Quantum diffusion models typically parameterize forward corruption by raw channel strength, even though equal parameter increments need not erase equal information or induce comparable inverse problems. We introduce the classical–quantum information decrement Δ_t=I(X:Q_t-1)-I(X:Q_t) as an intrinsic diffusion coordinate for labeled quantum ensembles. Along depolarization, equalizing Δ_t yields the unique minimax discretization of the forward path, while universal recoverability gives the same quantity an operational reverse interpretation as an attainable expected log-fidelity budget for a label-independent CPTP recovery channel. Complementary continuity and pairwise-geometric converses lower-bound the optimal common-channel recovery error. We further show that local calibration is fundamentally insufficient for stochastic generation: even in a fixed noncommuting two-qubit system, identical local-fidelity laws and budget-feasible risks can coexist with macroscopically different output distributions. This motivates a stochastic learner combining theorem-scaled recovery constraints with distribution matching, for which we establish finite-sample calibration and compositional trace-Wasserstein control. On four-qubit TFIM, a controlled capacity extension reduces endpoint from .622 to .424 on all ten matched seeds and achieves lower than official QuDDPM (.498) with fewer trainable parameters.

We study covert block-activity information transmission over a thermal-loss bosonic channel, where messages are encoded in weight-constrained activity patterns that Bob must recover through block-local decisions by prescribed deadlines. Shared circular Gaussian displacement modulation makes Willie's averaged lost-light state exactly thermal, with a strictly convex relative-entropy cost in signal energy, whereas a fixed Gaussian receiver at Bob yields a Gaussian mean-shift divergence linear in energy. This asymmetry produces an exact finite-block energy–information frontier and a detector-independent latency converse, matched in order by a block-reset cumulative-sum detector. For block covertness budget δ_b and error target ε_b, the required active length scales as δ_b^-1log^2(1/ε_b) up to an explicit channel–receiver factor. Lifting this local law to communication yields the matching transmission limit log M=Θ(√(n)/log n) for the symmetric coordinatewise architecture under a fixed total covertness budget, maximal-message covertness, vanishing maximal error, and local deadlines. A relaxed full-horizon reference with the same modulation family and fixed measurement supports Θ(√(n)), showing that covertness and the selected measurement alone do not impose the extra logarithmic factor. Finally, public quadrature phase-shift keying codebooks selected by O(√(n)) secret bits remove ideal continuous shared randomness without changing the payload order.

Forecast accuracy is often used as a proxy for temporal structure discovery, but predictive performance and structural identifiability are not equivalent. Different temporal mechanisms can achieve similar forecast errors, while small forecast differences may still contain sufficient information for recovery. In this work, we study when forecast-only structural selection can be trusted. We show that a vanishing forecast margin does not necessarily imply structural ambiguity, and establish a stability perspective that evaluates structural separation relative to uncertainty in the selection objective. This perspective provides both a sufficient condition for reliable selection and a continuous measure of selection difficulty. Experiments across controlled and end-to-end settings demonstrate that forecast margin alone is insufficient, while the proposed stability measure better characterizes when forecast-based structural selection succeeds or fails. Our results suggest that predictive accuracy should be treated as evidence for structure discovery only when its separation is sufficiently robust.

Abstract Artificial intelligence (AI) is increasingly permeating the drug development pipeline. Numerous algorithms for accelerating this multi-stage and multi-task process have been constructed, which depends heavily on expert design and labor-intensive task-specific optimization. Given that AI-driven acceleration of drug development is recognized as a cumulative, often synergistic, effect across multiple stages, the autonomous evolution of existing algorithms across the entire pipeline is demanded to achieve a holistic advancement. Here, we present DrugEvolve, a multi-role large language model system for systematic and autonomous algorithm evolution in drug development. DrugEvolve realizes a closed-loop evolution process by incorporating Researcher , Engineer , and Analyst domains, and enables an iterative design, implementation, evaluation, and refinement of algorithm by leveraging scientific knowledge and accumulated evolutionary experience. Across eleven representative tasks spanning target identification, drug discovery, preclinical study, and clinical trial, DrugEvolve autonomously evolved the corresponding task-specific algorithms and achieved substantial performance enhancement on 120 benchmark test sets. Moreover, it showed robust generalizabilities across heterogeneous data modalities (ranging from biological sequence and graph to molecular topology and textual language), and realized gains in both predictive and generative tasks. Collectively, this AI system can serve not only as an algorithmic infrastructure for drug development, but also as a transferable paradigm for broader scientific domains.

