arXiv subjects

Kai Wang

Publications and source records attributed to Kai Wang.

At least 523 records · Page 29Linked to original sources

arXiv2023

InfoBatch: Lossless Training Speed Up by Unbiased Dynamic Data Pruning

Data pruning aims to obtain lossless performances with less overall cost. A common approach is to filter out samples that make less contribution to the training. This could lead to gradient expectation bias compared to the original data. To solve this problem, we propose \textbf{InfoBatch}, a novel framework aiming to achieve lossless training acceleration by unbiased dynamic data pruning. Specifically, InfoBatch randomly prunes a portion of less informative samples based on the loss distribution and rescales the gradients of the remaining samples to approximate the original gradient. As a plug-and-play and architecture-agnostic framework, InfoBatch consistently obtains lossless training results on classification, semantic segmentation, vision pertaining, and instruction fine-tuning tasks. On CIFAR10/100, ImageNet-1K, and ADE20K, InfoBatch losslessly saves 40\% overall cost. For pertaining MAE and diffusion model, InfoBatch can respectively save 24.8\% and 27\% cost. For LLaMA instruction fine-tuning, InfoBatch is also able to save 20\% cost and is compatible with coreset selection methods. The code is publicly available at \href{https://github.com/henryqin1997/InfoBatch}{github.com/NUS-HPC-AI-Lab/InfoBatch}.

cs.CV↗

arXiv2023

A Conditional Abundance Matching Method of Extending Simulated Halo Merger Trees to Resolve Low-Mass Progenitors and Sub-halos

We present an algorithm to extend subhalo merger trees in a low-resolution dark-matter-only simulation by conditionally matching them to those in a high-resolution simulation. The algorithm is general and can be applied to simulation data with different resolutions using different target variables. We instantiate the algorithm by a case in which trees from ELUCID, a constrained simulation of $(500h^{-1}{\rm Mpc})^3$ volume of the local universe, are extended by matching trees from TNGDark, a simulation with much higher resolution. Our tests show that the extended trees are statistically equivalent to the high-resolution trees in the joint distribution of subhalo quantities and in important summary statistics relevant to modeling galaxy formation and evolution in halos. The extended trees preserve certain information of individual systems in the target simulation, including properties of resolved satellite subhalos, and shapes and orientations of their host halos. With the extension, subhalo merger trees in a cosmological scale simulation are extrapolated to a mass resolution comparable to that in a higher-resolution simulation carried out in a smaller volume, which can be used as the input for (sub)halo-based models of galaxy formation. The source code of the algorithm, and halo merger trees extended to a mass resolution of $\sim 2 \times 10^8 h^{-1}M_\odot$ in the entire ELUCID simulation, are available.

astro-ph.GA↗

arXiv2023

Massive Dark Matter Halos at High Redshift: Implications for Observations in the JWST Era

The presence of massive galaxies at high $z$ as recently observed by JWST appears to contradict the current $Λ$CDM cosmology. Here we aim to alleviate this tension by incorporating uncertainties from three sources in counting galaxies: cosmic variance, error in stellar mass estimation, and backsplash enhancement. Each of these factors significantly increases the cumulative stellar mass density $ρ_*(>M_*)$ at the high-mass end, and their combined effect can boost the density by more than one order of magnitude. Assuming a star formation efficiency of $ε_* \sim 0.5$, cosmic variance alone reduces the tension to a $2σ$ level, except for the most massive galaxy at $z=8$. Additionally, incorporating a 0.3 dex lognormal dispersion in the stellar mass estimation brings the observed $ρ_*(>M_*)$ at $z \sim 7 - 10$ within $2σ$. The tension is completely eliminated when we account for the gas stripped from backsplash halos. These results highlight the importance of fully modeling uncertainties when interpreting observational data of rare objects. We use the constrained simulation, ELUCID, to investigate the descendants of high-$z$ massive galaxies. Our findings reveal that a significant portion of these galaxies ultimately reside in massive halos at $z=0$ with $M_{\rm halo} > 10^{13} h^{-1}M_\odot$. Moreover, a large fraction of local central galaxies in $M_{\rm halo} \geqslant 10^{14.5} h^{-1}M_\odot$ halos are predicted to contain substantial amounts of ancient stars formed in massive galaxies at $z \sim 8$. This prediction can be tested by studying the structure and stellar population of central galaxies in present-day massive clusters.

