arXiv subjects

Aonan Zhang

Publications and source records attributed to Aonan Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Quantum-limited imaging using diffractive optical neural networks

We cast general imaging as multiparameter quantum estimation of band-limited spatial-frequency amplitudes. For separable (single-copy) measurements, we compute precision limits using semidefinite programming to evaluate the Nagaoka-Hayashi Cramér-Rao bound. We then introduce an architecture for a measurement apparatus based on diffractive optical neural networks and photon counting that saturates this bound. Extending the framework to arbitrary objects and many amplitudes, we show image reconstructions in which our architecture recovers fine features at the quantum limit, outperforming direct imaging. Together, these results open a scalable route to saturating multiparameter quantum limits in superresolution microscopy, telescopy, and remote sensing.

quant-ph↗

arXiv2026

Super-resolving frequency measurement with mode-selective quantum memory

High-precision optical frequency measurement underpins modern science and technology, yet conventional spectroscopic techniques struggle to resolve sub-linewidth spectral features. Here, we introduce a platform for super-resolved frequency estimation based on a mode-selective atomic Raman quantum memory implemented in warm caesium vapour. By precisely engineering the light-matter interaction, the memory coherently stores the optimal temporal mode with high fidelity and retrieves it on demand, achieving mode crosstalk as low as 0.34%. To estimate the separation between two spectral lines, we experimentally measure the mean squared error of the frequency estimate, reaching a sensitivity of 1/20 of the linewidth and a ($34\pm4$)-fold enhancement in precision over direct intensity measurements. This enhanced frequency resolution, combined with on-demand storage, retrieval, and mode-conversion capabilities, establishes a pathway toward multifunctional memory-based time-frequency sensors and their integration within quantum networks.

quant-ph↗

arXiv2026

Extremalization approach to black hole thermodynamics: perturbations around higher-derivative gravities

When higher-derivative terms are added to a gravitational action, black hole solutions and their thermodynamic properties are generally corrected. Recent progress has shown that, by treating higher-derivative operators as perturbations, the first-order corrections to black hole thermodynamics can be obtained without explicit knowledge of the corresponding perturbed black hole solutions. This result can be understood as a consequence of an extremalization principle underlying the Euclidean action formulation of black hole thermodynamics. In this paper, we emphasize that this extremalization approach is not restricted to perturbations around Einstein gravity. Instead, it can be applied to perturbations of more general higher-derivative gravity theories whose black hole solutions are already known and can be taken as the zeroth-order background. As an explicit illustration, we consider Einstein--Gauss--Bonnet gravity as the zeroth-order theory and study the first-order thermodynamic corrections induced by further higher-order curvature operators. We show that these corrections can be derived without solving the perturbed black hole solutions, both in asymptotically flat and asymptotically AdS spacetimes.

gr-qc↗

arXiv2026

Quantum gravitational corrections to Reissner-Nordström black hole thermodynamics and their implications for the weak gravity conjecture

In this paper, we investigate the quantum gravitational corrections to the thermodynamical quantities of Reissner-Nordström black holes within the framework of effective field theory. The effective action originates from integrating out massless particles, including gravitons, at the one-loop level. We perform a complete thermodynamic analysis for both non-extremal and extremal black holes, and are mainly concerned about the shift in the charge-to-mass ratio $q/M$ that plays an important role in analyzing the weak gravity conjuecture. For non-extremal black holes, we identify a relationship between the shift in the charge-to-mass ratio and the thermodynamic stability of the black holes. For extremal black holes, we show that quantum gravity effects naturally lead to the super-extremality $q/M>1$ of charged black holes.

gr-qc↗

arXiv2026

Improving the loss threshold for quantum advantage in photonic sensors by complete photon counting

