arXiv subjects

Yilun Wu

Publications and source records attributed to Yilun Wu.

At least 19 recordsLinked to original sources

arXiv2026

Graph-Based Financial Fraud Detection with Calibrated Risk Scoring and Structural Regularization

Financial transaction fraud prevention faces challenges such as complex relationship structures, concealed behavioral patterns, and dynamically changing data distribution. Discrimination models relying solely on independent sample features are insufficient to fully characterize the risks of group collaboration and chain transfers within transaction networks. This paper proposes a graph neural network representation learning and risk discrimination framework for financial transaction fraud prevention. It integrates transaction records and identity information into node attributes and constructs a transaction graph based on shared attributes and interaction consistency to explicitly model inter-transaction relationships. In model design, a multi-layer message passing mechanism is employed to aggregate neighborhood information, learn node embedding representations containing structural context semantics, and output transaction-level fraud probability and risk scores through a lightweight risk discrimination head. A weighted supervision objective is introduced to mitigate training bias caused by class imbalance, and structural consistency regularization constraints are combined to suppress the impact of noisy edges on representation drift, thereby improving the stability and usability of risk characterization. Experiments are conducted on a publicly available financial transaction dataset, comparing various methods in the same direction and comprehensively evaluating them under a unified evaluation protocol. The results show that the proposed method outperforms other methods in risk ranking and probability calibration quality, validating the effectiveness of graph structure modeling and representation learning collaboration in financial transaction fraud prevention.

cs.LG↗

arXiv2025

Validating a Machine Learning Approach to Identify Quenched Jets in Heavy-Ion Collisions

Jet quenching is a phenomenon in heavy-ion collisions arising from jet interactions with the quark-gluon plasma (QGP). Its study is complicated by the interplay of multiple physics processes that affect jet observables. In addition, detector effects may influence the results and must be accounted for when identifying quenched jets. We employ a Long Short-Term Memory (LSTM) neural network trained on jet substructure, incorporating parton shower history, to predict jet-by-jet quenching levels. Using photon-jet samples from the \textsc{Jewel} event generator, we show that the LSTM predictions strongly correlate with true jet energy loss. This validates that the model effectively learns the features of jet-QGP interaction. We simulate detector effects using \textsc{Delphes} simulation framework and demonstrate that the method identifies quenching effects in a realistic environment. We test the approach with photon-jet momentum imbalance, jet fragmentation function, and jet shape, which were not included in the training, confirming its ability to distinguish true quenching features.

hep-ph↗

arXiv2025

Adversarial Attacks on VQA-NLE: Exposing and Alleviating Inconsistencies in Visual Question Answering Explanations

Natural language explanations in visual question answering (VQA-NLE) aim to make black-box models more transparent by elucidating their decision-making processes. However, we find that existing VQA-NLE systems can produce inconsistent explanations and reach conclusions without genuinely understanding the underlying context, exposing weaknesses in either their inference pipeline or explanation-generation mechanism. To highlight these vulnerabilities, we not only leverage an existing adversarial strategy to perturb questions but also propose a novel strategy that minimally alters images to induce contradictory or spurious outputs. We further introduce a mitigation method that leverages external knowledge to alleviate these inconsistencies, thereby bolstering model robustness. Extensive evaluations on two standard benchmarks and two widely used VQA-NLE models underscore the effectiveness of our attacks and the potential of knowledge-based defenses, ultimately revealing pressing security and reliability concerns in current VQA-NLE systems.

