Searcharxiv⌕ Search

arXiv subjects

Kai Zhu

Publications and source records attributed to Kai Zhu.

At least 91 records · Page 5Linked to original sources
arXiv2023

Signal significance incorporating systematic uncertainty for continuous test

To properly estimate signal significance while accounting for both statistical and systematic uncertainties, we conducted a study to analyze the impact of typical systematic uncertainties, such as background shape, signal shape, and the number of backgrounds, on significance calculation using the continuous test method. Our investigation reveals unexpected and complex features, leading us to recommend a conservative approach: one should estimate signal significance by conducting trials with as many as possible combinations of various uncertainties associated with the fitting procedure, and then select the "worst" outcome as the final result.

physics.data-an↗
arXiv2023

CCM: Adding Conditional Controls to Text-to-Image Consistency Models

Consistency Models (CMs) have showed a promise in creating visual content efficiently and with high quality. However, the way to add new conditional controls to the pretrained CMs has not been explored. In this technical report, we consider alternative strategies for adding ControlNet-like conditional control to CMs and present three significant findings. 1) ControlNet trained for diffusion models (DMs) can be directly applied to CMs for high-level semantic controls but struggles with low-level detail and realism control. 2) CMs serve as an independent class of generative models, based on which ControlNet can be trained from scratch using Consistency Training proposed by Song et al. 3) A lightweight adapter can be jointly optimized under multiple conditions through Consistency Training, allowing for the swift transfer of DMs-based ControlNet to CMs. We study these three solutions across various conditional controls, including edge, depth, human pose, low-resolution image and masked image with text-to-image latent consistency models.

cs.CV↗
arXiv2023

Likelihood-Aware Semantic Alignment for Full-Spectrum Out-of-Distribution Detection

Full-spectrum out-of-distribution (F-OOD) detection aims to accurately recognize in-distribution (ID) samples while encountering semantic and covariate shifts simultaneously. However, existing out-of-distribution (OOD) detectors tend to overfit the covariance information and ignore intrinsic semantic correlation, inadequate for adapting to complex domain transformations. To address this issue, we propose a Likelihood-Aware Semantic Alignment (LSA) framework to promote the image-text correspondence into semantically high-likelihood regions. LSA consists of an offline Gaussian sampling strategy which efficiently samples semantic-relevant visual embeddings from the class-conditional Gaussian distribution, and a bidirectional prompt customization mechanism that adjusts both ID-related and negative context for discriminative ID/OOD boundary. Extensive experiments demonstrate the remarkable OOD detection performance of our proposed LSA especially on the intractable Near-OOD setting, surpassing existing methods by a margin of $15.26\%$ and $18.88\%$ on two F-OOD benchmarks, respectively.

cs.CV↗
arXiv2023

MaNGA DynPop -- III. Stellar dynamics versus stellar population relations in 6000 early-type and spiral galaxies: Fundamental Plane, mass-to-light ratios, total density slopes, and dark matter fractions

We present dynamical scaling relations, combined with the stellar population properties, for a subsample of about 6000 nearby galaxies with the most reliable dynamical models extracted from the full MaNGA sample of 10K galaxies. We show that the inclination-corrected mass plane (MP) for both early-type galaxies (ETGs) and late-type galaxies (LTGs), which links dynamical mass, projected half-light radius $R_{\rm e}$, and the second stellar velocity moment $σ_{\rm e}$ within $R_{\rm e}$, satisfies the virial theorem and is even tighter than the uncorrected one. We find a clear parabolic relation between $\lg(M/L)_{\rm e}$, the total mass-to-light ratio within a sphere of radius $R_{\rm e}$, and $\lgσ_{\rm e}$, with the $M/L$ increasing with $σ_{\rm e}$ and for older stellar populations. However, the relation for ETGs is linear and the one for the youngest galaxies is constant. We confirm and improve the relation between average logarithmic total density slopes $\overline{γ_{\rm T}}$ and $σ_{\rm e}$: $\overline{γ_{\rm T}}$ become steeper with increasing $σ_{\rm e}$ until $\lg(σ_{\rm e}/{\rm km\,s^{-1}})\approx 2.2$ and then remain constant around $\overline{γ_{\rm T}}\approx -2.2$. The $\overline{γ_{\rm T}}-σ_{\rm e}$ variation is larger for LTGs than ETGs. At fixed $σ_{\rm e}$ the total density profiles steepen with galaxy age and for ETGs. We find generally low dark matter fractions, median $f_{\rm DM}(<R_{\rm e})=8$ per cent, within a sphere of radius $R_{\rm e}$. However, we find that $f_{\rm DM}(<R_{\rm e})$ depends on $σ_{\rm e}$ better than stellar mass: dark matter increases to a median $f_{\rm DM}(<R_{\rm e})=33$ percent for galaxies with $σ_{\rm e}\lesssim100{\rm km\,s^{-1}}$. The increased $f_{\rm DM}(<R_{\rm e})$ at low $σ_{\rm e}$ explains the parabolic $\lg(M/L)_{\rm e}-\lgσ_{\rm e}$ relation.

