arXiv subjects

Keming Zhang

Publications and source records attributed to Keming Zhang.

At least 19 recordsLinked to original sources

arXiv2026

AgentSociety 2: An Integrated Research Environment for Executable Social Science

AI scientist systems are beginning to automate parts of scientific research, but social science poses a distinct challenge: its objects of inquiry are not merely datasets or laboratory protocols, but integrated social processes involving situated participants, interaction contexts, interventions, and outcomes. Yet a critical link is missing: existing systems either assist isolated research tasks or simulate agents as experimental subjects, leaving the research workflow and simulated society decoupled. Here we introduce AgentSociety 2, an Integrated Research Environment for executable social science. It couples two roles of LLM agents in the same runtime: AI social scientists that coordinate literature grounding, hypothesis generation, experiment design, simulation execution, result interpretation, and manuscript drafting; and silicon participants that generate behavioral responses within configurable social environments. This dual-role design turns hypotheses into auditable agent behaviors, environment rules, interventions, and measurements, thereby supporting an end-to-end workflow. Across seven illustrative studies spanning micro-level social-science laboratory experiments, meso-level dynamics in social media, and macro-level urban scenarios, we demonstrate its capacity to support diverse disciplinary questions, reproduce major qualitative patterns from prior studies, identify informative deviations, and enable large-scale simulations through optimized agent-environment interactions. By preserving human researchers' high-level agency while delegating procedural orchestration to agentic systems, it provides a human-in-the-loop and controllable infrastructure for next-generation computational social science, with broader applications in scalable computational social experimentation and AI-enabled social governance platforms.

cs.CY↗

arXiv2026

TrajRAG: Retrieving Geometric-Semantic Experience for Zero-Shot Object Navigation

Existing zero-shot Object Goal Navigation (ObjectNav) methods often exploit commonsense knowledge from large language or vision-language models to guide navigation. However, such knowledge arises from internet-scale text rather than embodied 3D experience, and episodic observations collected during navigation are typically discarded, preventing the accumulation of lifelong experience. To this end, we propose Trajectory RAG (TrajRAG), a retrieval-augmented generation framework that enhances large-model reasoning by retrieving geometric-semantic experiences. TrajRAG incrementally accumulates episodic observations from past navigation episodes. To structure these observations, we propose a topological-polar (topo-polar) trajectory representation that compactly encodes spatial layouts and semantic contexts, effectively removing redundancies in raw episodic observations. A hierarchical chunking structure further organizes similar topo-polar trajectories into unified summaries, enabling coarse-to-fine retrieval. During navigation, candidate frontiers generate multiple trajectory hypotheses that query TrajRAG for similar past trajectories, guiding large-model reasoning for waypoint selection. New experiences are continually consolidated into TrajRAG, enabling the accumulation of lifelong navigation experience. Experiments on MP3D, HM3D-v1, and HM3D-v2 show that TrajRAG effectively retrieves relevant geometric-semantic experiences and improves zero-shot ObjectNav performance.

cs.CV↗

arXiv2026

Multi-Scale Gaussian-Language Map for Zero-shot Embodied Navigation and Reasoning

Understanding the geometric and semantic structure of environments is essential for embodied navigation and reasoning. Existing semantic mapping methods trade off between explicit geometry and multi-scale semantics, and lack a native interface for large models, thus requiring additional training of feature projection for semantic alignment. To this end, we propose the multi-scale Gaussian-Language Map (GLMap), which introduces three key designs: (1) explicit geometry, (2) multi-scale semantics covering both instance and region concepts, and (3) a dual-modality interface where each semantic unit jointly stores a natural language description and a 3D Gaussian representation. The 3D Gaussians enable compact storage and fast rendering of task-relevant images via Gaussian splatting. To enable efficient incremental construction, we further propose a Gaussian Estimator that analytically derives Gaussian parameters from dense point clouds without gradient-based optimization. Experiments on ObjectNav, InstNav, and SQA tasks show that GLMap effectively enhances target navigation and contextual reasoning, while remaining compatible with large-model-based methods in a zero-shot manner. The code is available at https://github.com/sx-zhang/GLMap.