Cross-resolution change detection aims to identify changes between bitemporal images acquired from heterogeneous platforms (e.g., drones and satellites) with varying spatial resolutions. It plays a vital role in applications such as rapid disaster assessment. However, semantic inconsistencies caused by resolution differences and spatial misalignment resulting from imperfect registration pose significant challenges to robust change detection. To address these issues, this paper proposes a Semantic-Aware Alignment Network for cross-resolution change detection. For spatial alignment, a semantic-aware Transformer is introduced to establish global semantic correspondences and estimate geometric transformation parameters between the two images. To enforce semantic consistency, the high-resolution image is used as a reference and degraded to simulate low-resolution observations. By enforcing feature-level consistency between the high-resolution and low-resolution representations, the two inputs are mapped into a unified, resolution-invariant feature space, thereby substantially reducing semantic discrepancies. Benefiting from the joint semantic consistency learning and spatial alignment strategies, the proposed method outperforms existing state-of-the-art approaches, achieving F1-scores of 77.99%, 90.37%, and 51.40% on the HTCD, MRCDD, and DECD datasets, respectively.

Quantum diffusion models provide a physics-consistent route to generative learning by formulating noising and denoising directly on quantum states. However, applying such models to classical high-dimensional data is constrained by the qubit cost of state encoding and the computational burden of simulating large density operators. We propose a scalable hybrid generative pipeline that combines a classical autoencoder for dimensionality reduction with a mixed-state quantum denoising diffusion probabilistic model (MSQuDDPM) operating in the learned latent space. The autoencoder compresses data into compact latent codes that can be embedded into a small-qubit Hilbert space, after which the quantum diffusion model learns a generative distribution over latent density operators and decodes samples back to the original domain. Algorithmically, we simplify the reverse dynamics by predicting an estimate of the clean state ρ_0 at timestep t and computing the one-step reverse update via an analytic backward propagation rule, rather than learning an explicit predictor for ρ_t-1. We demonstrate the proposed approach on MNIST image generation and discuss how mixed-state quantum diffusion can serve as a practical backbone for hybrid quantum–classical generative modeling under realistic qubit budgets.

Fine-grained remote sensing datasets often use hierarchical label structures to differentiate objects in a coarse-to-fine manner, with each object annotated across multiple levels. However, embedding this semantic hierarchy into the representation learning space to improve fine-grained detection performance remains challenging. Previous studies have applied supervised contrastive learning at different hierarchical levels to group objects under the same parent class while distinguishing sibling subcategories. Nevertheless, they overlook two critical issues: (1) imbalanced data distribution across the label hierarchy causes high-frequency classes to dominate the learning process, and (2) learning semantic relationships among categories interferes with class-agnostic localization. To address these issues, we propose a balanced hierarchical contrastive loss combined with a decoupled learning strategy within the detection transformer (DETR) framework. The proposed loss introduces learnable class prototypes and equilibrates gradients contributed by different classes at each hierarchical level, ensuring that each hierarchical class contributes equally to the loss computation in every mini-batch. The decoupled strategy separates DETR's object queries into classification and localization sets, enabling task-specific feature extraction and optimization. Experiments on three fine-grained datasets with hierarchical annotations demonstrate that our method outperforms state-of-the-art approaches.

The scarcity of labeled samples results in the challenge of small sample size in hyperspectral image (HSI) classification. Transfer learning offers hope for solving this problem. In cross-domain transfer learning, the source domain boasts abundant labeled training samples, whereas the target domain comprises only limited labeled training samples. Leveraging the information from the source domain can benefit the classification of the target domain. However, inconsistencies in land-cover classes between source and target domains may hinder knowledge transfer between domains. Fortunately, few-shot learning (FSL) provides an effective solution to this challenge. In recent years, meta-learning has gained widespread attention as a mainstream approach within FSL. This article proposes a novel method for cross-domain heterogeneous HSI classification, called cross-domain meta-learning with feature alignment (CD-MFA). CD-MFA enhances the generalization performance of the inner-loop optimization by incorporating task-adaptive loss function. The adaptive weighting strategy (AWS) is used in the outer-loop optimization to balance the classification losses of the source and target domains to learn more discriminative features. Additionally, by aligning the features of the source and target domains under the guidance of the Gaussian prior, the impact of domain shift can be mitigated. It is worth noting that CD-MFA is trained concurrently on both the source and target domains so that the two domains are well bound, thereby enhancing the effectiveness of knowledge transfer. Experimental results on four publicly available HSI datasets validate the effectiveness of CD-MFA.