Tolerance to imperfections is a defining performance criterion for quantum sensors. The threshold for achieving a quantum advantage depends on the input state, sensor configuration, detection scheme, and, critically for optical platforms, photon loss. We consider a nonlinear interferometer in which two gain-optimized parametric nonlinear optical processes couple the state to the internal sensor and subsequently mix the reference and sensor beams. We demonstrate that measuring the full photon-number output statistics of this setup yields marked improvements in the loss threshold. Using photon-number-resolving detection (PNRD) based on transition-edge sensors (TESs), we experimentally reconstruct the joint photon-number statistics at the interferometer output. Subject to internal and external losses of approximately 25 % and 45 %, respectively -- and without any post-selection or loss correction -- we observe an unconditional violation of the shot-noise limit by $2.37 \pm 0.11$ dB. This translates to a 44 % enhancement in estimation precision over conventional click-detection strategies. We verify this performance by evaluating the classical Fisher information against both an analytical model of the joint photon-number distribution and the raw measured statistics. Ultimately, our results demonstrate that combining nonlinear interferometry with PNRD unlocks metrological information fundamentally inaccessible to click detectors, establishing a clear path toward practical, quantum-enhanced sensing under realistic loss conditions.

quant-ph↗

arXiv2026

Unveiling Hierarchical Invariants in Multiphoton Linear Optics

Linear optical networks driven by quantum states of light are important building blocks of photonic quantum technologies. They access large bosonic Hilbert spaces through multiphoton interference. At the same time, their dynamics are generated by single-particle mode transformations, thereby defining a highly structured subset of multiphoton unitaries and setting boundary on linear optics capability. To elucidate this boundary, we reveal an underlying fine-grained symmetry structure that partitions the multiphoton operator space into invariant subspaces and generates a hierarchy of invariants. We experimentally confirm the conservation of high-order invariants and demonstrate their operational utility in characterizing state reachability and the metrological capability of multiphoton probes. Our framework provides a symmetry-based perspective for understanding and harnessing structured multiphoton dynamics across photonic quantum technologies.

quant-ph↗

arXiv2026

Apple Intelligence Foundation Language Models

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

cs.AI↗

arXiv2026

Passive Imaging with Quantum Advantage

Far-field optical imaging inevitably involves low-pass spatial filtering, limiting the resolution. Moreover, conventional imaging suppresses high spatial frequency components close to the cutoff, making them invisible under noise, particularly the shot noise arising from discrete and random nature of quantum light. Here we propose and implement a method for reducing the effect of this noise by optically pre-processing the incoming light prior to detection, thereby optimizing the quantum measurement performed on it. Our scheme, termed Fourier Domain Division (FDD), partitions the Fourier plane into multiple regions for independent detection and subsequent post-processing for image reconstruction. By analyzing the quantum and classical Fisher information, we show that our method is advantageous with respect to direct imaging for high spatial-frequency components. As a result, the number of photons required to achieve a certain signal-to-noise-ratio in the Fourier domain is reduced, thus enhancing the overall resolution in the photon-starved regime. We demonstrate our method in microscopy, achieving 5-fold improvement of Fisher information on high spatial-frequency components. Unlike active super-resolution methods, FDD is passive, making it broadly applicable in microscopy and other imaging scenarios where active illumination is impractical, including astronomy and remote sensing. Our work establishes a general strategy for designing quantum optimized superresolution imaging systems, bridging fundamental quantum limits, practical image analysis and computer vision applications.

quant-ph↗

arXiv2026

An Atomic Interface for High-Dimensional Temporal Mode Quantum Networks

Temporal modes of photons are a promising encoding scheme for high-dimensional quantum networks due to their high channel capacity and fiber compatibility. However, realizing their full potential requires devices capable of synchronizing, processing and interfacing these modes across photonic and atomic bandwidths. In this work, we demonstrate a programmable high-dimensional temporal mode processor using a Raman quantum memory in warm cesium vapor. We exploit the single-mode nature of the Raman interaction kernel, dynamically shaping the control field to synthesize a tunable coherent filter that selectively addresses specific temporal waveforms. This mechanism enables on-demand storage, filtering, and conversion, providing a coherent interface between MHz- and GHz-bandwidth modes. We validate the platform's selectivity across a basis of 30 orthogonal Hermite-Gaussian modes and certify high-fidelity quantum operation via 5-dimensional process tomography. By combining deterministic mode conversion with bidirectional bandwidth interfacing, we establish the Raman memory as a critical active node for scalable quantum information processing.