astro-ph.GA↗

arXiv2023

Evidence of Hadronic Emission from the brightest-of-all-time GRB 221009A

Acceleration of hadrons in relativistic shocks has been long expected and invoked to model GRB high-energy photon and neutrino emissions. However, so far there has been no direct observational evidence of hadronic emission from GRBs. The B.O.A.T. ("brightest of all time") gamma-ray burst (GRB) 221009A had extreme energies (with an isotropic energy exceeding $10^{55}$ erg) and was detected in broad-band including the very-high-energy (VHE, $>100\,\rm GeV$) band up to $>10$ TeV. Here we perform a comprehensive spectral analysis of the GRB from keV to TeV energy range and perform detailed spectral and light curve modelings considering both the traditional synchrotron self-Compton process and the electromagnetic (EM) cascade process initiated by hadronic interactions by accelerated cosmic rays in the external shock. We find that the leptonic scenario alone is not adequate to account for the observations, whereas the proposed scenario with the combination of hadronic and leptonic components can well reproduce the multi-wavelength spectra and the light curve. This result reveals the existence of the accelerated hadronic component in the early afterglow of this extreme burst. According to this scenario, the observed TeV light curve should contain imprints of the prompt MeV emission.

astro-ph.HE↗

arXiv2023

From Halos to Galaxies. VII. The Connections Between Stellar Mass Growth History, Quenching History and Halo Assembly History for Central Galaxies

The assembly of galaxies over cosmic time is tightly connected to the assembly of their host dark matter halos. We investigate the stellar mass growth history and the chemical enrichment history of central galaxies in SDSS-MaNGA. We find that the derived stellar metallicity of passive central galaxies is always higher than that of the star-forming ones. This stellar metallicity enhancement becomes progressively larger towards low-mass galaxies (at a given epoch) and earlier epochs (at a given stellar mass), which suggests strangulation as the primary mechanism for star formation quenching in central galaxies not only in the local universe, but also very likely at higher redshifts up to $z\sim3$. We show that at the same present-day stellar mass, passive central galaxies assembled half of their final stellar mass $\sim 2$ Gyr earlier than star-forming central galaxies, which agrees well with semi-analytic model. Exploring semi-analytic model, we find that this is because passive central galaxies reside in, on average, more massive halos with a higher halo mass increase rate across cosmic time. As a consequence, passive central galaxies are assembled faster and also quenched earlier than their star-forming counterparts. While at the same present-day halo mass, different halo assembly history also produces very different final stellar mass of the central galaxy within, and halos assembled earlier host more massive centrals with a higher quenched fraction, in particular around the "golden halo mass" at $10^{12}\mathrm{M_\odot}$. Our results call attention back to the dark matter halo as a key driver of galaxy evolution.

astro-ph.GA↗

arXiv2023

Does Graph Distillation See Like Vision Dataset Counterpart?

Training on large-scale graphs has achieved remarkable results in graph representation learning, but its cost and storage have attracted increasing concerns. Existing graph condensation methods primarily focus on optimizing the feature matrices of condensed graphs while overlooking the impact of the structure information from the original graphs. To investigate the impact of the structure information, we conduct analysis from the spectral domain and empirically identify substantial Laplacian Energy Distribution (LED) shifts in previous works. Such shifts lead to poor performance in cross-architecture generalization and specific tasks, including anomaly detection and link prediction. In this paper, we propose a novel Structure-broadcasting Graph Dataset Distillation (SGDD) scheme for broadcasting the original structure information to the generation of the synthetic one, which explicitly prevents overlooking the original structure information. Theoretically, the synthetic graphs by SGDD are expected to have smaller LED shifts than previous works, leading to superior performance in both cross-architecture settings and specific tasks. We validate the proposed SGDD across 9 datasets and achieve state-of-the-art results on all of them: for example, on the YelpChi dataset, our approach maintains 98.6% test accuracy of training on the original graph dataset with 1,000 times saving on the scale of the graph. Moreover, we empirically evaluate there exist 17.6% ~ 31.4% reductions in LED shift crossing 9 datasets. Extensive experiments and analysis verify the effectiveness and necessity of the proposed designs. The code is available in the GitHub repository: https://github.com/RingBDStack/SGDD.