Hierarchical classification (HC) under real-world degradations (noise, occlusion, blur, low resolution) often hurts fine-grained accuracy and amplifies error propagation. We propose UGHM-Net, an HC framework that integrates uncertainty-quantification (UQ)-driven dynamic confidence adjustment, hierarchical backtracking, and a joint confidence–multi-level loss. UGHM-Net leverages a Vision Transformer (ViT) for global feature extraction and a Semantic-Guided Cross-Attention (SGCA) module for visual–semantic fusion between visual tokens and static label embeddings. Guided by UQ, the model adaptively selects reliable hierarchical paths and backtracks to higher levels when fine-level confidence is low, curbing over-classification and reducing cascading errors. The joint objective balances coarse- and fine-grained supervision. Experiments on degraded remote sensing and natural image datasets show that UGHM-Net consistently outperforms prior methods in ISDL, hierarchical precision (PH), hierarchical recall (RH), and level-wise accuracy, providing a robust and efficient HC solution via UQ-driven path control and visual–semantic fusion.

Structure-based virtual screening (VS) via molecular docking is a pivotal approach for hit identification. Many artificial intelligence (AI)-powered protein-ligand docking and scoring methods have demonstrated impressive speed and accuracy. Retrospective benchmarking studies using enrichment rate and computational efficiency on curated datasets have corroborated their potential for discovering bioactive compounds. However, determining which method suits a specific application and implementing it efficiently remains challenging. Here we present the Comprehensive VS Platform with AI Engine (CVSP-AIE) for drug discovery from compound libraries. It integrates three AI models: KarmaDock, a fast docking model that directly updates atomic coordinates; CarsiDock, an accurate docking model that predicts protein-ligand distances and reconstructs binding poses; and RTMScore, an accurate scoring model that learns residue-atom distance distributions for affinity prediction. Their hierarchical application enables dynamical balances in screening speed and accuracy. CVSP-AIE is available as an online web server ( https://cadd.zju.edu.cn/cvsp/ ) and a local software package. Users can efficiently initiate drug screening by uploading a protein and a known binder that defines the binding pocket. The following workflow involves (1) preprocessing, including protein structure repair and molecule standardization, (2) binding pose and affinity prediction powered by KarmaDock, CarsiDock and RTMScore and (3) postprocessing, comprising protein-ligand interaction calculation and visualization. It takes 30-45 min to hierarchically screen 100,000 compounds, and the output is a ranked list of molecules with predicted binding scores, intermolecular interaction profiles and interactive chemical space analysis. Users can also install locally the hierarchical screening module through command-line package for arbitrary-scale screening.

Hierarchical classification (HC) on degraded images presents challenges due to feature corruption, unreliable confidence estimation, and fine-grained misclassification. Existing methods often struggle to balance semantic consistency and adaptive decision paths under low-quality visual conditions. To address this, we propose HierUQ, a unified framework that integrates uncertainty quantification with adaptive granularity reconciliation. A Vision Transformer backbone extracts global features, which are fused with semantic embeddings via bilinear and semantic-guided cross-attentions. We develop a principled Hierarchical Uncertainty Quantification (HUQ) strategy based on label smoothing and proper scoring rules. When confidence is insufficient, a Confidence-Aware Path Adjustment (CAPA) mechanism adaptively rolls back predictions to higher-level nodes, mitigating overclassification and error propagation, stabilizing the learning trajectory, overcoming degradation-induced interference, and enhancing fine-grained classification accuracy. To enhance learning, we introduce a self-paced joint optimization (MLJO) over multi-level objectives with dynamic loss weighting. Experiments on degraded remote sensing and natural image benchmarks show that HierUQ achieves state-of-the-art performance with strong robustness and adaptability.

Spectral reconstruction aims to recover hyperspectral images (HSIs) from RGB inputs, a challenging task due to severe spectral downsampling. To address this, we propose leveraging spatial structures common to both RGB images and HSIs. Specifically, we introduce a spectral autoencoder that compresses and aligns spectral data with the RGB space, and incorporate spatial autoencoders from pre-trained RGB latent diffusion models (RGB LDMs) to capture robust spatial representations. This disentangled spectral-spatial representation enables more efficient learning and embeds both RGB images and HSIs into a shared latent space, facilitating effective cross-modal representation transfer. In this latent space, the RGB LDM is fine-tuned to learn the spectral-spatial joint distribution for reconstruction, preserving spatial information from the RGB LDM. During inference, the spectral-spatial features are estimated by sampling from this learned distribution and recovered using both autoencoders. Experiments demonstrate that our method outperforms state-of-the-art approaches, with additional tests on relighting tasks and ablation studies showcasing the practical benefits of RGB diffusion priors. The code is available at https://github.com/Dmsw/HS-SR-LDM.