cs.CV↗

arXiv2025

The Missing Covariate Indicator Method is Nearly Valid Almost Always

Background: Although the missing covariate indicator method (MCIM) has been shown to be biased under extreme conditions, the degree and determinants of bias have not been formally assessed. We derived the formula for the relative bias in the MCIM and systematically investigated conditions under which bias arises. We found that the extent of bias is independent of both the disease rate and the exposure-outcome association, but it is a function of 5 parameters: exposure and covariate prevalences, covariate missingness proportion, and associations of covariate with exposure and outcome. The MCIM was unbiased when the missing covariate is a risk factor for the outcome but not a confounder. The average median relative bias was zero across each of the parameters over a wide range of values considered. Our simulation study demonstrated that the mean and median of relative bias of MCIM was comparable to that of the no missingness method, which used the full sample with complete information for all variables, as long as the missingness of covariate is independent of the outcome. When missingness was no greater than 50%, less than 5% of the scenarios considered had relative bias greater than 10%. In several analyses of the Harvard cohort studies, the MCIM produced materially the same results as the multiple imputation method. In conclusion, the MCIM is nearly valid almost always in settings typically encountered in epidemiology and its continued use is recommended, unless the covariate is missing in an extreme proportion or acts as a strong confounder.

stat.AP↗

arXiv2025

Sustainable cooperation on the hybrid pollution-control game with heterogeneous players

This paper considers a hybrid pollution-control differential game with two farsighted players and one myopic player. Both the seasonal regime shifts in the state dynamics and the players' heterogeneous preferences are introduced into the model. The strategies under cooperative, noncooperative and partially cooperative scenarios are obtained by utilizing the Pontryagin's Maximum Principle. Under all feasible coalition structures, the convergence of the state variable is proved. A new sustainably--cooperative optimality principle is proposed according to the coalition structures, which belongs to the imputation set. The prerequisite for the existence of time-consistency in the sustainably-cooperative optimality principle is explicitly obtained. The seasonal imputation distribution procedure (IDP) is designed to maintain the time-consistentcy (dynamic stability) of cooperation over time.

math.OC↗

arXiv2025

On-Device Self-Supervised Learning of Low-Latency Monocular Depth from Only Events

Event cameras provide low-latency perception for only milliwatts of power. This makes them highly suitable for resource-restricted, agile robots such as small flying drones. Self-supervised learning based on contrast maximization holds great potential for event-based robot vision, as it foregoes the need for high-frequency ground truth and allows for online learning in the robot's operational environment. However, online, on-board learning raises the major challenge of achieving sufficient computational efficiency for real-time learning, while maintaining competitive visual perception performance. In this work, we improve the time and memory efficiency of the contrast maximization pipeline, making on-device learning of low-latency monocular depth possible. We demonstrate that online learning on board a small drone yields more accurate depth estimates and more successful obstacle avoidance behavior compared to only pre-training. Benchmarking experiments show that the proposed pipeline is not only efficient, but also achieves state-of-the-art depth estimation performance among self-supervised approaches. Our work taps into the unused potential of online, on-device robot learning, promising smaller reality gaps and better performance.

cs.RO↗

arXiv2024

Global Bifurcation of Steady Surface Capillary Waves on a $2D$ Droplet

We construct global curves of rotational traveling wave solutions to the $2D$ water wave equations on a compact domain. The real analytic interface is subject to surface tension, while gravitational effects are ignored. In contrast to the rotational surface waves, the fluid flow follows the incompressible, irrotational Euler equations. This model can provide a description for tiny water droplets in breaking waves and white caps. The primary tool we use is global bifurcation theory, via a conformal formulation of the problem. The obtained fluid domains have $m$-fold discrete rotational symmetry, as well as a reflection symmetry.

math.AP↗

arXiv2024

Lightweight Event-based Optical Flow Estimation via Iterative Deblurring

Inspired by frame-based methods, state-of-the-art event-based optical flow networks rely on the explicit construction of correlation volumes, which are expensive to compute and store, rendering them unsuitable for robotic applications with limited compute and energy budget. Moreover, correlation volumes scale poorly with resolution, prohibiting them from estimating high-resolution flow. We observe that the spatiotemporally continuous traces of events provide a natural search direction for seeking pixel correspondences, obviating the need to rely on gradients of explicit correlation volumes as such search directions. We introduce IDNet (Iterative Deblurring Network), a lightweight yet high-performing event-based optical flow network directly estimating flow from event traces without using correlation volumes. We further propose two iterative update schemes: "ID" which iterates over the same batch of events, and "TID" which iterates over time with streaming events in an online fashion. Our top-performing ID model sets a new state of the art on DSEC benchmark. Meanwhile, the base ID model is competitive with prior arts while using 80% fewer parameters, consuming 20x less memory footprint and running 40% faster on the NVidia Jetson Xavier NX. Furthermore, the TID model is even more efficient offering an additional 5x faster inference speed and 8 ms ultra-low latency at the cost of only a 9% performance drop, making it the only model among current literature capable of real-time operation while maintaining decent performance.

