negative loss (SPN) to enable it. This approach significantly enhances existing models when retrained with the proposed model contrastive paradigm. The results show significant improvements in image restoration across various tasks and architectures. For example, models retrained with SPN outperform the original FFANet and DehazeFormer by 3.41 dB and 0.57 dB on the RESIDE indoor dataset for image dehazing. Similarly, they achieve notable improvements of 0.47 dB on SPA-Data over IDT for image deraining and 0.12 dB on Manga109 for a 4x scale super-resolution over lightweight SwinIR, respectively. Code and retrained models are available at https://github.com/Aitical/MCLIR.

cs.CV↗

arXiv2023

AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models

Safety alignment of Large Language Models (LLMs) can be compromised with manual jailbreak attacks and (automatic) adversarial attacks. Recent studies suggest that defending against these attacks is possible: adversarial attacks generate unlimited but unreadable gibberish prompts, detectable by perplexity-based filters; manual jailbreak attacks craft readable prompts, but their limited number due to the necessity of human creativity allows for easy blocking. In this paper, we show that these solutions may be too optimistic. We introduce AutoDAN, an interpretable, gradient-based adversarial attack that merges the strengths of both attack types. Guided by the dual goals of jailbreak and readability, AutoDAN optimizes and generates tokens one by one from left to right, resulting in readable prompts that bypass perplexity filters while maintaining high attack success rates. Notably, these prompts, generated from scratch using gradients, are interpretable and diverse, with emerging strategies commonly seen in manual jailbreak attacks. They also generalize to unforeseen harmful behaviors and transfer to black-box LLMs better than their unreadable counterparts when using limited training data or a single proxy model. Furthermore, we show the versatility of AutoDAN by automatically leaking system prompts using a customized objective. Our work offers a new way to red-team LLMs and understand jailbreak mechanisms via interpretability.

cs.CR↗

arXiv2023

Distributed Two-tier DRL Framework for Cell-Free Network: Association, Beamforming and Power Allocation

Intelligent wireless networks have long been expected to have self-configuration and self-optimization capabilities to adapt to various environments and demands. In this paper, we develop a novel distributed hierarchical deep reinforcement learning (DHDRL) framework with two-tier control networks in different timescales to optimize the long-term spectrum efficiency (SE) of the downlink cell-free multiple-input single-output (MISO) network, consisting of multiple distributed access points (AP) and user terminals (UT). To realize the proposed two-tier control strategy, we decompose the optimization problem into two sub-problems, AP-UT association (AUA) as well as beamforming and power allocation (BPA), resulting in a Markov decision process (MDP) and Partially Observable MDP (POMDP). The proposed method consists of two neural networks. At the system level, a distributed high-level neural network is introduced to optimize wireless network structure on a large timescale. While at the link level, a distributed low-level neural network is proposed to mitigate inter-AP interference and improve the transmission performance on a small timescale. Numerical results show that our method is effective for high-dimensional problems, in terms of spectrum efficiency, signaling overhead as well as satisfaction probability, and generalize well to diverse multi-object problems.

eess.SP↗

arXiv2023

VaQuitA: Enhancing Alignment in LLM-Assisted Video Understanding

Recent advancements in language-model-based video understanding have been progressing at a remarkable pace, spurred by the introduction of Large Language Models (LLMs). However, the focus of prior research has been predominantly on devising a projection layer that maps video features to tokens, an approach that is both rudimentary and inefficient. In our study, we introduce a cutting-edge framework, VaQuitA, designed to refine the synergy between video and textual information. At the data level, instead of sampling frames uniformly, we implement a sampling method guided by CLIP-score rankings, which enables a more aligned selection of frames with the given question. At the feature level, we integrate a trainable Video Perceiver alongside a Visual-Query Transformer (abbreviated as VQ-Former), which bolsters the interplay between the input question and the video features. We also discover that incorporating a simple prompt, "Please be critical", into the LLM input can substantially enhance its video comprehension capabilities. Our experimental results indicate that VaQuitA consistently sets a new benchmark for zero-shot video question-answering tasks and is adept at producing high-quality, multi-turn video dialogues with users.

cs.CV↗

arXiv2023

Semi-randomized block Kaczmarz methods with simple random sampling for large-scale linear systems