Cross-domain hyperspectral image (HSI) classification algorithms utilize the information from the dataset with abundant samples (source domain, SD) to enhance the classification accuracy of the dataset with only a few samples (target domain, TD). This paper proposes a feature disentanglement and adaptation network (FDANet) for cross-domain HSI classification. A cross-domain mask mechanism (CDMM) is proposed to disentangle the domain-invariant features and domain-specific features from the original features. For the domain-invariant features, we design an attention-based cross-domain fusion (ACDF) module to obtain a hybrid representation. Moreover, a prediction consistency constraint (PCC) is introduced to ensure that the domain-invariant features have shared and discriminative information. Experiments on three real-world heterogeneous cross-domain HSI datasets show the effectiveness of the proposed method.

Existing remote sensing ship datasets typically organize fine-grained categories into a hierarchical structure by introducing superclasses at coarser levels. Hierarchical classification aims to encode semantic relationships in the label hierarchy to improve model performance across multiple granularity levels. However, recent studies mainly focus on parent-child or sibling classes, capturing only local semantic relationships within the label hierarchy. In this paper, we propose a novel loss function that formalizes the semantic distance between the ground truth node and all other nodes in the label hierarchy as the expected margin to separate hierarchical categories. The semantic distance, defined as the number of edges along the path connecting two nodes, is incorporated as a penalty term to adjust the relative magnitudes of class logits. Experiments on two widely used ship datasets demonstrate that our approach consistently outperforms state-of-the-art methods across all hierarchical levels.

Hyperspectral videos (HSVs) offer greater potential for object tracking than color videos due to their ability to identify materials through spectral information. With the development of sensing, HSVs are captured by different cameras across various domains, necessitating a universal representation that can be applied across all domains to enhance discriminative power. Existing methods either train separate models for each domain or co-train a single model across multiple domains. These approaches often fail to fully exploit both domain-shared and domain-specific properties, hindering the effective learning of a universal representation. Furthermore, the limited availability of multi-domain HSVs poses additional challenges for training robust universal representation. To address these challenges, this paper introduces DaSSP-Net, a universal representation learning framework for multi-domain hyperspectral tracking based on domain adaptation and spatial-spectral prompt learning. Specifically, we leverage a pre-trained foundation model trained on large-scale color videos and introduce a band-invariant spatial-spectral prompt module that guides the foundation model in learning universal representation across all domains. This spatial-spectral prompt module has less than 1% of model parameters to capture global spatial-spectral interactions between hyperspectral and false-color images. By spatial-spectral prompt learning, we leverage the robustness and generalizability of the foundation model along with the material identification capabilities of HSVs, addressing the issue of data scarcity in HSVs. Furthermore, to mitigate performance trade-offs that can arise from using a single universal model, we introduce a domain adapter module that learns the domain-specific distribution for each domain. Extensive experiments demonstrate that our DaSSP-Net achieves state-of-the-art tracking performance. The source code will be available at https://github.com/hscv/DaSSP-Net.

Ship targets in remote sensing images can be categorized at various granularities due to variations in image quality, ranging from general ship categories to fine-grained classes like Nimitz (NT)-class carriers. Traditional studies mainly focus on fine-grained ship classification (FGSC), often neglecting samples observed at coarser-grained levels. Samples distributed across multiple granularity levels exhibit semantic relationships among their annotated classes, enabling hierarchical knowledge transfer during model training. This article incorporates two semantic relationships into deep learning-based representation learning and class prediction: parent-child relationships across levels and mutual exclusivity among sibling categories. For hierarchical representation learning, the proposed hierarchical contrastive learning algorithm extracts category-specific representations from input images and aligns them with their semantic relationships, ensuring that parent and child categories share similarities while sibling categories remain distinct. For hierarchical class predictions, a novel consistency loss (CL) ensures coherence in probability distributions between parent and child categories. Specifically, cross-entropy (CE) loss is employed to impose mutual exclusivity among sibling categories. In this article, a multimodal dataset is also designedly developed for hierarchical classification, which integrates optical and synthetic aperture radar (SAR) images across multiple hierarchical levels. Experiments on two popular datasets and a multimodal dataset demonstrate that the proposed method outperforms state-of-the-art approaches in hierarchical multigranularity ship classification.

