arXiv subjects

Jungwoo Lee

Publications and source records attributed to Jungwoo Lee.

At least 19 recordsLinked to original sources

arXiv2026

Evidence for a charged vortex lattice in an exciton superfluid

Lattices of quantized vortices are a defining property of rotating superfluids and of type-II superconductors in magnetic fields. Realizing an analogous vortex lattice in an exciton superfluid-formed from bound electron-hole pairs-has remained challenging because excitons are electrically neutral. Building on previous theoretical studies of the low-energy excitations in an exciton superfluid, recent theory has predicted that in a strong magnetic field, a weakly charged exciton superfluid can stabilize a vortex-antivortex lattice arising from the mutual repulsion of fractionally charged vortex cores. Here we report evidence for this predicted vortex lattice in a bilayer exciton condensate by imaging its charged vortices with the scanning tunnelling microscope. The condensate forms between Landau levels of two graphene layers separated by a thin hexagonal boron nitride barrier in a perpendicular magnetic field. We find that the vortex lattice periodicity scales with the total excess charge and is insensitive to its distribution between the two layers, in quantitative agreement with theoretical predictions. Our observations provide evidence for phase coherence in an exciton superfluid system.

cond-mat.str-el↗

arXiv2026

RadLoc: Radar-based 3-DoF Global Localization via Fast, Robust, and Lightweight Spatial Descriptor Across Diverse Environmental Scenarios

While global localization using spinning radar has gained attention for its robustness to adverse weather and challenging environments, many studies have focused on individual components such as place recognition or pose estimation. In this paper, we take a holistic view of radar sensor-based global localization and present RadLoc, a fast, robust, and lightweight end-to-end pipeline from place recognition to 3-DoF pose estimation. RadLoc accelerates pre-processing using 1D CA-CFAR filtering and leverages the near-range dominance in spinning radar images to design a compact descriptor and an efficient hierarchical coarse-to-fine retrieval strategy. Moreover, coupled with phase correlation-based 3-DoF pose estimation, it forms a versatile global localization module applicable to SLAM and multi-session SLAM systems. Extensive experiments on 15 sequences across 5 datasets demonstrate that RadLoc achieves robust performance while maintaining the smallest descriptor size and fastest retrieval time among state-of-the-art approaches. The supplementary materials are available at https://sparolab.github.io/research/radloc/.

cs.RO↗

arXiv2026

Collision-aware Bayesian mitigation of photon loss in linear optical quantum circuits

Photon loss is the dominant noise source in linear optical quantum circuits. Each lost photon turns an ideal n-photon outcome into a pattern with fewer photons, so the detector counts no longer follow the ideal output distribution, and the lossless events that do follow it become exponentially rare as the photon number grows. Discarding the lossy events (postselection) therefore wastes most of the data at realistic loss rates, and the ideal distribution has to be inferred from the lossy counts themselves. Recycling-based mitigation (Mills & Mezher, Quantum 2026; Quandela's photon_recycling) reweights lossy samples to recover lossless probabilities, but its estimators are biased and restricted to the no-collision sector, where each output mode holds at most one photon. We instead treat the raw counts as Multinomial observations under per-photon binomial loss and infer the ideal output distribution and the loss rate jointly from a Bayesian posterior, which extends naturally to outputs with collisions. On data from the independent Perceval simulator at the scale of prior work (n=4 photons in m=20 and m=10 modes), the posterior has lower KL divergence than both recycling and postselection at every tested loss rate in the no-collision benchmark. In the collision regime it recovers the loss rate and the collision-pattern probabilities, which recycling cannot estimate, and stays below recycling at every loss rate and below postselection except at low loss, where the two are comparable. Its advantage over recycling grows with the sample count, and it supplies credible intervals that the baselines lack.

quant-ph↗

arXiv2026

Why Latent Actions Fail, and How to Prevent It

Latent action models (LAMs) aim to learn action-like representations from unlabeled videos by compressing frame-to-frame changes. The frames of in-the-wild videos, however, contain not only the agent's own state but exogenous state such as background clutter. Since the exogenous state introduces changes unrelated to actions, it hinders reliable latent action learning. This paper investigates this problem analytically by extending a linear LAM framework to explicitly model exogenous state. Our analysis reveals two insights: (1) minimizing the standard reconstruction objective produces latent actions that encode exogenous information from future observation; and (2) learning in a representation space that focuses on endogenous components is a key to mitigating the interference of noise. We further show that previously proposed auxiliary objectives, such as action-supervision, provably encourage latent actions to be consistent across exogenous states. These findings are validated through experiments on both linear and nonlinear LAMs, providing a unified theoretical analysis of how exogenous state hinders latent action learning and why common remedies work.

