arXiv subjects

Yuan Mei

Publications and source records attributed to Yuan Mei.

17 recordsLinked to original sources

arXiv2026

Deterministic Decomposition of Stochastic Generative Dynamics

Modern generative models can be understood as probability transport from a simple base distribution to a target data distribution. Deterministic transport models offer tractable velocity-field parameterizations, whereas stochastic generative models capture richer density evolution through drift and diffusion. Yet when stochastic dynamics are described through deterministic velocity fields, the effects of drift and diffusion are often compressed into a single effective field, obscuring the distinct roles of deterministic evolution and stochastic fluctuation. In this work, we show that the deterministic field $b_t$ of a stochastic generative process admits a natural transport--osmotic decomposition that separates deterministic transport from stochastic, diffusion-induced effects: $b_t = u_t + d_t$, where $u_t$ governs marginal probability transport and $d_t$ captures an osmotic effect induced by diffusion and determined by the marginal score. Based on this decomposition, we propose Bridge Matching, a flow-based framework for learning decomposed generative dynamics through both marginal and conditional formulations. In generative modeling experiments, we recombine the learned components as $b_t = u_t + \lambda_d d_t$, showing that the proposed decomposition enables interpretable and controllable sampling by adjusting the osmotic contribution in probability transport.

cs.LG↗

arXiv2026

M$^3$: Reframing Training Measures for Discretized Physical Simulations

Neural surrogate models for physical simulations are trained on discretized samples of continuous domains, where the induced empirical measure leads to uneven supervision, biasing optimization and causing spatial inconsistencies in physical fidelity. To mitigate this measure-induced bias, we propose M$^3$ (Multi-scale Morton Measure), a scalable framework that balances training measures by partitioning space according to physical variation and allocating supervision across multiple scales. Applied to three industrial-scale datasets with diverse discretizations, M$^3$ consistently improves predictions in the continuous physical domain, achieving up to 4.7$\times$ lower error in large-scale volumetric cases. These gains persist under aggressive subsampling (160M $\rightarrow$ 16M $\rightarrow$ 1.6M points), where M$^3$-trained models outperform those trained on higher-resolution data, reducing physics-weighted relative $L_2$ error by 3--4$\times$ and the corresponding MSE by up to 13$\times$. These results highlight data distribution as a key factor in operator learning and position M$^3$ as a scalable, data-efficient approach for physically consistent modeling. Code is available at https://github.com/PhysDataRefine/M3.

cs.AI↗

arXiv2026

UniStitch: Unifying Semantic and Geometric Features for Image Stitching

Traditional image stitching methods estimate warps from hand-crafted geometric features, whereas recent learning-based solutions leverage semantic features from neural networks instead. These two lines of research have largely diverged along separate evolution, with virtually no meaningful convergence to date. In this paper, we take a pioneering step to bridge this gap by unifying semantic and geometric features with UniStitch, a unified image stitching framework from multimodal features. To align discrete geometric features (i.e., keypoint) with continuous semantic feature maps, we present a Neural Point Transformer (NPT) module, which transforms unordered, sparse 1D geometric keypoints into ordered, dense 2D semantic maps. Then, to integrate the advantages of both representations, an Adaptive Mixture of Experts (AMoE) module is designed to fuse geometric and semantic representations. It dynamically shifts focus toward more reliable features during the fusion process, allowing the model to handle complex scenes, especially when either modality might be compromised. The fused representation can be adopted into common deep stitching pipelines, delivering significant performance gains over any single feature. Experiments show that UniStitch outperforms existing state-of-the-art methods with a large margin, paving the way for a unified paradigm between traditional and learning-based image stitching.

cs.CV↗

arXiv2025

Robust Image Stitching with Optimal Plane

We present \textit{RopStitch}, an unsupervised deep image stitching framework with both robustness and naturalness. To ensure the robustness of \textit{RopStitch}, we propose to incorporate the universal prior of content perception into the image stitching model by a dual-branch architecture. It separately captures coarse and fine features and integrates them to achieve highly generalizable performance across diverse unseen real-world scenes. Concretely, the dual-branch model consists of a pretrained branch to capture semantically invariant representations and a learnable branch to extract fine-grained discriminative features, which are then merged into a whole by a controllable factor at the correlation level. Besides, considering that content alignment and structural preservation are often contradictory to each other, we propose a concept of virtual optimal planes to relieve this conflict. To this end, we model this problem as a process of estimating homography decomposition coefficients, and design an iterative coefficient predictor and minimal semantic distortion constraint to identify the optimal plane. This scheme is finally incorporated into \textit{RopStitch} by warping both views onto the optimal plane bidirectionally. Extensive experiments across various datasets demonstrate that \textit{RopStitch} significantly outperforms existing methods, particularly in scene robustness and content naturalness. The code is available at {\color{red}https://github.com/MmelodYy/RopStitch}.

