Searcharxiv⌕ Search

arXiv subjects

Bin Zeng

Publications and source records attributed to Bin Zeng.

At least 19 recordsLinked to original sources
arXiv2026

Phys-3D: Physics-Constrained Real-Time Crowd Tracking and Counting on Railway Platforms

Accurate, real-time crowd counting on railway platforms is essential for safety and capacity management. We propose to use a single camera mounted in a train, scanning the platform while arriving. While hardware constraints are simple, counting remains challenging due to dense occlusions, camera motion, and perspective distortions during train arrivals. Most existing tracking-by-detection approaches assume static cameras or ignore physical consistency in motion modeling, leading to unreliable counting under dynamic conditions. We propose a physics-constrained tracking framework that unifies detection, appearance, and 3D motion reasoning in a real-time pipeline. Our approach integrates a transfer-learned YOLOv11m detector with EfficientNet-B0 appearance encoding within DeepSORT, while introducing a physics-constrained Kalman model (Phys-3D) that enforces physically plausible 3D motion dynamics through pinhole geometry. To address counting brittleness under occlusions, we implement a virtual counting band with persistence. On our platform benchmark, MOT-RailwayPlatformCrowdHead Dataset(MOT-RPCH), our method reduces counting error to 2.97%, demonstrating robust performance despite motion and occlusions. Our results show that incorporating first-principles geometry and motion priors enables reliable crowd counting in safety-critical transportation scenarios, facilitating effective train scheduling and platform safety management.

cs.CV↗
arXiv2026

Coarse-to-Fine Long-term Interest Modeling for Generative Recommendation

Leveraging long-term user behavioral patterns is a key trajectory for enhancing the accuracy of modern recommender systems. While generative recommender systems have emerged as a transformative paradigm, they face hurdles in effectively modeling extensive historical sequences. To address this challenge, we propose GLASS, a novel framework that integrates long-term user interests into the generative process via SID-Tier and Semantic Search. We first introduce SID-Tier, a module that maps long-term interactions into a unified interest vector to enhance the prediction of the initial SID token. Unlike traditional retrieval models that struggle with massive item spaces, SID-Tier leverages the compact nature of the semantic codebook to incorporate cross features between the user's long-term history and candidate semantic codes. Furthermore, we present semantic hard search, which utilizes generated coarse-grained semantic ID as dynamic keys to extract relevant historical behaviors, which are then fused via an adaptive gated fusion module to recalibrate the trajectory of subsequent fine-grained tokens. To address the inherent data sparsity in semantic hard search, we propose two strategies: semantic neighbor augmentation and codebook resizing. Extensive experiments on two large-scale real-world datasets, TAOBAO-MM and KuaiRec, demonstrate that GLASS outperforms state-of-the-art baselines, achieving significant gains in recommendation quality. Our codes are made publicly available to facilitate further research in generative recommendation.

cs.IR↗
arXiv2025

Did Models Learn Sufficiently? Attribution-Guided Training via Subset-Selected Counterfactual Augmentation

Current visual models often make predictions based on a limited set of discriminative visual cues. As a result, they may become unreliable when the distribution shifts or when these cues are missing. Faithful attribution methods can reveal such problematic reliance through localized explanations, but they are typically used post hoc and are not fed back into the model. To address this limitation, we propose Subset-Selected Counterfactual Augmentation (SS-CA), a training strategy that masks decision-relevant regions to construct counterfactual samples and guide the model toward more robust decision boundaries. Specifically, we extend LIMA, a subset-selection-based faithful attribution method, to Counterfactual LIMA to identify regions whose removal shifts the model toward a competing class. SS-CA then selects near-boundary masks that reduce the logit gap while preserving the original semantics, and applies an adaptive counterfactual filling strategy to replace the masked regions without introducing external semantics. Feeding these counterfactual samples back into training encourages the model to exploit the remaining informative evidence and shifts the decision boundary toward a more robust one. Extensive experiments across five ImageNet variants show that SS-CA effectively improves ID accuracy, OOD generalization, and perturbation robustness, achieving gains of 5.70%/18.04% on ImageNet-1k/ImageNet-R with CLIP ViT/32b, 9.52%/11.33% on ImageNet-R/ImageNet-S on TinyImageNet-200 with ResNet-101, and about 4% under Gaussian Noise corruption. The code will be released soon.