cs.CV↗

arXiv2025

The Brown-dwarf Desert Persists as a Mass-ratio Desert around Low-mass Stars

Sun-like stars are known to host a paucity of brown dwarf companions at close separations. Direct imaging surveys of intermediate-mass stars have suggested that the brown dwarf desert may be fundamentally a feature in the mass ratio. Microlensing surveys provide a unique opportunity to investigate the nature of this desert around low mass stars, as microlensing hosts have typical masses of 0.05-0.8 $M_\odot$. Here, we perform a statistical analysis of homogeneously selected binary-lens microlensing events in the literature, and identify a companion mass-ratio desert at $0.02\lesssim q\lesssim0.05$ and projected separations around 1-5 au. We derive a statistically significant truncation to the giant-planet mass-ratio distribution at $q\simeq0.02$, above which the occurrence rate density drops by approximately an order of magnitude. Due to the possibility that the small number of detected companions in this desert orbit white dwarfs, this mass-ratio desert may be closer to being completely dry around main-sequence stars. Below this desert, we derive a giant-planet ($q>4\times10^{-4}$) occurrence rate density of $5.8\pm0.8$% per decade of projected separation. Our analysis furthers the hypothesis that the brown dwarf desert is fundamentally a feature in the mass ratio, separating distinct populations of planetary and non-planetary companions that are likely formed via core accretion and gravitational instability, respectively. The persistence of this desert across stellar types suggests that both gas-giant planets and sub-stellar companions form in a scale-invariant fashion, with planets growing to a maximum of approximately 2% their host masses.

astro-ph.EP↗

arXiv2025

Predictions of the Nancy Grace Roman Space Telescope Galactic Exoplanet Survey. IV. Lens Mass and Distance Measurements

As part of the Galactic Bulge Time Domain Survey (GBTDS), the Nancy Grace Roman Galactic Exoplanet Survey (RGES) will use microlensing to discover cold outer planets and free-floating planets unbound to stars. NASA has established several science requirements for the GBTDS to ensure RGES success. A key advantage of RGES is Roman's high angular resolution, which will allow detection of flux from many host stars. One requirement specifies that Roman must measure the masses and distances of 40% of detected planet hosts with 20% precision or better. To test this, we simulated microlensing events toward the GBTDS fields and used Fisher matrix analysis to estimate light curve parameter uncertainties. Combining these with Roman imaging observables (lens flux, relative lens-source proper motion), we estimated the achievable precision of lens mass and distance measurements. Using pyLIMASS, a publicly available code for estimating lens properties, we applied this analysis to 3,000 simulated events. Assuming the Cassan et al. (2012) exoplanet mass function, we find that >40% of host stars meet the required 20% precision threshold, confirming that the GBTDS can satisfy the mission requirement. We validated our approach by comparing our inferred lens masses and distances to empirical measurements from detailed image-constrained light curve modeling of historical microlensing events with Hubble and Keck follow-up imaging. Our results agree within roughly 1 sigma, demonstrating that both approaches yield consistent and reliable mass and distance estimates, and confirming the robustness of our simulations for Roman-era microlensing science.

astro-ph.EP↗

arXiv2025

KMT-2018-BLG-0029Lb and OGLE-2019-BLG-0960Lb: Mass Measurements for Two Super-Earth Microlensing Planets