quant-ph↗

arXiv2026

Machine learning of quantum data using optimal similarity measurements

Quantum machine learning seeks a computational advantage in data processing by evaluating functions of quantum states, such as their similarity, that can be classically intractable to compute. For quantum advantage to be possible, however, it is essential to bypass costly characterisation of individual data instances in favour of efficient, direct similarity evaluation. Here we demonstrate a sample-optimal, hardware-efficient protocol for estimating quantum similarity -- the state overlap -- using bosonic quantum interference. The sample complexity of this approach is independent of the system dimension and is information-theoretically optimal up to a constant factor. Experimentally, we implement the scheme on \emph{Prakash-1}, a quantum computing platform based on a fully programmable integrated photonic processor. By preparing and interfering qudit states on the chip to directly extract their overlap, we demonstrate classification and online learning of quantum data with high accuracy in realistic noisy experiments. Our results establish joint overlap measurements as a scalable pathway to efficient quantum data analysis and a practical building block for network-integrated quantum machine learning.

quant-ph↗

arXiv2026

Validity of the Background Subtraction Method for Black Hole Thermodynamics in Matter-Coupled Gravity Theories

The background subtraction method has long served as a practical tool for computing the Euclidean action and thermodynamic quantities of black holes. While its equivalence to the Iyer--Wald formalism is well understood in pure gravity theories, its validity in matter-coupled theories remains less clear and has even been questioned in the literature. In this work, we revisit this issue and demonstrate that the equivalence between the Euclidean action method and the Iyer--Wald formalism persists in matter-coupled scenarios. We apply the resulting formulation to two representative examples of such theories, and in both cases, the Euclidean approach performs smoothly. We further identify situations where the method may encounter subtleties due to the special properties of certain matter fields. Our results clarify when background subtraction remains reliable beyond pure gravity and when additional care is necessary.

gr-qc↗

arXiv2025

Synthetic bootstrapped pretraining

We introduce Synthetic Bootstrapped Pretraining (SBP), a language model (LM) pretraining procedure that first learns a model of relations between documents from the pretraining dataset and then leverages it to synthesize a vast new corpus for joint training. While the standard pretraining teaches LMs to learn causal correlations among tokens within a single document, it is not designed to efficiently model the rich, learnable inter-document correlations that can potentially lead to better performance. We validate SBP by designing a compute-matched pretraining setup and pretrain a 3B-parameter and a 6B-parameter model on up to 1T tokens from scratch. We find SBP consistently improves upon a strong repetition baseline and delivers up to 60% of performance improvement attainable by an oracle upper bound with access to 20x more unique data. Qualitative analysis reveals that the synthesized documents go beyond mere paraphrases -- SBP first abstracts a core concept from the seed material and then crafts a new narration on top of it. Besides strong empirical performance, SBP admits a natural Bayesian interpretation: the synthesizer implicitly learns to abstract the latent concepts shared between related documents.

cs.CL↗

arXiv2025

A universal relation among Euclidean integrals for black holes in higher-derivative gravity theories

In this paper, we establish a universal equality governing Euclidean integrals of gravitational actions in higher-derivative theories. This relation is shown to hold universally for asymptotically flat black holes in pure gravity, and is generalized to asymptotically anti-de Sitter (AdS) spacetimes through appropriate regularization. We further examine its validity in systems with matter-gravity coupling, identifying that violations occur only when matter fields exhibit pathological behaviors: divergence at the horizon or non-decaying profiles at infinity. These findings reveal fundamental constraints on gravitational thermodynamics and provide diagnostic tools for identifying ill-behaved matter configurations.

