Searcharxiv⌕ Search

arXiv subjects

Geng Li

Publications and source records attributed to Geng Li.

At least 37 records · Page 2Linked to original sources
arXiv2026

OccamToken: Efficient VLM Inference with Training-Free and Budget-Adaptive Token Pruning

Vision-language models (VLMs) rely on long visual token sequences for visual understanding, making the prefill stage expensive in both computation and memory. Most existing pruning methods follow an absolute-ranking paradigm, assigning importance scores to visual tokens and retaining a fixed top-K subset. In this work, we argue that this paradigm is fundamentally brittle: attention sinks distort token importance rankings, while image redundancy and query-dependent visual evidence make fixed token budgets unreliable across inputs. We propose OccamToken, a training-free framework that replaces absolute token ranking with register-anchored relative evidence testing. Instead of asking which tokens are globally important, OccamToken evaluates whether a visual token provides information beyond a register-based reference. Our key insight is that register tokens naturally absorb low-information attention patterns, making them a stable reference for identifying genuinely informative visual evidence. Based on this principle, OccamToken performs both image-adaptive redundancy pruning and query-adaptive relevance pruning through dynamic thresholds derived from register attention. Across LLaVA-NeXT, LLaVA-v1.5, and Qwen3-VL, OccamToken consistently improves the accuracy-efficiency trade-off without additional training. Notably, on LLaVA-NeXT, it reduces 2,880 visual tokens to approximately 40 while preserving over 93% of the original accuracy, enabling stable visual token compression even in the extreme 1.4% retention regime.

cs.CV↗
arXiv2026

Gaze2Act: Gaze-Conditioned Vision-Language-Action Policies for Interactive Robot Manipulation

Vision-Language-Action (VLA) models have recently shown strong potential for robot learning by following language instructions. However, in practice, language alone is often insufficient to precisely convey human intent. It is difficult to describe which exact object to interact with among similar candidates, where to act on the object, or how the target may change during execution. To address this limitation, we propose Gaze2Act, a novel VLA framework that leverages human gaze as a dynamic and intuitive intent signal for complex interactive manipulation. Gaze2Act first bridges the ego-exo view gap by mapping first-person gaze into the robot's perspective through cross-view semantic matching, producing both an object mask and a gaze point for coarse-to-fine target specification. These cues are then integrated into the policy through perception-level prompting and action-level conditioning, allowing the robot to attend to relevant regions and execute precise interactions under dynamic intent. In a systematic evaluation across seven task categories and 16 real-robot tasks on a Unitree G1 humanoid, Gaze2Act achieves state-of-the-art performance in both intent accuracy and task success rate. It notably outperforms baselines in object disambiguation, fine-grained interaction, and dynamic intent steering. These results demonstrate that human gaze provides a natural, low-burden, and highly expressive modality for human-in-the-loop VLA control.

cs.RO↗
arXiv2026

Re-evaluation of bottleneck effect via a coupled monolayer WS_2/photonic crystal heterostructure

Exciton-polariton condensates is an important type of Bose-Einstein condensate whose realization requires efficient relaxation of polaritons to the band-energy minima. However, this process is often obstructed by bottleneck effect near the anticrossing region of polariton dispersion. Although the exciton-polariton bottleneck effect has been extensively observed in various polariton system, but there is no a unified views of physical origin. Here, we construct an exciton-trion-photon coupling system in monolayer WS_2/photonic-crystal slab heterostructures. Momentum-resolved photoluminescence reveals the anticrossing polariton dispersions for the exciton resonance with a ~57 meV Rabi splitting and there is no characteristic anticrossing for trion resonance with a ~5 meV splitting at ~12 K. Enhanced polariton emission is observed around the trion-polariton crossing with elevating temperature. We attributes this exotic phenomenon to bottleneck effect and indicating that small Rabi splitting is the unified origin of bottleneck effect in polariton systems.