cs.CV↗
arXiv2025

RECAP: Reproducing Copyrighted Data from LLMs Training with an Agentic Pipeline

If we cannot inspect the training data of a large language model (LLM), how can we ever know what it has seen? We believe the most compelling evidence arises when the model itself freely reproduces the target content. As such, we propose RECAP, an agentic pipeline designed to elicit and verify memorized training data from LLM outputs. At the heart of RECAP is a feedback-driven loop, where an initial extraction attempt is evaluated by a secondary language model, which compares the output against a reference passage and identifies discrepancies. These are then translated into minimal correction hints, which are fed back into the target model to guide subsequent generations. In addition, to address alignment-induced refusals, RECAP includes a jailbreaking module that detects and overcomes such barriers. We evaluate RECAP on EchoTrace, a new benchmark spanning over 30 full books, and the results show that RECAP leads to substantial gains over single-iteration approaches. For instance, with GPT-4.1, the average ROUGE-L score for the copyrighted text extraction improved from 0.38 to 0.47 - a nearly 24% increase.

cs.CL↗
arXiv2019

Modeling e-Learners' Cognitive and Metacognitive Strategy in Comparative Question Solving

Cognitive and metacognitive strategy had demonstrated a significant role in self-regulated learning (SRL), and an appropriate use of strategies is beneficial to effective learning or question-solving tasks during a human-computer interaction process. This paper proposes a novel method combining Knowledge Map (KM) based data mining technique with Thinking Map (TM) to detect learner's cognitive and metacognitive strategy in the question-solving scenario. In particular, a graph-based mining algorithm is designed to facilitate our proposed method, which can automatically map cognitive strategy to metacognitive strategy with raising abstraction level, and make the cognitive and metacognitive process viewable, which acts like a reverse engineering engine to explain how a learner thinks when solving a question. Additionally, we develop an online learning environment system for participants to learn and record their behaviors. To corroborate the effectiveness of our approach and algorithm, we conduct experiments recruiting 173 postgraduate and undergraduate students, and they were asked to complete a question-solving task, such as "What are similarities and differences between array and pointer?" from "The C Programming Language" course and "What are similarities and differences between packet switching and circuit switching?" from "Computer Network Principle" course. The mined strategies patterns results are encouraging and supported well our proposed method.

cs.CY↗
arXiv2018

Giant Anisotropic Magnetoresistance due to Purely Orbital Rearrangement in the Quadrupolar Heavy Fermion Superconductor PrV$_2$Al$_{20}$

We report the discovery of giant and anisotropic magnetoresistance due to the orbital rearrangement in a non-magnetic correlated metal. In particular, we measured the magnetoresistance under fields up to 31.4 T in the cubic Pr-based heavy fermion superconductor PrV$_2$Al$_{20}$ with a non-magnetic $Γ_3$ doublet ground state, exhibiting antiferro-quadrupole ordering below 0.7 K. For the [100] direction, we find that the high-field phase appears between 12 T and 25 T, accompanied by a large jump at 12 T in the magnetoresistance ($ΔMR \sim $ 100 $\% $) and in the anisotropic magnetoresistivity (AMR) ratio by $\sim $ 20 $\% $. These observations indicate that the strong hybridization between the conduction electrons and anisotropic quadrupole moments leads to the Fermi surface reconstruction upon crossing the field-induced antiferro-quadrupole (orbital) rearrangement.

cond-mat.str-el↗
arXiv2015

Non-stoichiometry and Defects in the Weyl Semimetals TaAs, TaP, NbP, and NbAs

We report a structural study of the Weyl semimetals TaAs, TaP, NbP, and NbAs, utilizing diffraction techniques (single crystal x-ray diffraction and energy dispersive spectroscopy) and imaging techniques (transmission electron microscopy/scanning transmission electron microscopy). We observe defects of various degrees, leading to non-stoichiometric single crystals of all four semimetals. While TaP displays a large pnictide deficiency with composition TaP$_{0.83(3)}$, and stacking faults accompanied by anti-site disorder and site vacancies, TaAs displays transition metal deficiency with composition Ta$_{0.92(2)}$As and a high density of stacking faults. NbP also displays pnictide deficiency, yielding composition NbP$_{0.95(2)}$, and lastly, NbAs display very little deviation from a 1:1 composition, NbAs$_{1.00(3)}$, and is therefore recommended to serve as the model compound for these semimetals.