cs.CV↗

arXiv2023

Existence of Rotating Magnetic Stars

We consider a star as a compressible fluid subject to gravitational and magnetic forces. This leads to an Euler-Poisson system coupled to a magnetic field, which may be regarded as an MHD model together with gravity. The star executes steadily rotating motion about a fixed axis. We prove, for the first time, the existence of such stars provided that the rotation speed and the magnetic field are sufficiently small.

math.AP↗

arXiv2023

Global Continuation and the Theory of Rotating Stars

This paper gives a condensed review of the history of solutions to the Euler-Poisson equations modeling equilibrium states of rotating stars and galaxies, leading to a recent result of Walter Strauss and the author. This result constructs a connected set of rotating star solutions for larger and larger rotation speed, so that the supports of the stars become unbounded if we assume an equation of state $p = ρ^γ$, $4/3<γ<2$. On the other hand, if $6/5<γ<4/3$, we show that either the supports of the stars become unbounded, or the density somewhere within the stars becomes unbounded. This is the first global continuation result for rotating stars that displays singularity formation within the solution set.

math.AP↗

arXiv2022

Existence of Rotating Stars with Variable Entropy

We model a rotating star as a compressible fluid subject to gravitational forces. In almost all the mathematical literature the entropy is considered to be constant. Here we allow it to be variable. We consider a star that steadily rotates differentially around a fixed axis, say the $z$-axis. We prove the existence of a family of such stars with small angular velocity $ω$ and small entropy variation $s$ and with an equation of state $p=Ke^sρ^γ$. Our analysis reduces to a hyperbolic equation for the modified entropy coupled to an elliptic equation for the modified density, together with a mass constraint. Due to the variable entropy and the consequent loss of both regularity and variational structure, all the methods in the previous literature fail. We develop a new ad hoc perturbative strategy that allows us to construct rotating stars that bifurcate from the non-rotating ones.

math.AP↗

arXiv2022

Global Continuation of a Vlasov Model of Rotating Galaxies

A typical galaxy consists of a huge number of stars attracted to each other by gravity. For instance, the Milky Way has about $10^{11}$ stars. Thus it is typically modeled by the Vlasov-Poisson system. We prove an existence theorem for axisymmetric steady states of galaxies that may rotate rapidly. Such states are given in terms of a fairly general function $ϕ$ of the particle energy and angular momentum. The set $\mathcal K$ of such states form a connected set in an appropriate function space. Along the set $\mathcal K$, we prove under some conditions that either (a) the supports of the galaxies become unbounded or (b) both the rotation speeds and the densities somewhere within the galaxy become unbounded.

math.AP↗

arXiv2020

ALPaCA vs. GP-based Prior Learning: A Comparison between two Bayesian Meta-Learning Algorithms

Meta-learning or few-shot learning, has been successfully applied in a wide range of domains from computer vision to reinforcement learning. Among the many frameworks proposed for meta-learning, bayesian methods are particularly favoured when accurate and calibrated uncertainty estimate is required. In this paper, we investigate the similarities and disparities among two recently published bayesian meta-learning methods: ALPaCA (Harrison et al. [2018]) and PACOH (Rothfuss et al. [2020]). We provide theoretical analysis as well as empirical benchmarks across synthetic and real-world dataset. While ALPaCA holds advantage in computation time by the usage of a linear kernel, general GP-based methods provide much more flexibility and achieves better result across datasets when using a common kernel such as SE (Squared Exponential) kernel. The influence of different loss function choice is also discussed.

cs.LG↗

arXiv2019

The Phoenix Drone: An Open-Source Dual-Rotor Tail-Sitter Platform for Research and Education