physics.optics↗
arXiv2026

FIKA-Bench: From Fine-grained Recognition to Fine-Grained Knowledge Acquisition

Fine-grained recognition in everyday life is often not a closed-book classification problem: when encountering unfamiliar objects, humans actively search, compare visual details, and verify evidence before deciding. Existing benchmarks primarily evaluate visually recognition, leaving this active external knowledge acquisition ability underexplored. We study fine-grained knowledge acquisition, where a system must seek, verify, and use external evidence to answer open-ended fine-grained recognition questions. We introduce FIKA-Bench, a leakage-aware and evidence-grounded collection of 311 public-source and real-life instances. To ensure high quality, every example is filtered against frontier closed-book models to remove memorized cases and audited to eliminate image-answer leakage, retaining only samples supported by verified evidence. Our evaluation of latest Large Multimodal Models (LMMs) and agents reveals that the task remains a formidable challenge: the best system reaches only 25.1% accuracy, with no model exceeding 30%. Crucially, we find that merely equipping models with tools is insufficient to bridge this gap; agent failures are predominantly driven by wrong entity retrieval and poor visual judgement. These results show that reliable knowledge acquisition needs better agent designs that focus on fine-grained recognition.

cs.CV↗
arXiv2026

Feedback World Model Enables Precise Guidance of Diffusion Policy

World models aim to improve robotic decision making by predicting the consequences of actions. However, in practice, their predictions often become unreliable once the robot encounters states outside the training distribution, limiting their effectiveness at deployment. We observe that execution itself provides a natural but underutilized signal: after each action, the robot directly observes the true next state, revealing the mismatch between predicted and actual outcomes. Building on this insight, we propose feedback world model, a new paradigm that closes the loop between prediction and observation at inference time. Instead of treating the world model as a static open-loop predictor, our method maintains a lightweight feedback state that is updated online to iteratively correct future predictions, compensating for model errors using real-time observations without additional training data or parameter updates. We show that this process can be interpreted as a latent-space observer and admits convergence guarantees under mild conditions. We further introduce action-aware guidance to better translate corrected predictions into control by emphasizing action-controllable components while suppressing irrelevant variations. Experiments on LIBERO-Plus, Robomimic, and real-world manipulation tasks demonstrate that our method substantially improves both prediction accuracy and policy performance under distribution shift. In particular, it reduces world model prediction error by up to 76.4% and improves out-of-distribution (OOD) success rate by 30%. These results show that incorporating real-time feedback at inference time provides a simple yet powerful alternative to static world modeling.

cs.RO↗
arXiv2026

EVA-0: Test-Time Model Evolution with Only Two Forward Passes per Sample

Test-time model evolution offers a promising way for deployed models to improve from unlabeled test-time experience, yet most existing methods depend on backpropagation (BP), which incurs substantial memory overhead and makes them difficult to deploy on edge devices, quantized models, specialized accelerators, or black-box models. In this work, we study test-time model evolution under a strict two-forward budget, a setting that pushes adaptation toward highly efficient real-world deployment. We reveal three key obstacles in zeroth-order test-time optimization: susceptibility to shortcut solutions, uncontrolled weight drift, and ineffective update direction estimation. To overcome them, we propose EVA-0, a minimal zeroth-order adaptation framework that: 1) keeps the loss scale-invariant to prevent shortcut solutions; 2) devises an anchor-guided optimization strategy to alleviate weight drift; 3) uses sample-wise symmetric two-sided perturbation for update direction estimation and inference. EVA-0 requires no BP and performs both inference and adaptation within only two forward passes per sample. Results on ImageNet-C & ViT-Base show that EVA-0 outperforms both BP-based DeYO and BP-free FOA, while achieving a 14x speed-up over FOA. Code will be released.

cs.LG↗
arXiv2026

Production Rate of Glueball-like $X(2370)$ in $J/ψ$ Radiative Decay

$X(2370)$ falls in the mass region of the lowest pseudoscalar glueball predicted by lattice QCD studies and its decay properties are similar to those of $η_c$. A previous lattice QCD study finds that the pseudoscalar glueball ($G_{0^-}$) and the lowest pseudoscalar charmonium ($c\bar{c}(0^-)$) can mix with a small mixing angle $\sinθ$. It is therefore possible that $X(2370)$ and $η_c$ are admixtures of $G_{0^-}$ and $c\bar{c}(0^-)$. In this picture, although $\mathrm{Br}(J/ψ\toγη_c)$ is insensitive to the small $\sinθ$, $\mathrm{Br}(J/ψ\to γX(2370))$ can be enlarged drastically by the mixing due to the much larger kinematic factor for $J/ψ\to γX(2370)$ and the much larger transition form factor for $J/ψ\to γ(c\bar{c}(0^{-}))$. Depending on the value of $\sinθ$, $\mathrm{Br}(J/ψ\to γX(2370))$ can be much larger than that of the pure pseudoscalar glueball, namely, $2.3(8)\times 10^{-4}$ that is predicted by a quenched lattice QCD calculation. Present results by BESIII favor a small mixing angle of $\mathcal{O}(1^\circ)$, which can be further constrained by more measurements of $X(2370)$ decays if the mixing picture applies here.