cond-mat.mtrl-sci↗
arXiv2015

Generation of elliptically polarized nitrogen-ion laser fields using two-color femtosecond laser pulses

We experimentally investigate generation of molecular nitrogen-ion lasers with two femtosecond laser pulses at different wavelengths. The first pulse serves as the pump which ionizes the nitrogen molecules and excites the molecular ions to excited electronic states. The second pulse serves as the probe which leads to stimulated emission from the excited molecular ions. We observe that changing the angle between the polarization directions of the two pulses gives rise to elliptically polarized molecular nitrogen-ion laser fields, which is interpreted as a result of strong birefringence of the gain medium near the wavelengths of the molecular nitrogen-ion laser.

physics.optics↗
arXiv2015

Time-resolved shadowgraphs of transient plasma induced by spatiotemporally focused femtosecond laser pulses in fused silica glass

We report on experimental observations of formation and evolution of transient plasma produced in fused silica glass with spatiotemporally focused (STF) femtosecond laser pulses using a pump-probe shadow imaging technique. Surprisingly, the observation shows that the track of the plasma is significantly curved, which is attributed to an asymmetric density distribution of the transient plasma produced in the focal volume caused by the pulse front tilt of the STF laser field.

physics.optics↗
arXiv2015

Population Redistribution among Multiple Electronic States of Molecular Nitrogen Ions in Strong Laser Fields

We carry out a combined theoretical and experimental investigation on the population distributions in the ground and excited states of tunnel ionized N2 molecules at various driver wavelengths in the near- and mid-infrared range. Our results reveal that efficient couplings (i.e., population exchanges) between the ground state and the excited states occur in strong laser fields. The couplings result in the population inversion between the ground and the excited states at the wavelengths near 800 nm, which is verified by our experiment by observing the amplification of a seed at ~391 nm. The result provides insight into the mechanism of free-space nitrogen ion lasers generated in remote air with strong femtosecond laser pulses.

physics.optics↗
arXiv2015

Magnetic torque anomaly in the quantum limit of the Weyl semi-metal NbAs

Electrons in materials with linear dispersion behave as massless Weyl- or Dirac-quasiparticles, and continue to intrigue physicists due to their close resemblance to elusive ultra-relativistic particles as well as their potential for future electronics. Yet the experimental signatures of Weyl-fermions are often subtle and indirect, in particular if they coexist with conventional, massive quasiparticles. Here we report a large anomaly in the magnetic torque of the Weyl semi-metal NbAs upon entering the "quantum limit" state in high magnetic fields, where topological corrections to the energy spectrum become dominant. The quantum limit torque displays a striking change in sign, signaling a reversal of the magnetic anisotropy that can be directly attributed to the topological properties of the Weyl semi-metal. Our results establish that anomalous quantum limit torque measurements provide a simple experimental method to identify Weyl- and Dirac- semi-metals.

cond-mat.mtrl-sci↗
arXiv2015

Spatiotemporal focusing does not always improve axial intensity localization

We report on an experimental comparison on critical intensities of nonlinear self-focusing in air with conventional focusing and spatiotemporal focusing schemes. Our results show that the conventional focusing with the focal lens completely filled with the incident beam allows for the strongest axial intensity confinement against the self-focusing effect. This is because that in the high-numerical-aperture condition, the focal spot will have a compact size which results in a high focal intensity. Meanwhile, the Rayleigh length of the focused beam will be substantially shortened which efficiently postpones the onset of self-focusing.

physics.optics↗
arXiv2015

Backward air lasing actions induced by femtosecond laser filamentation: influence of population inversion lifetime