cs.CV↗

arXiv2026

Mechanistic Circuit Identification for Controllable Data Generation

While recent advances in data synthesis aim to curate high-quality datasets, most generation pipelines still rely on heuristic prompt-based control. This black-box paradigm provides limited insight into how individual samples interact with a model's underlying learning dynamics. To bridge this gap, we propose a circuit-grounded framework that connects training-dynamics-based data valuation with mechanistic interpretability (MI). Specifically, we conceptualize data quality along three complementary utility axes, learnability, challenge, and alignment. First, we uncover specialized model-internal circuits that causally govern these utility signals. Then, moving beyond heuristic prompting toward mechanistic control, we leverage these circuits as controllable interfaces, actively steering generation to produce utility-targeted data. Building on this capability, we introduce SAMS (Stage-Aware Mechanistic Scheduling), which schedules circuit-steered data according to the model's evolving optimization needs. Experiments on multiple-choice QA tasks demonstrate that our approach yields precisely controlled data with greater diversity than prompt-based baselines, consistently improving downstream performance and calibration. Ultimately, this work establishes a principled white-box paradigm for interpretable data generation, pioneering the use of MI not just as an analytical tool, but as a practical, controllable interface.

cs.LG↗

arXiv2026

DCGC: Draft-Conditioned Global Correction for Complex Reasoning with Masked Diffusion Models

Correcting flawed reasoning traces remains a significant challenge for Large Language Models (LLMs), whose autoregressive generation can propagate early mistakes into subsequent reasoning. We introduce DCGC, a Masked Diffusion Model (MDM) framework for global correction that uses an imperfect solution draft from an upstream solver as auxiliary context. DCGC combines task-specific Supervised Fine-Tuning (SFT) with a novel inference-time mechanism called Dynamic Dual-CFG. This mechanism separates problem-only and joint problem-draft branches and scales the draft-conditioned residual using a relative confidence gap. Across math, code, and knowledge reasoning benchmarks, DCGC outperforms standard sampling and simpler CFG variants, with additional results suggesting transfer to different diffusion backbones. In test-time setting where ground-truth failure labels are unavailable, DCGC improves full test set accuracy by correcting low-consensus upstream outputs, highlighting its utility as a verifier-free global correction module for difficult reasoning instances.

cs.CL↗

arXiv2026

Re-uploading quantum data: a universal function approximator for quantum inputs

Quantum data re-uploading has proved powerful for classical inputs, where repeatedly encoding features into a small circuit yields universal function approximation. Extending this idea to quantum inputs remains underexplored, as the information contained in a quantum state is not directly accessible in classical form. We propose and analyze a quantum data re-uploading architecture in which a qubit interacts sequentially with fresh copies of an arbitrary input state. The circuit can approximate any bounded continuous function using only one ancilla qubit and single-qubit measurements, provided that sufficiently many copies of the input state and, in general, sufficiently many layers are available. By alternating entangling unitaries with mid-circuit resets of the input register, the architecture realizes a discrete cascade of completely positive and trace-preserving maps, analogous to collision models in open quantum system dynamics. Our framework provides a space-efficient and expressive approach to designing quantum machine learning models that operate directly on quantum data.