cs.LG↗

arXiv2023

DiM: Distilling Dataset into Generative Model

Dataset distillation reduces the network training cost by synthesizing small and informative datasets from large-scale ones. Despite the success of the recent dataset distillation algorithms, three drawbacks still limit their wider application: i). the synthetic images perform poorly on large architectures; ii). they need to be re-optimized when the distillation ratio changes; iii). the limited diversity restricts the performance when the distillation ratio is large. In this paper, we propose a novel distillation scheme to \textbf{D}istill information of large train sets \textbf{i}nto generative \textbf{M}odels, named DiM. Specifically, DiM learns to use a generative model to store the information of the target dataset. During the distillation phase, we minimize the differences in logits predicted by a models pool between real and generated images. At the deployment stage, the generative model synthesizes various training samples from random noises on the fly. Due to the simple yet effective designs, the trained DiM can be directly applied to different distillation ratios and large architectures without extra cost. We validate the proposed DiM across 4 datasets and achieve state-of-the-art results on all of them. To the best of our knowledge, we are the first to achieve higher accuracy on complex architectures than simple ones, such as 75.1\% with ResNet-18 and 72.6\% with ConvNet-3 on ten images per class of CIFAR-10. Besides, DiM outperforms previous methods with 10\% $\sim$ 22\% when images per class are 1 and 10 on the SVHN dataset.

cs.CV↗

arXiv2023

Expanding Small-Scale Datasets with Guided Imagination

The power of DNNs relies heavily on the quantity and quality of training data. However, collecting and annotating data on a large scale is often expensive and time-consuming. To address this issue, we explore a new task, termed dataset expansion, aimed at expanding a ready-to-use small dataset by automatically creating new labeled samples. To this end, we present a Guided Imagination Framework (GIF) that leverages cutting-edge generative models like DALL-E2 and Stable Diffusion (SD) to "imagine" and create informative new data from the input seed data. Specifically, GIF conducts data imagination by optimizing the latent features of the seed data in the semantically meaningful space of the prior model, resulting in the creation of photo-realistic images with new content. To guide the imagination towards creating informative samples for model training, we introduce two key criteria, i.e., class-maintained information boosting and sample diversity promotion. These criteria are verified to be essential for effective dataset expansion: GIF-SD obtains 13.5% higher model accuracy on natural image datasets than unguided expansion with SD. With these essential criteria, GIF successfully expands small datasets in various scenarios, boosting model accuracy by 36.9% on average over six natural image datasets and by 13.5% on average over three medical datasets. The source code is available at https://github.com/Vanint/DatasetExpansion.

cs.CV↗

arXiv2023

Can pre-trained models assist in dataset distillation?

Dataset Distillation (DD) is a prominent technique that encapsulates knowledge from a large-scale original dataset into a small synthetic dataset for efficient training. Meanwhile, Pre-trained Models (PTMs) function as knowledge repositories, containing extensive information from the original dataset. This naturally raises a question: Can PTMs effectively transfer knowledge to synthetic datasets, guiding DD accurately? To this end, we conduct preliminary experiments, confirming the contribution of PTMs to DD. Afterwards, we systematically study different options in PTMs, including initialization parameters, model architecture, training epoch and domain knowledge, revealing that: 1) Increasing model diversity enhances the performance of synthetic datasets; 2) Sub-optimal models can also assist in DD and outperform well-trained ones in certain cases; 3) Domain-specific PTMs are not mandatory for DD, but a reasonable domain match is crucial. Finally, by selecting optimal options, we significantly improve the cross-architecture generalization over baseline DD methods. We hope our work will facilitate researchers to develop better DD techniques. Our code is available at https://github.com/yaolu-zjut/DDInterpreter.

cs.CV↗

arXiv2023

Dynamic Prompt Learning: Addressing Cross-Attention Leakage for Text-Based Image Editing

Large-scale text-to-image generative models have been a ground-breaking development in generative AI, with diffusion models showing their astounding ability to synthesize convincing images following an input text prompt. The goal of image editing research is to give users control over the generated images by modifying the text prompt. Current image editing techniques are susceptible to unintended modifications of regions outside the targeted area, such as on the background or on distractor objects which have some semantic or visual relationship with the targeted object. According to our experimental findings, inaccurate cross-attention maps are at the root of this problem. Based on this observation, we propose Dynamic Prompt Learning (DPL) to force cross-attention maps to focus on correct noun words in the text prompt. By updating the dynamic tokens for nouns in the textual input with the proposed leakage repairment losses, we achieve fine-grained image editing over particular objects while preventing undesired changes to other image regions. Our method DPL, based on the publicly available Stable Diffusion, is extensively evaluated on a wide range of images, and consistently obtains superior results both quantitatively (CLIP score, Structure-Dist) and qualitatively (on user-evaluation). We show improved prompt editing results for Word-Swap, Prompt Refinement, and Attention Re-weighting, especially for complex multi-object scenes.

cs.CV↗

arXiv2023

Computational Design of Wiring Layout on Tight Suits with Minimal Motion Resistance

