国家预印本平台

Topological data analysis (TDA) provides a powerful framework for extracting information about the shape of complex, unstructured data, but the classical cost of computing high dimensional topological features limits its application. Quantum algorithms for TDA offer a route around this bottleneck, yet existing approaches typically focus on exact or high precision Betti number estimation, making the regime for practical quantum advantage appear narrow. Here, we instead frame quantum TDA as a feature-extraction method for downstream data analysis by extracting low-order spectral information from the combinatorial Laplacian as a proxy for high-dimensional topology. We support this perspective from both the application and algorithmic sides. First, we show that higher-order TDA features improve predictive performance in two time-series applications: functional MRI analysis for neurodegenerative disease classification and financial time-series analysis for identifying market instability. Second, we develop a moment-based quantum algorithm and show that low-order moments, including the relative trace, are strongly correlated with high-dimensional Betti information, even when the relative Betti number is small. Finally, we present circuit constructions, resource estimates, quantum-classical crossover projections, and experimental results from a Barium development system similar to the forthcoming IonQ Tempo line, extracting Laplacian-derived observables from graph instances and quantitatively comparing them with exact Betti information. Together, these results establish quantum TDA as a practical approach for extracting topological features from classically challenging data

Pre-training followed by fine-tuning has become the dominant recipe for learning performant policies, and in value-based reinforcement learning (RL) this raises a natural question: given a pretrained policy, should the Q-function be pretrained on offline data too? Conventional wisdom suggests it should, but recent results show that online RL with a randomly-initialized Q-function can result in highly performant and reliable policies without needing to pretrain the Q-function. In this paper, we systematically study whether pretraining the Q-function actually helps when fine-tuning on top of a pretrained base policy. We find, surprisingly, that naive Q-function pretraining often provides little benefit over random initialization. We show this stems from a fundamental mismatch: the Q-function learned during pretraining targets the pretrained policy's Q-function, not the Q-function that online fine-tuning converges to, and this gap persists even after offline value maximization. Motivated by this finding, we propose Initialization via Policy Ensemble (IPE), a simple method that trains multiple diverse policies and uses their pooled rollouts to bootstrap the Q-function learning in online RL. Across a suite of challenging continuous control benchmarks, IPE yields an average 1.26x improvement in fine-tuning performance over naive Q-function pre-training.

Recently, a fit of charmless $B\to PP$ decays ($B \in \{B^0, B^+, B_s^0\}$, $P \in \{ π, K, η, η' \}$) to the latest data was performed under the assumption of flavour SU(3) symmetry [SU(3)$_F$]. It was found that there is a $4.1σ$ disagreement with the SU(3)$_F$ limit of the Standard Model [$\rm SM_{SU(3)_F}$]. In this paper, we extend this analysis to charmless $B \to VV$ decays ($V \in \{ρ, K^*, ϕ, ω\}$). The fit examining $B \to ρK^*$ decays, assuming only isospin symmetry, is found to be acceptable. When we fit to $B \to VV$ decays with $V \in \{ ρ, K^* \}$ within SU(3)$_F$, we find a $5.2σ$ discrepancy with SM$_{SU(3)_F}$. Finally, when $B \to VV$ decays with $V \in \{ ρ, K^*, ϕ, ω\}$ are considered, the discrepancy grows to $>7σ$. The theoretical input in this analysis is modest, so our results are quite rigorous, group theoretically, and hold almost exactly in the SU(3)$_F$ limit. Although it seems unlikely that the introduction of $\sim 30$% SU(3)$_F$-breaking effects can account for this discrepancy, this must be verified.

Modular quantum computing is a leading paradigm for scaling quantum computation beyond the resource limitations of monolithic devices. In this architecture, multiple quantum processing units (QPUs), employing identical or distinct qubit modalities, are interconnected via shared entanglement. Here, we investigate how errors at module interfaces and within individual QPUs affect fault-tolerant computation when qubits are encoded using the rotated surface code. Going beyond the logical-memory benchmark, we perform circuit-level simulations of fault-tolerant nonlocal CNOT gates implemented via lattice surgery between QPUs connected by noisy Bell pairs, and analyze the resulting logical error rates. Our results show that interfaces can tolerate noise up to an order of magnitude higher than intra-QPU noise, with only a minor reduction in the fault-tolerance threshold. We further develop an efficient protocol for preparing distributed fault-tolerant logical GHZ states, reducing ancilla overhead, time, and nonlocal Bell-pair consumption. We show that ancilla minimization in this setting is equivalent to a vertex-cover problem on an associated graph, and introduce a polynomial-time heuristic algorithm for finding low-overhead solutions. Our results provide quantitative evidence that distributed quantum error correction can enable scalable, fault-tolerant quantum computation in modular architectures.

World models enable a predictive substrate for planning and action, yet existing formulations merely answer a physical question: what/where it is, and how will it evolve. Human behavior, however, is driven by hidden mental state (what a person believes, wants, intends, feels, and considers socially permissible), so a model that tracks the physical scene but not what each agent knows and believes about it predicts the wrong action for the right-looking scene. We formulate Mental World Modeling (MWM), a generic theoretical framework that makes mental variables core components of a world model rather than posthoc rationales: MWM aintains a coupled physical-mental world state, renders a target-specific partial observation, and simulates how candidate actions jointly update both components. We instantiate the framework in MENTIS, a training-free and fully inspectable baseline that decomposes the process into state parsing, target-observation generation, action decomposition, coupled physical and mental transition, and branch-level value evaluation. On a manually constructed, quality-controlled dataset of situated decision scenarios spanning text, image, and sounding-video stories, experiments with 8 modern LLM-based world models demonstrate that explicitly modeling the mental state is essential for predicting human decisions. Deeper analyses further expose the bottlenecks of current mental world modeling. We expect MWM as a next stage of world modeling, from simulating physical scenes to simulating the minds that act in them.