quant-ph↗

arXiv2026

Learning Robust Visuomotor Policies via Dynamics Alignment

Generative BC policies model expert action distributions well, but they do not explicitly reason about what scene a candidate action would produce, so they cannot tell whether it reaches a task-relevant state. We address this by coupling action generation with an action-conditioned dynamics model that predicts the future observations induced by actions. However, naively placing such a dynamics model inside a generative policy creates a circular dependency. Future observations are needed to guide action generation, yet those observations can only be predicted from the action samples being generated. We introduce Dynamics-Aligned Flow Matching Policy (DAP), where a flow matching policy and an action-conditioned flow matching dynamics model co-sample actions and future observations through flow extrapolation. At each step the two flows correct each other, and the samples converge to a self-consistent action and future pair on which the policy and dynamics model agree. Across four simulated tasks and three real-world tasks, DAP achieves the highest average success rate, with relative improvements of 56.5% in simulation and 33.3% in the real world over the strongest baseline in each setting. Moreover, when the future-observation estimate is corrupted at inference time, DAP recovers a self-consistent action-future pair through iterative refinement, whereas video-conditioned and post-hoc dynamics baselines, which use a fixed future surrogate, degrade or fail.

cs.RO↗

arXiv2026

Toward the Goldilocks Blind Compression of Quantum States

Quantum autoencoders (QAEs) are learning architectures that compress quantum data into a low-dimensional latent state while preserving the information needed for reconstruction. We study blind single-copy compression of quantum states through a $k$-qubit bottleneck and investigate the minimal circuit width required to attain the information-theoretic optimum under average infidelity. Between the conventional architecture, which is narrow but nonuniversal, and fully general \emph{completely positive and trace preserving} (CPTP) realizations, which are universal but overparameterized, we identify a balanced regime. We prove that for every distribution of pure $n$-qubit states, there exists a QAE with $k$ encoder ancillas and $n$ decoder ancillas that achieves the optimal fidelity over all CPTP encoder--decoder pairs. The encoder-side statement is sharp in that we construct source families for which every optimal scheme necessarily uses at least $k$ encoder ancillas, thereby determining the universal encoder threshold exactly. On the decoder side, we show that isometric decoders are optimal for several analytically tractable source families, but we also exhibit an explicit counterexample demonstrating that decoder isometry is not universally sufficient. Nevertheless, numerical experiments indicate that the performance gap is practically negligible.

quant-ph↗

arXiv2026

Learning Rate Transfer for Hybrid Transformer-SSM Architectures

We study learning rate (LR) scaling for hybrid architectures combining Transformer and State-Space Model (SSM) blocks, a class adopted by several recent production language models. In particular, we focus on the gap between the theoretical scaling rules derived for SSMs under zero-order-hold (ZOH) discretization at infinite width with growing state size, and the field-standard practical implementations using simplified-ZOH Mamba at fixed state size. Surprisingly, in this practical regime hybrid architectures achieve a near-zero LR transfer gap across widths 256-2048 and depths 4-32 up to billion-parameter scale using only the original $μ$P prescription, even though SSM operations fall outside its Tensor Programs representability conditions and every parameterization we test fails the standard coordinate-check diagnostic of $μ$P correctness. We attribute this to a two-condition decomposition of LR transfer in hybrid architectures: a global update-to-weight invariance, enforced by $μ$P's initialization and LR scaling; and a local per-component balance, provided by AdamW's per-parameter normalization. Our observations show that the optimal LR is invariant to width up to 8$\times$, that this width invariance holds across depth, sequence length, batch size, and Transformer-to-SSM ratio, and that it transfers to Nemotron-H, a production hybrid outside our custom architecture set. We hope these findings fill the gap between theoretical scaling rules and practical hybrid implementations, and stimulate further research toward bridging it.

cs.LG↗

arXiv2026

Rethinking Memorization Mitigation in Diffusion Models: Reinforcing Text Conditioning

Text-to-image diffusion models have achieved remarkable progress in image synthesis, yet can exhibit memorization by closely reproducing individual training examples. Effective mitigation must preserve useful prompt information to guide alternative depictions. We introduce a training-free method that redistributes cross-attention with Gaussian smoothing before reinforcing content-token contributions and attenuating padding contributions, without additional denoiser evaluations. With this intervention, stronger content conditioning can improve prompt alignment at comparable training-image similarity. A local analysis identifies when reinforcement preserves shared value information while redistribution reduces localized attention mass. On Stable Diffusion v1.4 and v2.0, all evaluated smoothing widths lie on the empirical Pareto frontiers for training-image similarity versus both prompt alignment and image preference. A configuration selected on Stable Diffusion reduces template reproduction in DeepFloyd IF without further tuning. These findings support jointly controlling conditioning allocation and strength to generate prompt-consistent alternatives.