cs.CV↗

arXiv2024

First operation of a multi-channel Q-Pix prototype: measuring transverse electron diffusion in a gas time projection chamber

We report measurements of the transverse diffusion of electrons in P-10 gas (90% Ar, 10% CH4) in a laboratory-scale time projection chamber (TPC) utilizing a novel pixelated signal capture and digitization technique known as Q-Pix. The Q-Pix method incorporates a precision switched integrating transimpedance amplifier whose output is compared to a threshold voltage. Upon reaching the threshold, a comparator sends a 'reset' signal, initiating a discharge of the integrating capacitor. The time difference between successive resets is inversely proportional to the average current at the pixel in that time interval, and the number of resets is directly proportional to the total collected charge. We developed a 16-channel Q-Pix prototype fabricated from commercial off-the-shelf components and coupled them to 16 concentric annular anode electrodes to measure the spatial extent of the electron swarm that reaches the anode after drifting through the uniform field of the TPC. The swarm is produced at a gold photocathode using pulsed UV light. The measured transverse diffusion agrees with simulations in PyBoltz across a range of operating pressures (200-1500 Torr). These results demonstrate that a Q-Pix readout can successfully reconstruct the ionization topology in a TPC.

hep-ex↗

arXiv2023

Demonstrating the Q-Pix front-end using discrete OpAmp and CMOS transistors

Using Commercial Off-The-Shelf (COTS) Operational Amplifiers (OpAmps) and Complementary Metal-Oxide Semiconductor (CMOS) transistors, we present a demonstration of the Q-Pix front-end architecture, a novel readout solution for kiloton-scale Liquid Argon Time Projection Chamber (LArTPC) detectors. The Q-Pix scheme employs a Charge-Integrate/Reset process based on the Least Action principle, enabling pixel-scale self-triggering charge collection and processing, minimizing energy consumption, and maximizing data compression. We examine the architecture's sensitivity, linearity, noise, and other features at the circuit board level and draw comparisons to SPICE simulations. Furthermore, we highlight the resemblance between the Q-Pix front-end and Sigma-Delta modulator, emphasizing that digital data processing techniques for Sigma-Delta can be directly applied to Q-Pix, resulting in enhanced signal-to-noise performance. These insights will inform the development of Q-Pix front-end designs in integrated circuits (IC) and guide data collection and processing for future large-scale LArTPC detectors in neutrino physics and other high-energy physics experiments.

physics.ins-det↗

arXiv2022

Monolithic Active Pixel Sensors on CMOS technologies

Collider detectors have taken advantage of the resolution and accuracy of silicon detectors for at least four decades. Future colliders will need large areas of silicon sensors for low mass trackers and sampling calorimetry. Monolithic Active Pixel Sensors (MAPS), in which Si diodes and readout circuitry are combined in the same pixels, and can be fabricated in some of standard CMOS processes, are a promising technology for high-granularity and light detectors. In this paper we review 1) the requirements on MAPS for trackers and electromagnetic calorimeters (ECal) at future colliders experiments, 2) the ongoing efforts towards dedicated MAPS for the Electron-Ion Collider (EIC) at BNL, for which the EIC Silicon Consortium was already instantiated, and 3) space-born applications for MeV $γ$-ray experiments with MAPS based trackers (AstroPix).

physics.ins-det↗

arXiv2022

Topmetal-M: a novel pixel sensor for compact tracking applications

The Topmetal-M is a large area pixel sensor (18 mm * 23 mm) prototype fabricated in a new 130 nm high-resistivity CMOS process in 2019. It contains 400 rows * 512 columns square pixels with the pitch of 40 μm. In Topmetal-M, a novel charge collection method combing the Monolithic Active Pixel Sensor (MAPS) and the Topmetal sensor has been proposed for the first time. Both the ionized charge deposited by the particle in the sensor and along the track over the sensor can be collected. The in-pixel circuit mainly consists of a low-noise charge sensitive amplifier to establish the signal for the energy reconstruction, and a discriminator with a Time-to-Amplitude Converter (TAC) for the Time of Arrival (TOA) measurement. With this mechanism, the trajectory, particle hit position, energy and arrival time of the particle can be measured. The analog signal from each pixel is accessible through time-shared multiplexing over the entire pixel array. This paper will discuss the design and preliminary test results of the Topmetal-M sensor.