We experimentally investigate generation of backward 357 nm N2 laser in a gas mixture of N2/Ar using 800-nm femtosecond laser pulses, and examine the involved gain dynamics based on pump-probe measurements. Our findings show that a minimum lifetime of population inversion in the excited N2 molecules is required for generating intense backward nitrogen lasers, which is ~0.8 ns under our experimental conditions. The results shed new light on the mechanism for generating intense backward lasers from ambient air, which are highly in demand for high sensitivity remote atmospheric sensing application.

physics.optics↗
arXiv2015

Field induced density wave in the heavy fermion compound CeRhIn5

Metals containing Ce often show strong electron correlations due to the proximity of the 4f state to the Fermi energy, leading to strong coupling with the conduction electrons. This coupling typically induces a variety of competing ground states, including heavy-fermion metals, magnetism and unconventional superconductivity. The d-wave superconductivity in CeTMIn5 (TM=Co, Rh, Ir) has attracted significant interest due to its qualitative similarity to the cuprate high-Tc superconductors. Here, we show evidence for a field induced phase-transition to a state akin to a density-wave (DW) in the heavy fermion CeRhIn5, existing in proximity to its unconventional superconductivity. The DW state is signaled by a hysteretic anomaly in the in-plane resistivity accompanied by the appearance of non-linear electrical transport at high magnetic fields (>27T), which are the distinctive characteristics of density-wave states. The unusually large hysteresis enables us to directly investigate the Fermi surface of a supercooled electronic system and to clearly associate a Fermi surface reconstruction with the transition. Key to our observation is the fabrication of single crystal microstructures, which are found to be highly sensitive to "subtle" phase transitions involving only small portions of the Fermi surface. Such subtle order might be a common feature among correlated electron systems, and its clear observation adds a new perspective on the similarly subtle CDW state in the cuprates.

cond-mat.str-el↗
arXiv2014

Field-Induced Quadrupolar Quantum Criticality in PrV2Al20

PrV2Al20 is the heavy fermion superconductor based on the cubic Gamma3 doublet that exhibits non- magnetic quadrupolar ordering below ~ 0.6 K. Our magnetotransport study on PrV2Al20 reveals field-induced quadrupolar quantum criticality at Hc ~ 11 T applied along the [111] direction. Near the critical field Hc required to suppress the quadrupolar state, we find a marked enhancement of the resistivity rho(H, T), a divergent effective mass of quasiparticles and concomitant non-Fermi liquid (NFL) behavior (i.e. rho(T) ~ T^n with n < 0.5). We also observe the Shubnikov de Haas-effect above ?Hc, indicating the enhanced effective mass m/m0 ~ 10. This reveals the competition between the nonmagnetic Kondo effect and the intersite quadrupolar coupling, leading to the pronounced NFL behavior in an extensive region of T and H emerging from the quantum critical point.

cond-mat.str-el↗
arXiv2014

Interferometric characterization of pulse front tilt of spatiotemporally focused femtosecond laser pulses for three-dimensional micromachining application

We report on an experimental measurement of the pulse front tilt (PFT) of spatiotemporally focused femtosecond laser pulses in the focal plane in both air and bulk transparent materials, which is achieved by examination of the interference pattern between the spatiotemporally focused pulse and a conventional focused reference pulse as a function of time delay between the two pulses. Our simulation results agree well with the experimental observations.

physics.optics↗
arXiv2014

Gain dynamics of a free-space nitrogen laser pumped by circularly polarized femtosecond laser pulses

We experimentally demonstrate ultrafast dynamic of generation of a strong 337-nm nitrogen laser by injecting an external seed pulse into a femtosecond laser filament pumped by a circularly polarized laser pulse. In the pump-probe scheme, it is revealed that the population inversion between the excited and ground states of N2 for the free-space 337-nm laser is firstly built up on the timescale of several picoseconds, followed by a relatively slow decay on the timescale of tens of picoseconds, depending on the nitrogen gas pressure. By measuring the intensities of 337-nm signal from nitrogen gas mixed with different concentrations of oxygen gas, it is also found that oxygen molecules have a significant quenching effect on the nitrogen laser signal. Our experimental observations agree with the picture of electron-impact excitation.

physics.optics↗