astro-ph.GA↗
arXiv2023

MaNGA DynPop -- II. Global stellar population, gradients, and star-formation histories from integral-field spectroscopy of 10K galaxies: link with galaxy rotation, shape, and total-density gradients

This is the second paper of the MaNGA DynPop series, which analyzes the global stellar population, radial gradients, and non-parametric star-formation history of $\sim 10$K galaxies from the MaNGA Survey final data release 17 (DR17) and relates them with dynamical properties of galaxies. We confirm the correlation between the stellar population properties and the stellar velocity dispersion $σ_{\rm e}$, but also find that younger galaxies are more metal-poor at fixed $σ_{\rm e}$. Stellar age, metallicity, and mass-to-light ratio $M_{\ast}/L$ all decrease with increasing galaxy rotation, while their radial gradients become more negative (i.e. lower value at the outskirts). The exception is the slow rotators, which also appear to have significantly negative metallicity gradients, confirming the mass-metallicity gradient correlation. Massive disk galaxies in the green valley, on the $(σ_{\rm e},\rm Age)$ plane, show the most negative age and metallicity gradients, consistent with their old central bulges surrounded by young star-forming disks and metal-poor gas accretion. Galaxies with high $σ_{\rm e}$, steep total mass-density slope, low dark matter fraction, high $M_{\ast}/L$, and high metallicity have the highest star-formation rate at earlier times, and are currently quenched. We also discover a population of low-mass star-forming galaxies with low rotation but physically distinct from the massive slow rotators. A catalogue of these stellar population properties is provided publicly.

astro-ph.GA↗
arXiv2023

Fraction Constraint in Partial Wave Analysis

To resolve the non-convex optimization problem in partial wave analysis, this paper introduces a novel approach that incorporates fraction constraints into the likelihood function. This method offers significant improvements in both the efficiency of pole searching and the reliability of resonance selection within partial wave analysis.

hep-ex↗
arXiv2023

Background Activation Suppression for Weakly Supervised Object Localization and Semantic Segmentation

Weakly supervised object localization and semantic segmentation aim to localize objects using only image-level labels. Recently, a new paradigm has emerged by generating a foreground prediction map (FPM) to achieve pixel-level localization. While existing FPM-based methods use cross-entropy to evaluate the foreground prediction map and to guide the learning of the generator, this paper presents two astonishing experimental observations on the object localization learning process: For a trained network, as the foreground mask expands, 1) the cross-entropy converges to zero when the foreground mask covers only part of the object region. 2) The activation value continuously increases until the foreground mask expands to the object boundary. Therefore, to achieve a more effective localization performance, we argue for the usage of activation value to learn more object regions. In this paper, we propose a Background Activation Suppression (BAS) method. Specifically, an Activation Map Constraint (AMC) module is designed to facilitate the learning of generator by suppressing the background activation value. Meanwhile, by using foreground region guidance and area constraint, BAS can learn the whole region of the object. In the inference phase, we consider the prediction maps of different categories together to obtain the final localization results. Extensive experiments show that BAS achieves significant and consistent improvement over the baseline methods on the CUB-200-2011 and ILSVRC datasets. In addition, our method also achieves state-of-the-art weakly supervised semantic segmentation performance on the PASCAL VOC 2012 and MS COCO 2014 datasets. Code and models are available at https://github.com/wpy1999/BAS-Extension.

cs.CV↗
arXiv2023

A complete catalogue of broad-line AGNs and double-peaked emission lines from MaNGA integral-field spectroscopy of 10K galaxies: stellar population of AGNs, supermassive black holes, and dual AGNs