physics.ins-det↗

arXiv2021

A direct-sampling RF receiver for MOLLER beam charge measurement

We have developed and tested a direct-sampling RF receiver capable of measuring the amplitude of a 1497 MHz sinusoidal signal in 0.5 ms integration windows to within <10 ppm relative uncertainty. The receiver is intended for measuring signals from beam current monitoring cavities on the beamline of the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Laboratory. The signal strength, frequency, and integration window are consistent with the thus far unmet requirements of the upcoming MOLLER experiment to measure the beam charge for different helicity states.

physics.ins-det↗

arXiv2020

PFL-MoE: Personalized Federated Learning Based on Mixture of Experts

Federated learning (FL) is an emerging distributed machine learning paradigm that avoids data sharing among training nodes so as to protect data privacy. Under coordination of the FL server, each client conducts model training using its own computing resource and private data set. The global model can be created by aggregating the training results of clients. To cope with highly non-IID data distributions, personalized federated learning (PFL) has been proposed to improve overall performance by allowing each client to learn a personalized model. However, one major drawback of a personalized model is the loss of generalization. To achieve model personalization while maintaining generalization, in this paper, we propose a new approach, named PFL-MoE, which mixes outputs of the personalized model and global model via the MoE architecture. PFL-MoE is a generic approach and can be instantiated by integrating existing PFL algorithms. Particularly, we propose the PFL-MF algorithm which is an instance of PFL-MoE based on the freeze-base PFL algorithm. We further improve PFL-MF by enhancing the decision-making ability of MoE gating network and propose a variant algorithm PFL-MFE. We demonstrate the effectiveness of PFL-MoE by training the LeNet-5 and VGG-16 models on the Fashion-MNIST and CIFAR-10 datasets with non-IID partitions.

cs.LG↗

arXiv2020

Topmetal CMOS direct charge sensing plane for neutrinoless double-beta decay search in high-pressure gaseous TPC

We propose a novel charge sensing concept for high-pressure Time Projection Chamber (TPC) to search for Neutrinoless Double-Beta Decay (NLDBD) with ton-scale isotope mass and beyond. A meter-sized plane, tiled with an array of CMOS integrated sensors called Topmetal that directly collect charge without gas avalanche gain, is to be deployed into a high-pressure gaseous TPC with working gases containing suitable NLDBD candidate isotopes such as Xe-136 and Se-82. The Topmetal sensor has an electronic noise <30 e- per pixel, which allows the detector to reach <1% FWHM energy resolution at the NLDBD Q-value for both Xe-136 and 82SeF6 gases by measuring ionization charges alone. The elimination of charge avalanche gain allows the direct sensing of slow-drifting ions, which enables the use of highly electronegative gas SeF6 in which free electrons do not exist. It supports the swapping of working gases without hardware modification, which is a unique way to validate signals against radioactive backgrounds. Since the sensor manufacturing and plane assembling could leverage unaltered industrial mass-production processes, stability, uniformity, scalability, and cost-effectiveness that are required for ton-scale experiments could all be reached. The strengths of TPC such as 3D ionization tracking and decay daughter tagging are retained. This development could lead to a competitive NLDBD experiment at and above ton-scale. The conceptual considerations, simulations, and initial prototyping are discussed.

physics.ins-det↗

arXiv2020

Cryogenic Characterization of 180 nm CMOS Technology at 100 mK

Conventional CMOS technology operated at cryogenic conditions has recently attracted interest for its uses in low-noise electronics. We present one of the first characterizations of 180 nm CMOS technology at a temperature of 100 mK, extracting I/V characteristics, threshold voltages, and transconductance values, as well as observing their temperature dependence. We find that CMOS devices remain fully operational down to these temperatures, although we observe hysteresis effects in some devices. The measurements described in this paper can be used to inform the future design of CMOS devices intended to be operated in this deep cryogenic regime.

physics.ins-det↗

arXiv2018

Q-Pix: Pixel-scale Signal Capture for Kiloton Liquid Argon TPC Detectors: Time-to-Charge Waveform Capture, Local Clocks, Dynamic Networks