KMT-2018-BLG-0029Lb and OGLE-2019-BLG-0960Lb were the lowest mass-ratio microlensing planets at the time of discovery. For both events, microlensing parallax measurements from the Spitzer Space Telescope implied lens systems that were more distant and massive than those inferred from the ground-based parallax. Here, we report on the detection of excess flux aligned to the event locations using Keck Adaptive Optics imaging, which is consistent with the expected brightness of main-sequence hosts under the ground-based parallax, but inconsistent with that predicted by Spitzer. Based on the excess flux, ground-based parallax, and angular Einstein radius, we determine KMT-2018-BLG-0029Lb to be a $4.2\pm0.5 M_\oplus$ planet orbiting a $0.70\pm0.07 M_\odot$ host at a projected separation of $3.1\pm0.3$ au, and OGLE-2019-BLG-0960Lb to be a $2.0\pm0.2 M_\oplus$ planet orbiting a $0.40\pm0.03 M_\odot$ host at a projected separation of $1.7\pm0.1$ au. We report on additional light-curve models for KMT-2018-BLG-0029 under the generalized inner-outer (offset) degeneracy, which were not reported in the original analysis. We point out inconsistencies in the inner/outer labeling of the degenerate models in the lens and source planes, and advocate for the lens-plane convention, which refers to the planet being closer or further to the host star compared to the image it perturbs. Lastly, we discuss the possibility of breaking this degeneracy via ground concurrent observations with the Roman Space Telescope.

astro-ph.EP↗

arXiv2024

An Earth-Mass Planet and a Brown Dwarf in Orbit Around a White Dwarf

Terrestrial planets born beyond 1-3 AU have been theorized to avoid being engulfed during the red-giant phases of their host stars. Nevertheless, only a few gas-giant planets have been observed around white dwarfs (WDs) -- the end product left behind by a red giant. Here we report on evidence that the lens system that produced the microlensing event KMT-2020-BLG-0414 is composed of a WD orbited by an Earth-mass planet and a brown dwarf (BD) companion, as shown by the non-detection of the lens flux using Keck Adaptive Optics (AO). From microlensing orbital motion constraints, we determine the planet to be a $1.9\pm0.2$ Earth-mass ($M_\oplus$) planet at a physical separation of $2.1\pm0.2$ au from the WD during the event. By considering the system evolutionary history, we determine the BD companion to have a projected separation of 22 au from the WD, and reject an alternative model that places the BD at 0.2 au. Given planetary orbital expansion during the final evolutionary stages of the host star, this Earth-mass planet may have existed in an initial orbit close to 1 au, thereby offering a glimpse into the possible survival of planet Earth in the distant future.

astro-ph.EP↗

arXiv2024

An Analysis of the Recovery Path of the Consumer Sector in the Post-Pandemic Era

This paper proposes a referencable pattern of the recovery of the consumption sector, a new dimension to observe and evaluate the intrinsic value of the consumption sector, and proposes the concept of sensory-based consumption and the ranking of the weights of different categories;creates the concept of digital consumption index, coupled with digital RMB index and China-style digital economy index. Finally we explain the internal logic of digital consumption as a consumption upgrade tool and a higher valuation target in the context of China's economic performance in 2022 and the Chinese government's policy in 2023, leading to the investment strategy of roller conduction effect.

econ.GN↗

arXiv2023

Stellar Spectra Fitting with Amortized Neural Posterior Estimation and nbi

Modern surveys often deliver hundreds of thousands of stellar spectra at once, which are fit to spectral models to derive stellar parameters/labels. Therefore, the technique of Amortized Neural Posterior Estimation (ANPE) stands out as a suitable approach, which enables the inference of large number of targets as sub-linear/constant computational costs. Leveraging our new nbi software package, we train an ANPE model for the APOGEE survey and demonstrate its efficacy on both mock and real APOGEE stellar spectra. Unique to the nbi package is its out-of-the-box functionality on astronomical inverse problems with sequential data. As such, we have been able to acquire the trained model with minimal effort. We introduce an effective approach to handling the measurement noise properties inherent in spectral data, which utilizes the actual uncertainties in the observed data. This allows training data to resemble observed data, an aspect that is crucial for ANPE applications. Given the association of spectral data properties with the observing instrument, we discuss the utility of an ANPE "model zoo," where models are trained for specific instruments and distributed under the nbi framework to facilitate real-time stellar parameter inference.