We analyse the integral-field spectroscopy data for the $\approx10,000$ galaxies in final data release of the MaNGA survey. We identify 188 galaxies for which the emission lines cannot be described by single Gaussian components. These galaxies can be classified into (1) 38 galaxies with broad $Hα$ and [OIII] $λ$5007 lines, (2) 101 galaxies with broad $Hα$ lines but no broad [OIII] $λ$5007 lines, and (3) 49 galaxies with double-peaked narrow emission lines. Most of the broad line galaxies are classified as Active Galactic Nuclei (AGN) from their line ratios. The catalogue helps us further understand the AGN-galaxy coevolution through the stellar population of broad-line region host galaxies and the relation between broad lines' properties and the host galaxies' dynamical properties. The stellar population properties (including mass, age and metallicity) of broad-line host galaxies suggest there is no significant difference between narrow-line Seyfert-2 galaxies and Type-1 AGN with broad $Hα$ lines. We use the broad-$Hα$ line width and luminosity to estimate masses of black hole in these galaxies, and test the $M_{BH}-σ_{e}$ relation in Type-1 AGN host galaxies. Furthermore we find three dual AGN candidates supported by radio images from the VLA FIRST survey. This sample may be useful for further studies on AGN activities and feedback processes.

astro-ph.GA↗
arXiv2023

Regularized Mask Tuning: Uncovering Hidden Knowledge in Pre-trained Vision-Language Models

Prompt tuning and adapter tuning have shown great potential in transferring pre-trained vision-language models (VLMs) to various downstream tasks. In this work, we design a new type of tuning method, termed as regularized mask tuning, which masks the network parameters through a learnable selection. Inspired by neural pathways, we argue that the knowledge required by a downstream task already exists in the pre-trained weights but just gets concealed in the upstream pre-training stage. To bring the useful knowledge back into light, we first identify a set of parameters that are important to a given downstream task, then attach a binary mask to each parameter, and finally optimize these masks on the downstream data with the parameters frozen. When updating the mask, we introduce a novel gradient dropout strategy to regularize the parameter selection, in order to prevent the model from forgetting old knowledge and overfitting the downstream data. Experimental results on 11 datasets demonstrate the consistent superiority of our method over previous alternatives. It is noteworthy that we manage to deliver 18.73% performance improvement compared to the zero-shot CLIP via masking an average of only 2.56% parameters. Furthermore, our method is synergistic with most existing parameter-efficient tuning methods and can boost the performance on top of them. Project page can be found here (https://wuw2019.github.io/R-AMT/).

cs.CV↗
arXiv2023

Velocity Dispersion $σ_{\rm aper}$ Aperture Corrections as a Function of Galaxy Properties from Integral-field Stellar Kinematics of 10,000 MaNGA Galaxies

The second moment of the stellar velocity within the effective radius, denoted by $σ_{\rm e}^2$, is a crucial quantity in galaxy studies as it provides insight into galaxy properties and their mass distributions. However, large spectroscopic surveys typically do not measure $σ_{\rm e}$ directly, instead providing $σ_{\rm aper}$, the second moment of the stellar velocity within a fixed fiber aperture. In this paper, we derive an empirical aperture correction formula, given by $σ_{\rm aper}/σ_{\rm e}=(R_{\rm aper}/R_{\rm e})^α$, using spatially resolved stellar kinematics extracted from approximately 10,000 Sloan Digital Sky Survey-Mapping Nearby Galaxies at Apache Point Observatory (SDSS-MaNGA) integral field unit observations. Our analysis reveals a strong dependence of $α$ on the $r$-band absolute magnitude $M_{\rm r}$, $g-i$ color, and Sersic index $n_{\rm Ser}$, where $α$ values are lower for brighter, redder galaxies with higher Sersic indices. Our results demonstrate that the aperture correction derived from previous literature on early-type galaxies cannot be applied to predict the aperture corrections for galaxies with intermediate Sersic indices. We provide a lookup table of $α$ values for different galaxy types, with parameters in the ranges of $-18>M_{\rm r}>-24$, $0.4<g-i<1.6$, and $0<n_{\rm Ser}<8$. A Python script is provided to obtain the correction factors from the lookup table.

astro-ph.GA↗
arXiv2023

Global Inequality in Cooling from Urban Green Spaces and its Climate Change Adaptation Potential