hep-ph↗
arXiv2026

Assessing the Robustness of Climate Foundation Models under No-Analog Distribution Shifts

The accelerating pace of climate change introduces profound non-stationarities that challenge the ability of Machine Learning based climate emulators to generalize beyond their training distributions. While these emulators offer computationally efficient alternatives to traditional Earth System Models, their reliability remains a potential bottleneck under "no-analog" future climate states, which we define here as regimes where external forcing drives the system into conditions outside the empirical range of the historical training data. A fundamental challenge in evaluating this reliability is data contamination; because many models are trained on simulations that already encompass future scenarios, true out-of-distribution (OOD) performance is often masked. To address this, we benchmark the OOD robustness of three state-of-the-art architectures: U-Net, ConvLSTM, and the ClimaX foundation model specifically restricted to a historical-only training regime (1850-2014). We evaluate these models using two complementary strategies: (i) temporal extrapolation to the recent climate (2015-2023) and (ii) cross-scenario forcing shifts across divergent emission pathways. Our analysis within this experimental setup reveals an accuracy vs. stability trade-off: while the ClimaX foundation model achieves the lowest absolute error, it exhibits higher relative performance changes under distribution shifts, with precipitation errors increasing by up to 8.44% under extreme forcing scenarios. These findings suggest that when restricted to historical training dynamics, even high-capacity foundation models are sensitive to external forcing trajectories. Our results underscore the necessity of scenario-aware training and rigorous OOD evaluation protocols to ensure the robustness of climate emulators under a changing climate.

cs.LG↗
arXiv2026

Discovery of parity-violating chiral polar-nematic charge density wave and superconductivity in kagome metals

Nonmagnetic kagome metals and superconductors AV3Sb5 (A = K, Rb, Cs) host unconventional charge density wave (CDW) and superconducting (SC) phases accompanied by multiple electronic symmetry breaking. Due to the centrosymmetric crystal structure, inversion symmetry has generally been assumed to hold. Here, using scanning tunneling microscopy complemented by atomic force microscopy and optical second-harmonic generation, we directly reveal that inversion symmetry in the kagome plane is spontaneously broken in the CDW state. The mixed-parity CDW state exhibits ferroelectric dipolar and nematic quadrupolar ordered moments. The coexistence and coupling between the dipole and quadrupole favor noncollinear ferro-polar and nematic alignment that breaks all mirror symmetries and gives rise to robust electronic chirality in the 3Q CDW. The multipolar coupling to in-plane electric field enables electric field control and manipulation of the chiral polar-nematic CDW state, including its chirality. Below the SC transition, we observe parity-violating pair density modulations at both the original and the CDW lattice wavevectors. Our findings of parity-violating electronic chiral multipolar order provide microscopic insights into the magnetoelectric and nonreciprocal transport, loop current order, pairing density waves, and unconventional superconductivity in kagome metals and related quantum materials.

cond-mat.supr-con↗
arXiv2026

Coupling of phase transition, anharmonicity, and thermal transport in CaSnF$_6$

Understanding the coupling between structural phase transitions and thermal transport is essential for designing functional materials with tunable properties. Here, we investigate this interplay in CaSnF$_6$ by combining first-principles calculations with a machine-learned neuroevolution potential that enables large-scale molecular dynamics simulations across a wide temperature range. The simulations accurately capture the first-order structural phase transition and associated lattice dynamics. We show that the negative thermal expansion originates from low-energy rigid unit modes involving cooperative rotations of corner-sharing [CaF$_6$]$^{4-}$ octahedra, which induce bond-angle bending and volume contraction. At the same time, strong anharmonicity, dominated by four-phonon scattering, plays a central role in suppressing lattice thermal conductivity ($κ_L$). Crucially, non-equilibrium simulations reveal a pronounced non-monotonic anomaly in $κ_L$ near the phase transition, deviating from the conventional $\sim 1/T^α$ behavior and providing direct transport evidence of lattice reconstruction. These results establish a unified mechanism linking lattice geometry, anharmonic vibrational dynamics, and thermal transport, and highlight the potential of machine-learned potentials for bridging atomic-scale phase transitions with macroscopic transport properties.