astro-ph.SR↗

arXiv2023

nbi: the Astronomer's Package for Neural Posterior Estimation

Despite the promise of Neural Posterior Estimation (NPE) methods in astronomy, the adaptation of NPE into the routine inference workflow has been slow. We identify three critical issues: the need for custom featurizer networks tailored to the observed data, the inference inexactness, and the under-specification of physical forward models. To address the first two issues, we introduce a new framework and open-source software nbi (Neural Bayesian Inference), which supports both amortized and sequential NPE. First, nbi provides built-in "featurizer" networks with demonstrated efficacy on sequential data, such as light curve and spectra, thus obviating the need for this customization on the user end. Second, we introduce a modified algorithm SNPE-IS, which facilities asymptotically exact inference by using the surrogate posterior under NPE only as a proposal distribution for importance sampling. These features allow nbi to be applied off-the-shelf to astronomical inference problems involving light curves and spectra. We discuss how nbi may serve as an effective alternative to existing methods such as Nested Sampling. Our package is at https://github.com/kmzzhang/nbi.

astro-ph.IM↗

arXiv2023

A Shock Flash Breaking Out of a Dusty Red Supergiant

Shock breakout emission is light that arises when a shockwave, generated by core-collapse explosion of a massive star, passes through its outer envelope. Hitherto, the earliest detection of such a signal was at several hours after the explosion, though a few others had been reported. The temporal evolution of early light curves should reveal insights into the shock propagation, including explosion asymmetry and environment in the vicinity, but this has been hampered by the lack of multiwavelength observations. Here we report the instant multiband observations of a type II supernova (SN 2023ixf) in the galaxy M101 (at a distance of 6.85+/-0.15 Mpc), beginning at about 1.4 hours after the explosion. The exploding star was a red supergiant with a radius of about 440 solar radii. The light curves evolved rapidly, on timescales of 1-2 hours, and appeared unusually fainter and redder than predicted by models within the first few hours, which we attribute to an optically thick dust shell before it was disrupted by the shockwave. We infer that the breakout and perhaps the distribution of the surrounding dust were not spherically symmetric.

astro-ph.HE↗

arXiv2023

The Progenitor Star of SN 2023ixf: A Massive Red Supergiant with Enhanced, Episodic Pre-Supernova Mass Loss

We identify the progenitor star of SN 2023ixf in the nearby galaxy Messier 101 using Keck/NIRC2 adaptive optics imaging and pre-explosion HST/ACS images. The supernova position, localized with diffraction-spike pattern and high precision relative astrometry, unambiguously coincides with a single progenitor candidate of m_F814W=24.96(-0.04)(+0.05). Forced photometry further recovers 2-sigma detections in the F673N and F675W bands and imposes robust flux limits on the bluer bands. Given the reported infrared excess and semi-regular variability of the progenitor, we fit a time-dependent spectral energy distribution (SED) model of a dusty red supergiant (RSG) to a combined dataset of HST photometry, as well as ground-based near-infrared and Spitzer/IRAC [3.6], [4.5] photometry from the literature. The progenitor closely resembles a RSG of T_eff=3343+/-27 K and logL=5.10+/-0.02, with a 0.11+/-0.01 dex (25.2+/-1.7 per cent) variation over the mean luminosity at a period of P=1128.3+/-6.5 days, heavily obscured by a dust envelope with an optical depth of tau=2.83+/-0.03 at 1 micron (or A_V=10.28+/-0.11 mag). Such observed signatures match a post-main sequence star of 18.1(-1.2)(+0.7) Msun, close to the most massive SN II progenitor, with a pulsation-enhanced mass-loss rate of M_dot=(3.58+/-0.15) x 10^(-4) Msun/yr. The dense and confined circumstellar material is likely ejected during the last episode of radial pulsation before the explosion. Notably, we find strong evidence for periodic variation of tau (or both T_eff and tau) along with luminosity, a necessary assumption to reproduce the wavelength dependence of the variability, which implies dust sublimation and condensation during radial pulsations. Given the observed SED, partial dust obscuration remains a possible scenario, but any unobstructed binary companion over 7.1 Msun can be ruled out.