An increasing number of electronics are directly embedded on the clothing to monitor human status (e.g., skeletal motion) or provide haptic feedback. A specific challenge to prototype and fabricate such a clothing is to design the wiring layout, while minimizing the intervention to human motion. We address this challenge by formulating the topological optimization problem on the clothing surface as a deformation-weighted Steiner tree problem on a 3D clothing mesh. Our method proposed an energy function for minimizing strain energy in the wiring area under different motions, regularized by its total length. We built the physical prototype to verify the effectiveness of our method and conducted user study with participants of both design experts and smart cloth users. On three types of commercial products of smart clothing, the optimized layout design reduced wire strain energy by an average of 77% among 248 actions compared to baseline design, and 18% over the expert design.

cs.HC↗

arXiv2023

When Large Language Models Meet Citation: A Survey

Citations in scholarly work serve the essential purpose of acknowledging and crediting the original sources of knowledge that have been incorporated or referenced. Depending on their surrounding textual context, these citations are used for different motivations and purposes. Large Language Models (LLMs) could be helpful in capturing these fine-grained citation information via the corresponding textual context, thereby enabling a better understanding towards the literature. Furthermore, these citations also establish connections among scientific papers, providing high-quality inter-document relationships and human-constructed knowledge. Such information could be incorporated into LLMs pre-training and improve the text representation in LLMs. Therefore, in this paper, we offer a preliminary review of the mutually beneficial relationship between LLMs and citation analysis. Specifically, we review the application of LLMs for in-text citation analysis tasks, including citation classification, citation-based summarization, and citation recommendation. We then summarize the research pertinent to leveraging citation linkage knowledge to improve text representations of LLMs via citation prediction, network structure information, and inter-document relationship. We finally provide an overview of these contemporary methods and put forth potential promising avenues in combining LLMs and citation analysis for further investigation.

cs.DL↗

arXiv2023

Plasticity-Optimized Complementary Networks for Unsupervised Continual Learning

Continuous unsupervised representation learning (CURL) research has greatly benefited from improvements in self-supervised learning (SSL) techniques. As a result, existing CURL methods using SSL can learn high-quality representations without any labels, but with a notable performance drop when learning on a many-tasks data stream. We hypothesize that this is caused by the regularization losses that are imposed to prevent forgetting, leading to a suboptimal plasticity-stability trade-off: they either do not adapt fully to the incoming data (low plasticity), or incur significant forgetting when allowed to fully adapt to a new SSL pretext-task (low stability). In this work, we propose to train an expert network that is relieved of the duty of keeping the previous knowledge and can focus on performing optimally on the new tasks (optimizing plasticity). In the second phase, we combine this new knowledge with the previous network in an adaptation-retrospection phase to avoid forgetting and initialize a new expert with the knowledge of the old network. We perform several experiments showing that our proposed approach outperforms other CURL exemplar-free methods in few- and many-task split settings. Furthermore, we show how to adapt our approach to semi-supervised continual learning (Semi-SCL) and show that we surpass the accuracy of other exemplar-free Semi-SCL methods and reach the results of some others that use exemplars.

cs.LG↗

arXiv2023

Classification of integers based on residue classes via modern deep learning algorithms

Judging whether an integer can be divided by prime numbers such as 2 or 3 may appear trivial to human beings, but can be less straightforward for computers. Here, we tested multiple deep learning architectures and feature engineering approaches on classifying integers based on their residues when divided by small prime numbers. We found that the ability of classification critically depends on the feature space. We also evaluated Automated Machine Learning (AutoML) platforms from Amazon, Google and Microsoft, and found that they failed on this task without appropriately engineered features. Furthermore, we introduced a method that utilizes linear regression on Fourier series basis vectors, and demonstrated its effectiveness. Finally, we evaluated Large Language Models (LLMs) such as GPT-4, GPT-J, LLaMA and Falcon, and demonstrated their failures. In conclusion, feature engineering remains an important task to improve performance and increase interpretability of machine-learning models, even in the era of AutoML and LLMs.