Heat extremes are projected to severely impact humanity and with increasing geographic disparities. Global South countries are more exposed to heat extremes and have reduced adaptation capacity. One documented source of such adaptation inequality is a lack of resources to cool down indoor temperatures. Less is known about the capacity to ameliorate outdoor heat stress. Here, we assess global inequality in green infrastructure, on which urban residents critically rely to ameliorate lethal heat stress outdoors. We use satellite-derived indicators of land surface temperature and urban green space area to quantify the daytime cooling capacity of urban green spaces in the hottest months across ~500 cities with population size over 1 million per city globally. Our results show a striking contrast with an about two-fold lower cooling capacity in Global South cities compared to the Global North (2.1 degrees Celsius vs. 3.8 degrees Celsius). A similar gap occurs for the cooling adaptation benefits received by an average urban resident (Global South 1.9 degrees Celsius vs. North 3.6 degrees Celsius), i.e., accounting for relative spatial distributions of people and urban green spaces. This cooling adaptation inequality is attributed to the discrepancies in urban green space quantity and quality between Global North and South cities, jointly shaped by natural and socioeconomic factors. Our analyses suggest vast potential for enhancing outdoor cooling adaptation while reducing its global inequality through expanding and optimizing urban green infrastructure.

stat.AP↗
arXiv2023

MaNGA DynPop -- I. Quality-assessed stellar dynamical modelling from integral-field spectroscopy of 10K nearby galaxies: a catalogue of masses, mass-to-light ratios, density profiles and dark matter

This is the first paper in our series on the combined analysis of the Dynamics and stellar Population (DynPop) for the MaNGA survey in the final SDSS Data Release 17 (DR17). Here we present a catalogue of dynamically-determined quantities for over 10000 nearby galaxies based on integral-field stellar kinematics from the MaNGA survey. The dynamical properties are extracted using the axisymmetric Jeans Anisotropic Modelling (JAM) method, which was previously shown to be the most accurate for this kind of study. We assess systematic uncertainties using eight dynamical models with different assumptions. We use two orientations of the velocity ellipsoid: either cylindrically-aligned JAM$_{\rm cyl}$ or spherically-aligned JAM$_{\rm sph}$. We also make four assumptions for the models' dark vs. luminous matter distributions: (1) mass-follows-light, (2) free NFW dark halo, (3) cosmologically-constrained NFW halo, (4) generalized NFW dark halo, i.e. with free inner slope. In this catalogue, we provide the quantities related to the mass distributions (e.g. the density slopes and enclosed mass within a sphere of a given radius for total mass, stellar mass, and dark matter mass components). We also provide the complete models which can be used to compute the full luminous and mass distribution of each galaxy. Additionally, we visually assess the qualities of the models to help with model selections. We estimate the observed scatter in the measured quantities which decreases as expected with improvements in quality. For the best data quality, we find a remarkable consistency of measured quantities between different models, highlighting the robustness of the results.

astro-ph.GA↗
arXiv2023

Cones 2: Customizable Image Synthesis with Multiple Subjects

Synthesizing images with user-specified subjects has received growing attention due to its practical applications. Despite the recent success in single subject customization, existing algorithms suffer from high training cost and low success rate along with increased number of subjects. Towards controllable image synthesis with multiple subjects as the constraints, this work studies how to efficiently represent a particular subject as well as how to appropriately compose different subjects. We find that the text embedding regarding the subject token already serves as a simple yet effective representation that supports arbitrary combinations without any model tuning. Through learning a residual on top of the base embedding, we manage to robustly shift the raw subject to the customized subject given various text conditions. We then propose to employ layout, a very abstract and easy-to-obtain prior, as the spatial guidance for subject arrangement. By rectifying the activations in the cross-attention map, the layout appoints and separates the location of different subjects in the image, significantly alleviating the interference across them. Both qualitative and quantitative experimental results demonstrate our superiority over state-of-the-art alternatives under a variety of settings for multi-subject customization.

cs.CV↗
arXiv2023

A Transient Electrical-Thermal Co-Simulation Method with LTS for Multiscale Structures

In this article, an efficient transient electricalthermal co-simulation method based on the finite element method (FEM) and the discontinuous Galerkin time-domain (DGTD) method is developed for electrical-thermal coupling analysis of multiscale structures. Two Independent meshes are adopted by the steady electrical analysis and the transient thermal simulation to avoid redundant overhead. In order to enhance the feasibility and efficiency of solving multiscale and sophisticated structures, a local time stepping (LTS) technique coupled with an interpolation method is incorporated into the co-simulation method. Several numerical examples from simple structures to complex multiscale PDN structures are carried out to demonstrate the accuracy and efficiency of the proposed method by comparing with the COMSOL. Finally, two practical numerical examples are considered to confirm the performance of the proposed method for complex and multiscale structures.