astro-ph.SR↗

arXiv2022

On the Perturbative Picture and the Chang-Refsdal Lens Approximation for Planetary Microlensing

Under the perturbative picture of planetary microlensing, the planet is considered to act as a uniform-shear Chang-Refsdal lens on one of the two images produced by the host star that comes close to the angular Einstein radius of the planet, leaving the other image unaffected. However, this uniform-shear approximation is only valid for isolated planetary caustics and breaks down in the resonant regime. Recently, the planetary-caustic degeneracy arising from the above formalism is found to generalize to the regime of central and resonant caustics, indicating that the perturbative picture and Chang-Refsdal lens approximation may have been under-explored in the past. Here, I introduce a new variable-shear Chang-Refsdal lens approximation, which not only supports central and resonant caustics, but also enables full magnification maps to be calculated analytically. Moreover, I introduce the generalized perturbative picture, which relaxes the required proximity between the planet and the image being perturbed in the previous work. Specifically, the planet always perturbs the image in the same half of the lens plane as the planet itself, leaving the other image largely unaffected. It is demonstrated how this new framework results in the offset degeneracy as a consequence of physical symmetry. The generalized perturbative picture also points to an approach to solve the two-body lens equation semi-analytically. The analytic and semi-analytic microlensing solutions associated with this work may allow for substantially faster light-curve calculations and modeling of observed events. A python implementation is provided.

astro-ph.EP↗

arXiv2022

A Mathematical Treatment of the Offset Microlensing Degeneracy

The offset microlensing degeneracy, recently proposed by Zhang et al. (2022), has been shown to generalize the close-wide and inner-outer caustic degeneracies into a unified regime of magnification degeneracy in the interpretation of 2-body planetary microlensing observations. While the inner-outer degeneracy expects the source trajectory to pass equidistant to the planetary caustics of the degenerate lens configurations, the offset degeneracy states that the same mathematical expression applies to any combination of the close, wide, and resonant caustic topologies, where the projected star-planet separations differ by an offset ($s_{\rm A}\neq s_{\rm B}$) that depends on where the source trajectory crosses the star-planet axis. An important implication is that the $s_{\rm A}=1/s_{\rm B}$ solution of the close-wide degeneracy never strictly manifests in observations except when the source crosses a singular point near the primary. Nevertheless, the offset degeneracy was proposed upon numerical calculations, and no theoretical justification was given. Here, we provide a theoretical treatment of the offset degeneracy, which demonstrates its nature as a mathematical degeneracy. From first principles, we show that the offset degeneracy formalism is exact to zeroth-order in the mass ratio ($q$) for two cases: when the source crosses the lens-axis inside of caustics, and for $(s_{\rm A}-s_{\rm B})^6\ll1$ when crossing outside of caustics. The extent to which the offset degeneracy persists in oblique source trajectories is explored numerically. Lastly, it is shown that the superposition principle allows for a straightforward generalization to $N$-body microlenses with $N-1$ planetary lens components ($q\ll1$), which results in a $2^{N-1}$-fold degeneracy.

astro-ph.EP↗

arXiv2022

Portfolio selection models based on interval-valued conditional value at risk (ICVaR) and empirical analysis

Risk management is very important for individual investors or companies. There are many ways to measure the risk of investment. Prices of risky assets vary rapidly and randomly due to the complexity of finance market. Random interval is a good tool to describe uncertainty with both randomness and imprecision. Considering the uncertainty of financial market, we employ random intervals to describe the returns of a risk asset and consider the tail risk, which is called the interval-valued Conditional Value at Risk (ICVaR, for short). Such an ICVaR is a risk measure and satisfies subadditivity. Under the new risk measure ICVaR, as a manner similar to the classical portfolio model of Markowitz, optimal interval-valued portfolio selection models are built. Based on the real data from mainland Chinese stock market, the case study shows that our models are interpretable and consistent with the practical scenarios.