cond-mat.mtrl-sci↗
arXiv2026

Shortcuts to state transitions for active matter

Shortcut schemes can accelerate quasi-static processes in passive systems by adding auxiliary controls to realize swift transitions between equilibrium states. In active systems, however, inherently directed motion driven by free energy consumption continually drives the system away from equilibrium. In this work, we develop a shortcut framework to realize swift state transitions for active systems operating in the weak activity regime. An auxiliary potential is introduced to guide the system along a predefined distribution path, allowing it to reach the target state within a finite time. Considering unavoidable energy cost in such a finite-time process, we derive a thermodynamic metric from the dissipative work to induce a Riemann manifold on the space spanned by the control parameters. The optimal protocol with minimum dissipative work is then identical to the geodesic path in the geometric space. We demonstrate this framework by considering active systems confined in an external harmonic trap and interacting via two distinct internal potentials, respectively: an attractive harmonic coupling and a repulsive pairwise Gaussian-core coupling. The strengths of both the external trap and the internal interactions are controllable. For the latter case, since the auxiliary potential can not be derived precisely, we adopt a variational method to obtain an approximate auxiliary control. Compared to linear protocols, the geodesic protocols can effectively reduce dissipation.

cond-mat.stat-mech↗
arXiv2026

Realization of a Wigner-Mott insulator in 6R-TaS$_2$ superconductor

Wigner-Mott insulating states represent a paradigmatic manifestation of strong electronic correlations, in which long-range Coulomb interactions drive spontaneous charge ordering and enable Mott localization at fractional electronic fillings. Such states have been theoretically proposed to arise from the cooperative interplay between onsite and inter-site Coulomb interactions. However, experimental realizations of the simultaneous microscopic observation of interaction-driven charge order and genuine Mott localization, which are the defining hallmarks of a Wigner-Mott insulator, have remained elusive. Here we report the observation of a Wigner-Mott insulating state in 6R-TaS$_2$ using scanning tunneling microscopy. By locally injecting electrons into the depleted 1T layer, we induce distinct Star-of-David charge-ordered superstructures and realize a cascade of insulating phases. In particular, a $\sqrt{3}\times \sqrt{3}$ charge-ordered superstructure at one-third filling hosts a robust Mott gap despite fractional filling. The spontaneous relaxation from excited states back to the ground state demonstrates that this Wigner-Mott phase is stabilized by the cooperative effects of onsite and inter-site Coulomb interactions. Our results provide direct microscopic evidence for a Wigner-Mott mechanism and establish 6R-TaS$_2$ as a platform for the controlled realization and investigation of Wigner-Mott insulating states.

cond-mat.str-el↗
arXiv2025

Do machine learning climate models work in changing climate dynamics?

Climate change is accelerating the frequency and severity of unprecedented events, deviating from established patterns. Predicting these out-of-distribution (OOD) events is critical for assessing risks and guiding climate adaptation. While machine learning (ML) models have shown promise in providing precise, high-speed climate predictions, their ability to generalize under distribution shifts remains a significant limitation that has been underexplored in climate contexts. This research systematically evaluates state-of-the-art ML-based climate models in diverse OOD scenarios by adapting established OOD evaluation methodologies to climate data. Experiments on large-scale datasets reveal notable performance variability across scenarios, shedding light on the strengths and limitations of current models. These findings underscore the importance of robust evaluation frameworks and provide actionable insights to guide the reliable application of ML for climate risk forecasting.

cs.LG↗
arXiv2025

Flavor Physics at the CEPC: a General Perspective

We discuss the landscape of flavor physics at the Circular Electron-Positron Collider (CEPC), based on the nominal luminosity outlined in its Technical Design Report. The CEPC is designed to operate in multiple modes to address a variety of tasks. At the $Z$ pole, the expected production of 4 Tera $Z$ bosons will provide unique and highly precise measurements of $Z$ boson couplings, while the substantial number of boosted heavy-flavored quarks and leptons produced in clean $Z$ decays will facilitate investigations into their flavor physics with unprecedented precision. We investigate the prospects of measuring various physics benchmarks and discuss their implications for particle theories and phenomenological models. Our studies indicate that, with its highlighted advantages and anticipated excellent detector performance, the CEPC can explore beauty and $τ$ physics in ways that are superior to or complementary with the Belle II and Large-Hadron-Collider-beauty experiments, potentially enabling the detection of new physics at energy scales of 10 TeV and above. This potential also extends to the observation of yet-to-be-discovered rare and exotic processes, as well as testing fundamental principles such as lepton flavor universality, lepton and baryon number conservation, etc., making the CEPC a vibrant platform for flavor physics research. The $WW$ threshold scan, Higgs-factory operation and top-pair productions of the CEPC further enhance its merits in this regard, especially for measuring the Cabibbo-Kobayashi-Maskawa matrix elements, and Flavor-Changing-Neutral-Current physics of Higgs boson and top quarks. We outline the requirements for detector performance and considerations for future development to achieve the anticipated scientific goals.