math.NA↗
arXiv2023

Uncertainty-Aware Optimal Transport for Semantically Coherent Out-of-Distribution Detection

Semantically coherent out-of-distribution (SCOOD) detection aims to discern outliers from the intended data distribution with access to unlabeled extra set. The coexistence of in-distribution and out-of-distribution samples will exacerbate the model overfitting when no distinction is made. To address this problem, we propose a novel uncertainty-aware optimal transport scheme. Our scheme consists of an energy-based transport (ET) mechanism that estimates the fluctuating cost of uncertainty to promote the assignment of semantic-agnostic representation, and an inter-cluster extension strategy that enhances the discrimination of semantic property among different clusters by widening the corresponding margin distance. Furthermore, a T-energy score is presented to mitigate the magnitude gap between the parallel transport and classifier branches. Extensive experiments on two standard SCOOD benchmarks demonstrate the above-par OOD detection performance, outperforming the state-of-the-art methods by a margin of 27.69% and 34.4% on FPR@95, respectively.

cs.CV↗
arXiv2023

Cones: Concept Neurons in Diffusion Models for Customized Generation

Human brains respond to semantic features of presented stimuli with different neurons. It is then curious whether modern deep neural networks admit a similar behavior pattern. Specifically, this paper finds a small cluster of neurons in a diffusion model corresponding to a particular subject. We call those neurons the concept neurons. They can be identified by statistics of network gradients to a stimulation connected with the given subject. The concept neurons demonstrate magnetic properties in interpreting and manipulating generation results. Shutting them can directly yield the related subject contextualized in different scenes. Concatenating multiple clusters of concept neurons can vividly generate all related concepts in a single image. A few steps of further fine-tuning can enhance the multi-concept capability, which may be the first to manage to generate up to four different subjects in a single image. For large-scale applications, the concept neurons are environmentally friendly as we only need to store a sparse cluster of int index instead of dense float32 values of the parameters, which reduces storage consumption by 90\% compared with previous subject-driven generation methods. Extensive qualitative and quantitative studies on diverse scenarios show the superiority of our method in interpreting and manipulating diffusion models.

cs.CV↗
arXiv2022

Neural Dependencies Emerging from Learning Massive Categories

This work presents two astonishing findings on neural networks learned for large-scale image classification. 1) Given a well-trained model, the logits predicted for some category can be directly obtained by linearly combining the predictions of a few other categories, which we call \textbf{neural dependency}. 2) Neural dependencies exist not only within a single model, but even between two independently learned models, regardless of their architectures. Towards a theoretical analysis of such phenomena, we demonstrate that identifying neural dependencies is equivalent to solving the Covariance Lasso (CovLasso) regression problem proposed in this paper. Through investigating the properties of the problem solution, we confirm that neural dependency is guaranteed by a redundant logit covariance matrix, which condition is easily met given massive categories, and that neural dependency is highly sparse, implying that one category correlates to only a few others. We further empirically show the potential of neural dependencies in understanding internal data correlations, generalizing models to unseen categories, and improving model robustness with a dependency-derived regularizer. Code for this work will be made publicly available.

cs.LG↗
arXiv2022

Mapping the Pathways of Photo-induced Ion Migration in Organic-inorganic Hybrid Halide Perovskites

Organic-inorganic hybrid perovskites (OIHPs) exhibiting exceptional photovoltaic and optoelectronic properties are of fundamental and practical interest, owing to their tunability and low manufacturing cost. For practical applications, however, challenges such as material instability and the photocurrent hysteresis occurring in perovskite solar cells under light exposure need to be understood and addressed. While extensive investigations have suggested that ion migration is a plausible origin of these detrimental effects, detailed understanding of the ion migration pathways remains elusive. Here, we report the characterization of photo-induced ion migration in OIHPs using \textit{in situ} laser illumination inside a scanning electron microscope, coupled with secondary electron imaging, energy-dispersive X-ray spectroscopy and cathodoluminescence with varying primary electron energies. Using methylammonium lead iodide (MAPbI$_3$), formamidinium lead iodide (FAPbI$_3$) and hybrid formamidinium-methylammonium lead iodide as model systems, we observed photo-induced long-range migration of halide ions over hundreds of micrometers and elucidated the transport pathways of various ions both near the surface and inside the bulk of the OIHPs, including a surprising finding of the vertical migration of lead ions. Our study provides insights into ion migration processes in OIHPs that can aid OIHP material design and processing in future applications.

physics.app-ph↗