q-fin.PM↗

arXiv2021

A Ubiquitous Unifying Degeneracy in Two-Body Microlensing Systems

While gravitational microlensing by planetary systems provides unique vistas on the properties of exoplanets, observations of a given 2-body microlensing event can often be interpreted with multiple distinct physical configurations. Such ambiguities are typically attributed to the close-wide and inner-outer types of degeneracies that arise from transformation invariances and symmetries of microlensing caustics. However, there remain unexplained inconsistencies between aforementioned theories and observations. Here, leveraging a fast machine learning inference framework, we present the discovery of the offset degeneracy, which concerns a magnification-matching behaviour on the lens-axis and is formulated independent of caustics. This offset degeneracy unifies the close-wide and inner-outer degeneracies, generalises to resonant topologies, and upon reanalysis, not only appears ubiquitous in previously published planetary events with 2-fold degenerate solutions, but also resolves prior inconsistencies. Our analysis demonstrates that degenerate caustics do not strictly result in degenerate magnifications and that the commonly invoked close-wide degeneracy essentially never arises in actual events. Moreover, it is shown that parameters in offset degenerate configurations are related by a simple expression. This suggests the existence of a deeper symmetry in the equations governing 2-body lenses than previously recognised.

astro-ph.EP↗

arXiv2021

Real-Time Likelihood-Free Inference of Roman Binary Microlensing Events with Amortized Neural Posterior Estimation

Fast and automated inference of binary-lens, single-source (2L1S) microlensing events with sampling-based Bayesian algorithms (e.g., Markov Chain Monte Carlo; MCMC) is challenged on two fronts: high computational cost of likelihood evaluations with microlensing simulation codes, and a pathological parameter space where the negative-log-likelihood surface can contain a multitude of local minima that are narrow and deep. Analysis of 2L1S events usually involves grid searches over some parameters to locate approximate solutions as a prerequisite to posterior sampling, an expensive process that often requires human-in-the-loop domain expertise. As the next-generation, space-based microlensing survey with the Roman Space Telescope is expected to yield thousands of binary microlensing events, a new fast and automated method is desirable. Here, we present a likelihood-free inference (LFI) approach named amortized neural posterior estimation, where a neural density estimator (NDE) learns a surrogate posterior $\hat{p}(\theta|x)$ as an observation-parametrized conditional probability distribution, from pre-computed simulations over the full prior space. Trained on 291,012 simulated Roman-like 2L1S simulations, the NDE produces accurate and precise posteriors within seconds for any observation within the prior support without requiring a domain expert in the loop, thus allowing for real-time and automated inference. We show that the NDE also captures expected posterior degeneracies. The NDE posterior could then be refined into the exact posterior with a downstream MCMC sampler with minimal burn-in steps.

astro-ph.IM↗

arXiv2020

Classification of Periodic Variable Stars with Novel Cyclic-Permutation Invariant Neural Networks

Neural networks (NNs) have been shown to be competitive against state-of-the-art feature engineering and random forest (RF) classification of periodic variable stars. Although previous work utilising NNs has made use of periodicity by period folding multiple-cycle time-series into a single cycle -- from time-space to phase-space -- no approach to date has taken advantage of the fact that network predictions should be invariant to the initial phase of the period-folded sequence. Initial phase is exogenous to the physical origin of the variability and should thus be factored out. Here, we present cyclic-permutation invariant networks, a novel class of NNs for which invariance to phase shifts is guaranteed through polar coordinate convolutions, which we implement by means of "Symmetry Padding." Across three different datasets of variable star light curves, we show that two implementations of the cyclic-permutation invariant network: the iTCN and the iResNet, consistently outperform non-invariant baselines and reduce overall error rates by between 4% to 22%. Over a 10-class OGLE-III sample, the iTCN/iResNet achieves an average per-class accuracy of 93.4%/93.3%, compared to RNN/RF accuracies of 70.5%/89.5% in a recent study using the same data. Finding improvement on a non-astronomy benchmark, we suggest that the methodology introduced here should also be applicable to a wide range of science domains where periodic data abounds due to physical symmetries.

astro-ph.IM↗