cs.CV↗

arXiv2026

Optimal dense materialization of the stabilizer formalism without polynomial overhead

Stabilizer states and Clifford transformations constitute the tractable backbone of quantum information science, from error correction and fault tolerance to benchmarking and simulation. Although these objects admit compact classical descriptions, many physical and computational workflows still require their explicit dense forms such as a full wavefunction for a stabilizer state or a full matrix for a Clifford transformation. In such explicit output tasks, exponential scaling is unavoidable because the outputs themselves have sizes $2^n$ and $4^n$. The fundamental question is therefore whether compact stabilizer and Clifford descriptions can be expanded with no additional polynomial overhead. Here we answer this question affirmatively. We present optimal algorithms that materialize an $n$-qubit stabilizer state vector in $O(2^n)$ time and a full dense Clifford matrix in $O(4^n)$ time. The same framework also yields an optimal conversion from standard stabilizer check matrices to state vectors and, for every fixed odd prime qudit dimension $\ell$, gives $O(\ell^n)$-time materialization of qudit stabilizer states. As an additional compact-to-compact result, we design a sign-aware Four Russians method for converting stabilizer check matrices to quadratic forms faster than Gaussian elimination. These results close the asymptotic gap between compact descriptions of the stabilizer formalism and their dense representations.

quant-ph↗

arXiv2026

Robust Rate-Splitting Design for Mixed Dual-Polarized Integrated Satellite-Terrestrial Networks Under Polarization Mismatch

Dual-polarized transmission offers a promising approach to improve spectral efficiency in multiantenna networks by reusing frequency and time resources across orthogonal polarization domains. Building upon this advantage, this paper investigates interference management in mixed dual-polarized integrated satellite-terrestrial networks (MDP-ISTN), comprising a circularly polarized (CP) satellite sub-network and a linearly polarized (LP) terrestrial sub-network. To this end, we employ rate-splitting multiple access (RSMA), which enables flexible non-orthogonal transmission through partial interference decoding and partial interference treating-as-noise. Specifically, to jointly mitigate both inter-network interference between the CP low Earth orbit (LEO) satellite and LP terrestrial sub-networks as well as intra-network interference within each sub-network, we propose an MDP-RSMA framework that incorporates inter-network rate-splitting (RS) with a super-common message together with intra-network RS. Moreover, we account for practical challenges in MDP-ISTN, including polarization mismatch, channel depolarization, and imperfect channel state information at the transmitter. To maximize the minimum user rate among all satellite and terrestrial users, we formulate a robust precoder optimization problem and develop a weighted minimum mean square error (WMMSE)-based algorithm tailored to the proposed MDP-RSMA. Numerical results demonstrate that the proposed scheme significantly improves the minimum user rate over several baseline schemes across diverse MDP-ISTN scenarios.

cs.IT↗

arXiv2026

Continual Data Unlearning in Diffusion Models via Transition-based Regularization

Data unlearning in diffusion models aims to remove the influence of specific training examples without suppressing the broader concepts they represent. However, when deletion requests arrive sequentially, updates for new requests can degrade generative utility and undermine earlier deletions. We propose a continual data unlearning framework that uses completed deletion transitions as directional references to regularize future updates. For each request, we record changes in denoiser responses on the same fixed noisy inputs before and after unlearning. Rather than matching full post-deletion responses, we apply a one-sided penalty that discourages reversal along the recorded directions relative to the post-deletion references, while leaving orthogonal response changes and progress beyond these references unpenalized. To keep storage independent of the number of requests, we maintain a fixed-capacity bank of representative transition records. Records are selected based on the local sensitivity of progress along their recorded directions to parameter updates, allowing them to be retained even when their penalties are inactive. Empirical evaluations show that the proposed framework achieves a better balance between deletion persistence and generative utility than existing unlearning baselines as requests accumulate, using only a small transition memory.