hep-ex↗
arXiv2025

Single-photon detection enabled by negative differential conductivity in moiré superlattices

Detecting individual light quanta is essential for quantum information, space exploration, advanced machine vision, and fundamental science. Here, we introduce a novel single photon detection mechanism using highly photosensitive non-equilibrium electron phases in moiré materials. Using tunable bands in bilayer graphene/hexagonal-boron nitride superlattices, we engineer negative differential conductance and a sensitive bistable state capable of detecting single photons. Operating in this regime, we demonstrate single-photon counting at mid-infrared (11.3 microns) and visible wavelengths (675 nanometres) and temperatures up to 25 K. This detector offers new prospects for broadband, high-temperature quantum technologies with CMOS compatibility and seamless integration into photonic integrated circuits (PICs). Our analysis suggests the mechanism underlying our device operation originates from negative differential velocity, and represents an important milestone in the field of high-bias transport in two-dimensional moiré quantum materials.

cond-mat.mes-hall↗
arXiv2025

DyFo: A Training-Free Dynamic Focus Visual Search for Enhancing LMMs in Fine-Grained Visual Understanding

Humans can effortlessly locate desired objects in cluttered environments, relying on a cognitive mechanism known as visual search to efficiently filter out irrelevant information and focus on task-related regions. Inspired by this process, we propose Dyfo (Dynamic Focus), a training-free dynamic focusing visual search method that enhances fine-grained visual understanding in large multimodal models (LMMs). Unlike existing approaches which require additional modules or data collection, Dyfo leverages a bidirectional interaction between LMMs and visual experts, using a Monte Carlo Tree Search (MCTS) algorithm to simulate human-like focus adjustments. This enables LMMs to focus on key visual regions while filtering out irrelevant content, without introducing additional training caused by vocabulary expansion or the integration of specialized localization modules. Experimental results demonstrate that Dyfo significantly improves fine-grained visual understanding and reduces hallucination issues in LMMs, achieving superior performance across both fixed and dynamic resolution models. The code is available at https://github.com/PKU-ICST-MIPL/DyFo_CVPR2025

cs.CV↗
arXiv2025

FingER: Content Aware Fine-grained Evaluation with Reasoning for AI-Generated Videos

Recent advances in video generation have posed great challenges in the assessment of AI-generated content, particularly with the emergence of increasingly sophisticated models. The various inconsistencies and defects observed in such videos are inherently complex, making overall scoring notoriously difficult. In this paper, we emphasize the critical importance of integrating fine-grained reasoning into video evaluation, and we propose $\textbf{F}$ing$\textbf{ER}$, a novel entity-level reasoning evaluation framework that first automatically generates $\textbf{F}$ine-grained $\textbf{E}$ntity-level questions, and then answers those questions by a $\textbf{R}$easoning model with scores, which can be subsequently weighted summed to an overall score for different applications. Specifically, we leverage LLMs to derive entity-level questions across five distinct perspectives, which (i) often focus on some specific entities of the content, thereby making answering or scoring much easier by MLLMs, and (ii) are more interpretable. Then we construct a FingER dataset, consisting of approximately 3.3k videos and corresponding 60k fine-grained QA annotations, each with detailed reasons. Based on that, we further investigate various training protocols to best incentivize the reasoning capability of MLLMs for correct answer prediction. Extensive experiments demonstrate that a reasoning model trained using Group Relative Policy Optimization (GRPO) with a cold-start strategy achieves the best performance. Notably, our model surpasses existing methods by a relative margin of $11.8\%$ on GenAI-Bench and $5.5\%$ on MonetBench with only 3.3k training videos, which is at most one-tenth of the training samples utilized by other methods. Our code and dataset will be released soon.

cs.CV↗