In this paper, we introduce the Phoenix drone: the first completely open-source tail-sitter micro aerial vehicle (MAV) platform. The vehicle has a highly versatile, dual-rotor design and is engineered to be low-cost and easily extensible/modifiable. Our open-source release includes all of the design documents, software resources, and simulation tools needed to build and fly a high-performance tail-sitter for research and educational purposes. The drone has been developed for precision flight with a high degree of control authority. Our design methodology included extensive testing and characterization of the aerodynamic properties of the vehicle. The platform incorporates many off-the-shelf components and 3D-printed parts, in order to keep the cost down. Nonetheless, the paper includes results from flight trials which demonstrate that the vehicle is capable of very stable hovering and accurate trajectory tracking. Our hope is that the open-source Phoenix reference design will be useful to both researchers and educators. In particular, the details in this paper and the available open-source materials should enable learners to gain an understanding of aerodynamics, flight control, state estimation, software design, and simulation, while experimenting with a unique aerial robot.

cs.RO↗

arXiv2019

Rapidly Rotating White Dwarfs

A rotating star may be modeled as a continuous system of particles attracted to each other by gravity and with a given total mass and prescribed angular velocity. Mathematically this leads to the Euler-Poisson system. A white dwarf star is modeled by a very particular, and rather delicate, equation of state for the pressure as a function of the density. We prove an existence theorem for rapidly rotating white dwarfs that depend continuously on the speed of rotation. The key tool is global continuation theory, combined with a delicate limiting process. The solutions form a connected set $\mathcal K$ in an appropriate function space. As the speed of rotation increases, we prove that either the supports of white dwarfs in $\mathcal K$ become unbounded or their densities become unbounded. We also discuss the polytropic case with the critical exponent $γ=4/3$.

math.AP↗

arXiv2018

Rapidly Rotating Stars

A rotating star may be modeled as a continuous system of particles attracted to each other by gravity and with a given total mass and prescribed angular velocity. Mathematically this leads to the Euler-Poisson system. We prove an existence theorem for such stars that are rapidly rotating, depending continuously on the speed of rotation. This solves a problem that has been open since Lichtenstein's work in 1933. The key tool is global continuation theory, combined with a delicate limiting process. The solutions form a connected set $\mathcal K$ in an appropriate function space. As the speed of rotation increases, we prove that {\it either the supports of the stars in $\mathcal K$ become unbounded or the density somewhere within the stars becomes unbounded}. We permit any equation of state of the form $p=ρ^γ,\ 6/5<γ<2$, so long as $γ\ne4/3$. We consider two formulations, one where the angular velocity is prescribed and the other where the angular momentum per unit mass is prescribed.

math.AP↗

arXiv2017

Jost Solutions and the Direct Scattering Problem of the Benjamin--Ono Equation

In this paper, we present a rigorous study of the direct scattering problem that arises from the complete integrability of the Benjamin--Ono (BO) equation. In particular, we establish existence, uniqueness, and asymptotic properties of the Jost solutions to the scattering operator in the Fokas--Ablowitz inverse scattering transform (IST). Formulas relating different scattering coefficients are proven, together with their asymptotic behavior with respect to the spectral parameter. This work is an initial step toward the construction of general solutions to the BO equation by IST.

math.AP↗

arXiv2017

Steady States of Rotating Stars and Galaxies

A rotating continuum of particles attracted to each other by gravity may be modeled by the Euler-Poisson system. The existence of solutions is a very classical problem. Here it is proven that a curve of solutions exists, parametrized by the rotation speed, with a fixed mass independent of the speed. The rotation is allowed to vary with the distance to the axis. A special case is when the equation of state is $p=ρ^γ,\ 6/5<γ<2,\ γ\ne4/3$, in contrast to previous variational methods which have required $4/3 < γ$. The continuum of particles may alternatively be modeled microscopically by the Vlasov-Poisson system. The kinetic density is a prescribed function. We prove an analogous theorem asserting the existence of a curve of solutions with constant mass. In this model the whole range $(6/5,2)$ is allowed, including $γ=4/3$.

math.AP↗