gr-qc↗

arXiv2025

Enhancing Quantum Memories with Light-Matter Interference

Future optical quantum technologies, such as quantum networks, distributed quantum computing and sensing, demand efficient, broadband quantum memories. However, achieving high efficiency without introducing noise, reducing bandwidth, or limiting scalability remains a challenge. Here, we present a new approach to enhance quantum memory protocols by leveraging constructive light-matter interference, leading to an increase in memory efficiency without increasing atomic density or laser intensity. We implement this method in a Raman quantum memory in warm Cesium vapor, and achieve more than a three-fold improvement in total efficiency reaching $(34.3\pm8.4)\%$, while retaining GHz-bandwidth operation and low noise levels. Numerical simulations predict that this approach can boost efficiencies in systems limited by atomic density, such as cold atomic ensembles, from $65\%$ to beyond $96\%$, while in warm atomic vapors it could reduce the laser intensity needed to reach a given efficiency by over an order-of-magnitude, exceeding $95\%$ total efficiency. Furthermore, our method preserves the single-mode nature of the memory at high efficiencies. This new protocol is applicable to various memory architectures, paving the way toward scalable, efficient, low-noise, and high-bandwidth quantum memories.

quant-ph↗

arXiv2025

Apple Intelligence Foundation Language Models: Tech Report 2025

We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: i a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and ii a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple's Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users' privacy with innovations like Private Cloud Compute.

cs.LG↗

arXiv2025

Topological network analysis using a programmable photonic quantum processor

Understanding topological features in networks is crucial for unravelling complex phenomena across fields such as neuroscience, condensed matter, and high-energy physics. However, identifying higher-order topological structures -- such as $k$-cliques, fundamental building blocks of complex networks -- remains a significant challenge. Here we develop a universal programmable photonic quantum processor that enables the encoding of arbitrary complex-weight networks, providing a direct pathway to uncovering their topological structures. We demonstrate how this quantum approach can identify weighted $k$-cliques and estimate Betti numbers by leveraging the Gaussian boson sampling algorithm's ability to preferentially select high-weight, dense subgraphs. The unique capabilities of our programmable quantum processor allow us to observe topological phase transitions and identify clique percolation phenomena directly from the entropy of the sampling results. These findings showcase how photonic quantum computing can be applied to analyse the topological characteristics of real-world complex networks, opening new possibilities for quantum-enhanced data analysis.

quant-ph↗

arXiv2025

Step-by-Step Reasoning for Math Problems via Twisted Sequential Monte Carlo

Augmenting the multi-step reasoning abilities of Large Language Models (LLMs) has been a persistent challenge. Recently, verification has shown promise in improving solution consistency by evaluating generated outputs. However, current verification approaches suffer from sampling inefficiencies, requiring a large number of samples to achieve satisfactory performance. Additionally, training an effective verifier often depends on extensive process supervision, which is costly to acquire. In this paper, we address these limitations by introducing a novel verification method based on Twisted Sequential Monte Carlo (TSMC). TSMC sequentially refines its sampling effort to focus exploration on promising candidates, resulting in more efficient generation of high-quality solutions. We apply TSMC to LLMs by estimating the expected future rewards at partial solutions. This approach results in a more straightforward training target that eliminates the need for step-wise human annotations. We empirically demonstrate the advantages of our method across multiple math benchmarks, and also validate our theoretical analysis of both our approach and existing verification methods.

cs.LG↗

arXiv2025

Experimental benchmarking of quantum state overlap estimation strategies with photonic systems

Accurately estimating the overlap between quantum states is a fundamental task in quantum information processing. While various strategies using distinct quantum measurements have been proposed for overlap estimation, the lack of experimental benchmarks on estimation precision limits strategy selection in different situations. Here we compare the performance of four practical strategies for overlap estimation, including tomography-tomography, tomography-projection, Schur collective measurement and optical swap test using photonic quantum systems. We encode the quantum states on the polarization and path degrees of freedom of single photons. The corresponding measurements are performed by photon detection on certain modes following single-photon mode transformation or two-photon interference. We further propose an adaptive strategy with optimized precision in full-range overlap estimation. Our results shed new light on extracting the parameter of interest from quantum systems, prompting the design of efficient quantum protocols.

quant-ph↗