We describe a novel ionization signal capture and waveform digitization scheme for kiloton-scale liquid argon Time Projection Chamber (TPC) detectors. The scheme is based on a pixel-scale self-triggering `charge integrate/reset' block, local clocks running at unconstrained frequencies and dynamically established data networks. The scheme facilitates detailed capture of waveforms of arbitrary complexity from a sequence of varying time intervals, each of which corresponds to a fixed charge integral. An absolute charge auto-calibration process based on intrinsic $^{39}$Ar decay current is a major benefit. A flat electronic architecture with self-guided network generation provides exceptionally high resilience against single-point failure. Integrated photon detection, although highly speculative, may also be possible. The goal is optimized discovery potential. Much might be at stake.

physics.ins-det↗

arXiv2016

PandaX-III: Searching for Neutrinoless Double Beta Decay with High Pressure $^{136}$Xe Gas Time Projection Chambers

Searching for the Neutrinoless Double Beta Decay (NLDBD) is now regarded as the topmost promising technique to explore the nature of neutrinos after the discovery of neutrino masses in oscillation experiments. PandaX-III (Particle And Astrophysical Xenon Experiment III) will search for the NLDBD of $^{136}$Xe at the China Jin Ping underground Laboratory (CJPL). In the first phase of the experiment, a high pressure gas Time Projection Chamber (TPC) will contain 200 kg, 90% $^{136}$Xe enriched gas operated at 10 bar. Fine pitch micro-pattern gas detector (Microbulk Micromegas) will be used at both ends of the TPC for the charge readout with a cathode in the middle. Charge signals can be used to reconstruct tracks of NLDBD events and provide good energy and spatial resolution. The detector will be immersed in a large water tank to ensure $\sim$5 m of water shielding in all directions. The second phase, a ton-scale experiment, will consist of five TPCs in the same water tank, with improved energy resolution and better control over backgrounds.

physics.ins-det↗

arXiv2016

Detailed study of the column-based priority logic readout of Topmetal-II- CMOS pixel direct charge sensor

We present the detailed study of the digital readout of Topmetal-II- CMOS pixel direct charge sensor. Topmetal-II- is an integrated sensor with an array of 72X72 pixels each capable of directly collecting external charge through exposed metal electrodes in the topmost metal layer. In addition to the time-shared multiplexing readout of the analog output from Charge Sensitive Amplifiers in each pixel, hits are also generated through comparators with individually DAC settable thresholds in each pixel. The hits are read out via a column-based priority logic structure, retaining both hit location and time information. The in-array column-based priority logic is fully combinational hence there is no clock distributed in the pixel array. Sequential logic and clock are placed on the peripheral of the array. We studied the detailed working behavior and performance of this readout, and demonstrated its potential in imaging applications.

physics.ins-det↗

arXiv2015

A Low-Noise CMOS Pixel Direct Charge Sensor, Topmetal-II-

We report the design and characterization of a CMOS pixel direct charge sensor, Topmetal-II-, fabricated in a standard 0.35um CMOS Integrated Circuit process. The sensor utilizes exposed metal patches on top of each pixel to directly collect charge. Each pixel contains a low-noise charge-sensitive preamplifier to establish the analog signal and a discriminator with tunable threshold to generate hits. The analog signal from each pixel is accessible through time-shared multiplexing over the entire array. Hits are read out digitally through a column-based priority logic structure. Tests show that the sensor achieved a <15e- analog noise and a 200e- minimum threshold for digital readout per pixel. The sensor is capable of detecting both electrons and ions drifting in gas. These characteristics enable its use as the charge readout device in future Time Projection Chambers without gaseous gain mechanism, which has unique advantages in low background and low rate-density experiments.

physics.ins-det↗

arXiv2014

Development of a highly pixelated direct charge sensor, Topmetal-I, for ionizing radiation imaging

Using industrial standard 0.35μm CMOS Integrated Circuit process, we realized a highly pixelated sensor that directly collects charge via metal nodes placed on the top of each pixel and forms two dimensional images of charge cloud distribution. The first version, Topmetal-I, features a 64x64 pixel array of 80μm pitch size. Direct charge calibration reveals an average capacitance of 210fF per pixel. The charge collection noise is near the thermal noise limit. With the readout, individual pixel channels exhibit a most probable equivalent noise charge of 330e-.

physics.ins-det↗