cs.LG↗

arXiv2023

Region Generation and Assessment Network for Occluded Person Re-Identification

Person Re-identification (ReID) plays a more and more crucial role in recent years with a wide range of applications. Existing ReID methods are suffering from the challenges of misalignment and occlusions, which degrade the performance dramatically. Most methods tackle such challenges by utilizing external tools to locate body parts or exploiting matching strategies. Nevertheless, the inevitable domain gap between the datasets utilized for external tools and the ReID datasets and the complicated matching process make these methods unreliable and sensitive to noises. In this paper, we propose a Region Generation and Assessment Network (RGANet) to effectively and efficiently detect the human body regions and highlight the important regions. In the proposed RGANet, we first devise a Region Generation Module (RGM) which utilizes the pre-trained CLIP to locate the human body regions using semantic prototypes extracted from text descriptions. Learnable prompt is designed to eliminate domain gap between CLIP datasets and ReID datasets. Then, to measure the importance of each generated region, we introduce a Region Assessment Module (RAM) that assigns confidence scores to different regions and reduces the negative impact of the occlusion regions by lower scores. The RAM consists of a discrimination-aware indicator and an invariance-aware indicator, where the former indicates the capability to distinguish from different identities and the latter represents consistency among the images of the same class of human body regions. Extensive experimental results for six widely-used benchmarks including three tasks (occluded, partial, and holistic) demonstrate the superiority of RGANet against state-of-the-art methods.

cs.CV↗

arXiv2023

FarSight: A Physics-Driven Whole-Body Biometric System at Large Distance and Altitude

Whole-body biometric recognition is an important area of research due to its vast applications in law enforcement, border security, and surveillance. This paper presents the end-to-end design, development and evaluation of FarSight, an innovative software system designed for whole-body (fusion of face, gait and body shape) biometric recognition. FarSight accepts videos from elevated platforms and drones as input and outputs a candidate list of identities from a gallery. The system is designed to address several challenges, including (i) low-quality imagery, (ii) large yaw and pitch angles, (iii) robust feature extraction to accommodate large intra-person variabilities and large inter-person similarities, and (iv) the large domain gap between training and test sets. FarSight combines the physics of imaging and deep learning models to enhance image restoration and biometric feature encoding. We test FarSight's effectiveness using the newly acquired IARPA Biometric Recognition and Identification at Altitude and Range (BRIAR) dataset. Notably, FarSight demonstrated a substantial performance increase on the BRIAR dataset, with gains of +11.82% Rank-20 identification and +11.3% TAR@1% FAR.

cs.CV↗

arXiv2023

Recurrence-Free Survival Prediction for Anal Squamous Cell Carcinoma Chemoradiotherapy using Planning CT-based Radiomics Model

Objectives: Approximately 30% of non-metastatic anal squamous cell carcinoma (ASCC) patients will experience recurrence after chemoradiotherapy (CRT), and currently available clinical variables are poor predictors of treatment response. We aimed to develop a model leveraging information extracted from radiation pretreatment planning CT to predict recurrence-free survival (RFS) in ASCC patients after CRT. Methods: Radiomics features were extracted from planning CT images of 96 ASCC patients. Following pre-feature selection, the optimal feature set was selected via step-forward feature selection with a multivariate Cox proportional hazard model. The RFS prediction was generated from a radiomics-clinical combined model based on an optimal feature set with five repeats of five-fold cross validation. The risk stratification ability of the proposed model was evaluated with Kaplan-Meier analysis. Results: Shape- and texture-based radiomics features significantly predicted RFS. Compared to a clinical-only model, radiomics-clinical combined model achieves better performance in the testing cohort with higher C-index (0.80 vs 0.73) and AUC (0.84 vs 0.79 for 1-year RFS, 0.84 vs 0.78 for 2-year RFS, and 0.86 vs 0.83 for 3-year RFS), leading to distinctive high- and low-risk of recurrence groups (p<0.001). Conclusions: A treatment planning CT based radiomics and clinical combined model had improved prognostic performance in predicting RFS for ASCC patients treated with CRT as compared to a model using clinical features only.

cs.CV↗

arXiv2023

ScrollNet: Dynamic Weight Importance for Continual Learning

The principle underlying most existing continual learning (CL) methods is to prioritize stability by penalizing changes in parameters crucial to old tasks, while allowing for plasticity in other parameters. The importance of weights for each task can be determined either explicitly through learning a task-specific mask during training (e.g., parameter isolation-based approaches) or implicitly by introducing a regularization term (e.g., regularization-based approaches). However, all these methods assume that the importance of weights for each task is unknown prior to data exposure. In this paper, we propose ScrollNet as a scrolling neural network for continual learning. ScrollNet can be seen as a dynamic network that assigns the ranking of weight importance for each task before data exposure, thus achieving a more favorable stability-plasticity tradeoff during sequential task learning by reassigning this ranking for different tasks. Additionally, we demonstrate that ScrollNet can be combined with various CL methods, including regularization-based and replay-based approaches. Experimental results on CIFAR100 and TinyImagenet datasets show the effectiveness of our proposed method. We release our code at https://github.com/FireFYF/ScrollNet.git.

cs.CV↗