cs.LG↗

arXiv2026

CHSH inequality always holds in bipartite qutrits with spin-1 observables

Bell inequalities reveal when quantum correlations cannot be explained by any local hidden-variable model, and the CHSH inequality is the most fundamental test of this phenomenon. While entangled two-qubit states can violate CHSH inequalities with suitable local measurements, the situation is more subtle for higher-dimensional systems when the allowed observables are physically restricted. In this work, we study bipartite qutrit systems in which each party is restricted to spin-1 observables $S(u)=u_xS_x+u_yS_y+u_zS_z$, with $u\in\mathbb R^3$ a unit vector. Hanotel and Loubenets conjectured that nonseparable pure states of two qutrits do not violate the CHSH inequality under this measurement restriction. We prove the stronger result that every bipartite state on $\mathbb C^3\otimes\mathbb C^3$, pure or mixed, satisfies the CHSH inequality for all choices of local spin-1 observables. Thus the absence of CHSH violation is not a property of special states, but a structural limitation of the spin-1 measurement family. Our result clarifies a distinction between two-qutrit entanglement and CHSH nonlocality under natural spin-component measurements.

quant-ph↗

arXiv2026

Refining transposition bounds from traceless quantum perturbations

The transpose map is the canonical nonphysical map in quantum information that is central to entanglement tests and to converse bounds for quantum capacity. It is a longstanding conjecture that transposition satisfies a dimension-independent strict submultiplicativity for physically relevant channel differences. We answer this affirmatively by proving that for every quantum channel $T$ on $\mathsf{L}(\mathbb{C}^d)$, $\|Θ\circ(\mathrm{id}-T)\|_\diamond \le \frac{1}{\sqrt{2}}\|Θ\|_\diamond \|\mathrm{id}-T\|_\diamond$. The same bound holds for all Hermiticity-preserving trace-annihilating maps. Because a trace-preserving channel produces a traceless deviation from the identity, transposition cannot act as strongly on such physically generated errors as it can on arbitrary perturbations. Thus, although transposition is extremal in the unrestricted diamond-norm geometry, physical constraints mandate a universal margin in the submultiplicative bound. As one immediate consequence, the finite-error Holevo--Werner converse extends from decoding errors $\varepsilon<1/2$ to $\varepsilon<1/\sqrt{2}$. The same tracelessness mechanism also yields rank-sensitive continuity refinements for negativity and logarithmic negativity, demonstrating its potential for broader application in quantum information.

quant-ph↗

arXiv2026

One-parameter counterexamples to the refined Bessis-Moussa-Villani conjecture

Positivity of matrix trace exponentials is a basic structural principle behind finite-temperature quantum statistical mechanics. The Bessis-Moussa-Villani conjecture, a central manifestation of this principle, was proved by Stahl after an influential reformulation by Lieb and Seiringer. A later refinement asks whether the normalized average over all words with $n$ letters $A$ and $m$ letters $B$ is always bounded above by $\mathrm{tr}(A^nB^m)$ and below by $\mathrm{tr}\exp(n\log A+m\log B)$. In this work, we study a specific one-parameter family $(A_x, B_x)$ and show that the correct small-$x$ invariant of a word is not its degree of fragmentation, but a weighted shortest-bridge cost on its cyclic run decomposition. Our results yield a class of counterexamples to the suggested refinement. Remarkably, the ratio of the normalized word average to the trace $\mathrm{tr}(A^nB^m)$ can become arbitrarily large.

quant-ph↗

arXiv2026

Towards Understanding Pause Token Fine-Tuning Dynamics: A Mode Retention Perspective

Pause-token methods improve LLM reasoning by inserting special tokens into sequences. Prior work explains these gains through computational expressivity. However, there is relatively little investigation into the training dynamics of pause tokens. We explore how pause tokens reshape the training dynamics of fine-tuning. Two controlled pilots expose distinct asymmetries. On a synthetic continual-learning task, masked pauses overwrite a previously-learned distribution roughly 4x less at matched final adaptation (H1, mode retention); on a synthetic math-reasoning probe, the boundary-adjacent token comes to encode substantially more downstream-step information (H2, non-myopic compression). We formalize a training rule consistent with both - Masked Boundary Pause (MBP), pause tokens placed at reasoning-step boundaries with their loss masked. Across 1B-8B Qwen and Llama models, MBP consistently improves reasoning, achieving gains of up to 6 points on math and 2.5 points on code, while preserving general language understanding abilities. We further demonstrate that this mode-preserving strategy extend gains to GRPO. These results recast pause tokens as a training-dynamics intervention on the retention-adaptation trade-off, rather than merely an inference-time computation device.

cs.CL↗