Randomized block Kaczmraz method plays an important role in solving large-scale linear system. One of the key points of this type of methods is how to effectively select working rows. However, in most of the state-of-the-art randomized block Kaczmarz-type methods, one has to scan all the rows of the coefficient matrix in advance for computing probabilities or paving, or to compute the residual vector of the linear system in each iteration to determine the working rows. Thus, we have to access all the rows of the data matrix in these methods, which are unfavorable for big-data problems. Moreover, to the best of our knowledge, how to efficiently choose working rows in randomized block Kaczmarz-type methods for multiple linear systems is still an open problem. In order to deal with these problems, we propose semi-randomized block Kaczmarz methods with simple random sampling for linear systems with single and multiple right-hand sides, respectively. In these methods, there is no need to scan or pave all the rows of the coefficient matrix, nor to compute probabilities and the residual vector of the linear system in each step. Specifically, one can update all the solutions of a large-scale linear system with multiple right-hand sides simultaneously. The convergence of the proposed methods is considered. Numerical experiments on both real-world and synthetic data sets show that the proposed methods are superior to many state-of-the-art randomized Kaczmarz-type methods for large-scale linear systems.

math.NA↗

arXiv2023

A semi-randomized and augmented Kaczmarz method with simple random sampling for large-scale inconsistent linear systems

A greedy randomized augmented Kaczmarz (GRAK) method was proposed in [Z.-Z. Bai and W.-T. WU, SIAM J. Sci. Comput., 43 (2021), pp. A3892-A3911] for large and sparse inconsistent linear systems. However, one has to construct two new index sets via computing residual vector with respect to the augmented linear system in each iteration. Thus, the computational overhead of this method is large for extremely large-scale problems. Moreover, there is no reliable stopping criterion for this method. In this work, we are interested in solving large-scale sparse or dense inconsistent linear systems, and try to enhance the numerical performance of the GRAK method. First, we propose an accelerated greedy randomized augmented Kaczmarz method. Theoretical analysis indicates that it converges faster than the GRAK method under very weak assumptions. Second, in order to further release the overhead, we propose a semi-randomized augmented Kaczmarz method with simple random sampling. Third, to the best of our knowledge, there are no practical stopping criteria for all the randomized Kaczmarz-type methods till now. To fill-in this gap, we introduce a practical stopping criterion for Kaczmarz-type methods, and show its rationality from a theoretical point of view. Numerical experiments are performed on both real-world and synthetic data sets, which demonstrate the efficiency of the proposed methods and the effectiveness of our stopping criterion.

math.NA↗

arXiv2023

The Qitai Radio Telescope

This study presents a general outline of the Qitai radio telescope (QTT) project. Qitai, the site of the telescope, is a county of Xinjiang Uygur Autonomous Region of China, located in the east Tianshan Mountains at an elevation of about 1800 m. The QTT is a fully steerable, Gregorian type telescope with a standard parabolic main reflector of 110 m diameter. The QTT has adopted an um-brella support, homology-symmetric lightweight design. The main reflector is active so that the deformation caused by gravity can be corrected. The structural design aims to ultimately allow high-sensitivity observations from 150 MHz up to 115 GHz. To satisfy the requirements for early scientific goals, the QTT will be equipped with ultra-wideband receivers and large field-of-view mul-ti-beam receivers. A multi-function signal-processing system based on RFSoC and GPU processor chips will be developed. These will enable the QTT to operate in pulsar, spectral line, continuum and Very Long Baseline Interferometer (VLBI) observing modes. Electromagnetic compatibility (EMC) and radio frequency interference (RFI) control techniques are adopted throughout the system design. The QTT will form a world-class observational platform for the detection of low-frequency (nanoHertz) gravitational waves through pulsar timing array (PTA) techniques, pulsar surveys, the discovery of binary black-hole systems, and exploring dark matter and the origin of life in the universe.

astro-ph.IM↗

arXiv2023

Dual-Granularity Contrastive Learning for Session-based Recommendation

Session-based recommendation systems(SBRS) are more suitable for the current e-commerce and streaming media recommendation scenarios and thus have become a hot topic. The data encountered by SBRS is typically highly sparse, which also serves as one of the bottlenecks limiting the accuracy of recommendations. So Contrastive Learning(CL) is applied in SBRS owing to its capability of improving embedding learning under the condition of sparse data. However, existing CL strategies are limited in their ability to enforce finer-grained (e.g., factor-level) comparisons and, as a result, are unable to capture subtle differences between instances. More than that, these strategies usually use item or segment dropout as a means of data augmentation which may result in sparser data and thus ineffective self-supervised signals. By addressing the two aforementioned limitations, we introduce a novel multi-granularity CL framework. Specifically, two extra augmented embedding convolution channels with different granularities are constructed and the embeddings learned by them are compared with those learned from original view to complete the CL tasks. At factor-level, we employ Disentangled Representation Learning to obtain finer-grained data(e.g. factor-level embeddings), with which we can construct factor-level convolution channels. At item-level, the star graph is deployed as the augmented data and graph convolution on it can ensure the effectiveness of self-supervised signals. Compare the learned embeddings of these two views with the learned embeddings of the basic view to achieve CL at two granularities. Finally, the more precise item-level and factor-level embeddings obtained are referenced to generate personalized recommendations for the user. The proposed model is validated through extensive experiments on two benchmark datasets, showcasing superior performance compared to existing methods.