Incremental learning is a strategy that continuously incorporates new data to tackle emerging tasks without the need for retraining the model. While effective, it encounters the challenge of catastrophic forgetting. Hierarchical Classification (HC) enhances classification accuracy and efficiency by assigning objects to multiple labels within a hierarchical structure. This paper introduces LRGIC, a novel approach specifically designed for incremental hierarchical classification of remote sensing images. LRGIC combines class hierarchy (CH), a Feature Pyramid Network (FPN), and a Learning Without Forgetting (LWF) strategy, while also incorporating a HEX graph to constrain labels and encode hierarchical knowledge. These elements, integrated within a hierarchical residual network, significantly boost classification performance. The FPN captures multi-scale features, and the LWF strategy facilitates the learning of new categories without reusing old samples. Experimental results demonstrate that LRGIC effectively classifies new categories and their hierarchical relationships, while preserving the performance of existing categories, underscoring its substantial research and practical value.

Remote sensing categories can be organized into a hierarchical label structure, with objects distributed across multiple granularity levels due to substantial differences in resolution and data modalities. However, traditional hierarchical classification performs mandatory leaf-node prediction and cannot stop at intermediate levels when predicting unseen or uncertain targets. In contrast, multigranularity classification (MGC) can infer targets at any class in the label hierarchy, making it suitable for predicting multigranularity remote sensing objects. To the best of our knowledge, this is the first study to address the MGC challenge in remote sensing images. Previous studies mainly focus on selecting an optimal threshold for stopping prediction. However, such fixed thresholds cannot flexibly adapt to different tasks and scenarios. This article proposes an accuracy-specificity curve that measures the probability distribution over the label hierarchy and enables the adaptive selection of task-dependent thresholds. The accuracy-specificity curve first ranks hierarchical labels by probability and information content and then evaluates their accuracy and specificity at different thresholds. In top-down prediction along a label tree, increasing label specificity improves information richness but also raises prediction risk, leading to reduced accuracy. Lowering the threshold increases specificity but reduces accuracy, enabling the selection of the appropriate threshold based on the required accuracy or specificity. Moreover, a novel loss function models the semantic distance between the ground-truth node and other nodes in the label hierarchy as a penalty term, adjusting the relative magnitudes of class logits. Experiments on three hierarchical ship datasets show that the proposed method outperforms state-of-the-art methods using conventional and proposed metrics.

Advancements in artificial intelligence (AI) and emerging technologies are rapidly expanding the exploration of chemical space, facilitating innovative drug discovery. However, the transformation of novel compounds into safe and effective drugs remains a lengthy, high-risk, and costly process. Comprehensive early-stage evaluation is essential for reducing costs and improving the success rate of drug development. Despite this need, no comprehensive tool currently supports systematic evaluation and efficient screening. Here, we present druglikeFilter, a deep learning-based framework designed to assess drug-likeness across four critical dimensions: 1) physicochemical rule evaluated by systematic determination, 2) toxicity alert investigated from multiple perspectives, 3) binding affinity measured by dual-path analysis, and 4) compound synthesizability assessed by retro-route prediction. By enabling automated, multidimensional filtering of compound libraries, druglikeFilter not only streamlines the drug development process but also plays a crucial role in advancing research efforts towards viable drug candidates, which can be freely accessed at https://idrblab.org/drugfilter/.

Object tracking with hyperspectral videos (HSVs) offers significant advantages due to the captured spectral fingerprint information, which provides detailed physical material characteristics. While correlation filter (CF)-based tracking methods align well with the high-dimensional nature of HSVs, they often fall short of fully utilizing the spatial-spectral-temporal structure inherent in these data. In this article, we introduce a spatial-spectral-temporal CF (SSTCF) framework to address these limitations. SSTCF employs the spatial-spectral histogram of gradients and fractional abundances as features to characterize the spatial-spectral structure of the object. A low-rank constraint is integrated into the CF framework to enhance the global spectral semantic dependencies among learned filters. In addition, a temporal constraint is incorporated to ensure filter consistency across consecutive frames, further improving tracking continuity between nearby frames. Extensive experiments demonstrate that our SSTCF tracker achieves more accurate and stable performance. The source code will be publicly available at https://github.com/bearshng/SSTCF