cs.IR↗

arXiv2023

First-order perturbation theory of trust-region subproblems

Trust-region subproblem (TRS) is an important problem arising in many applications such as numerical optimization, Tikhonov regularization of ill-posed problems, and constrained eigenvalue problems. In recent decades, extensive works focus on how to solve the trust-region subproblem efficiently. To the best of our knowledge, there are few results on perturbation analysis of the trust-region subproblem. In order to fill in this gap, we focus on first-order perturbation theory of the trust-region subproblem. The main contributions of this paper are three-fold. First, suppose that the TRS is in \emph{easy case}, we give a sufficient condition under which the perturbed TRS is still in easy case. Second, with the help of the structure of the TRS and the classical eigenproblem perturbation theory, we perform first-order perturbation analysis on the Lagrange multiplier and the solution of the TRS, and define their condition numbers. Third, we point out that the solution and the Lagrange multiplier could be well-conditioned even if TRS is in {\it nearly hard case}. The established results are computable, and are helpful to evaluate ill-conditioning of the TRS problem beforehand. Numerical experiments show the sharpness of the established bounds and the effectiveness of the proposed strategies.

math.NA↗

arXiv2023

Syntax-aware Hybrid prompt model for Few-shot multi-modal sentiment analysis

Multimodal Sentiment Analysis (MSA) has been a popular topic in natural language processing nowadays, at both sentence and aspect level. However, the existing approaches almost require large-size labeled datasets, which bring about large consumption of time and resources. Therefore, it is practical to explore the method for few-shot sentiment analysis in cross-modalities. Previous works generally execute on textual modality, using the prompt-based methods, mainly two types: hand-crafted prompts and learnable prompts. The existing approach in few-shot multi-modality sentiment analysis task has utilized both methods, separately. We further design a hybrid pattern that can combine one or more fixed hand-crafted prompts and learnable prompts and utilize the attention mechanisms to optimize the prompt encoder. The experiments on both sentence-level and aspect-level datasets prove that we get a significant outperformance.

cs.CL↗

arXiv2023

ALMA and VLBA views on the outflow associated with an O-type protostar in G26.50+0.28

Protostellar jets and outflows are essential ingredients of the star formation process. A better understanding of this phenomenon is important in its own right as well as for many fundamental aspects of star formation. Jets and outflows associated with O-type protostars are rarely studied with observations reaching the close vicinity of the protostars. In this work, we report high-resolution ALMA and VLBA observations to reveal a clear and consistent picture of an outflow associated with an O-type protostar candidate in the G26.50+0.28 region. These observations reveal, for the first time, a collimated jet located in the middle of the outflow cavity. The jet is found to be perpendicular to an elongated disk/toroid and its velocity gradient. The collimated jet appears to show a small amplitude ($α$$\approx$0$\,.\!\!^{\circ}$06) counterclockwise precession, when looking along the blueshifted jet axis from the strongest continuum source MM1, with a precession length of 0.22 pc. The inclination of the jet is likely to be very low ($\approx$8$^{\circ}$), which makes it a promising target to study its transverse morphologies and kinematics. However, no clear evidence of jet rotation is found in the ALMA and VLBA observations. The three-dimensional velocities of the water maser spots appear to show the same absolute speed with respect to different opening angles, suggesting the jet winds may be launched in a relatively small region. This favors the X-wind model, that is, jets are launched in a small area near the inner disk edge.

astro-ph.GA↗

arXiv2023

A Practical Contrastive Learning Framework for Single-Image Super-Resolution

Contrastive learning has achieved remarkable success on various high-level tasks, but there are fewer contrastive learning-based methods proposed for low-level tasks. It is challenging to adopt vanilla contrastive learning technologies proposed for high-level visual tasks to low-level image restoration problems straightly. Because the acquired high-level global visual representations are insufficient for low-level tasks requiring rich texture and context information. In this paper, we investigate the contrastive learning-based single image super-resolution from two perspectives: positive and negative sample construction and feature embedding. The existing methods take naive sample construction approaches (e.g., considering the low-quality input as a negative sample and the ground truth as a positive sample) and adopt a prior model (e.g., pre-trained VGG model) to obtain the feature embedding. To this end, we propose a practical contrastive learning framework for SISR, named PCL-SR. We involve the generation of many informative positive and hard negative samples in frequency space. Instead of utilizing an additional pre-trained network, we design a simple but effective embedding network inherited from the discriminator network which is more task-friendly. Compared with existing benchmark methods, we re-train them by our proposed PCL-SR framework and achieve superior performance. Extensive experiments have been conducted to show the effectiveness and technical contributions of our proposed PCL-SR thorough ablation studies. The code and pre-trained models can be found at https://github.com/Aitical/PCL-SISR.

cs.CV↗