With the rapid development of generative models and multimodal content editing technologies, the key challenge faced by synthetic image detection (SID) lies in cross-distribution generalization to unknown generation sources. In recent years, visual foundation models (VFM), which acquire rich visual priors through large scale image-text alignment pretraining, have become a promising technical route for improving the generalization ability of SID. However, existing VFM-based methods remain relatively coarse-grained in their adaptation strategies. They typically either directly use the final layer representations of VFM or simply fuse multi layer features, lacking explicit modeling of the optimal representational hierarchy for transferable forgery cues. Meanwhile, although directly fine-tuning VFM can enhance task adaptation, it may also damage the cross-modal pretrained structure that supports open-set generalization. To address this task specific tension, we reformulate VFM adaptation for SID as a joint optimization problem: it is necessary both to identify the critical representational layer that is more suitable for carrying forgery discriminative information and to constrain the disturbance caused by task knowledge injection to the pretrained structure. Based on this, we propose I2P, an SID framework centered on intrinsic importance perception. I2P first adaptively identifies the critical layer representations that are most discriminative for SID, and then constrains task-driven parameter updates within a low sensitivity parameter subspace, thereby improving task specificity while preserving the transferable structure of pretrained representations as much as possible.

Deep models are vulnerable to performance degradation caused by various factors, such as imbalanced samples, inaccurate labels, and backdoor attacks. However, existing optimization methods that address these issues are typically designed in a scenario- or architecture-specific manner, and each optimization often requires costly training. To this end, inspired by image denoising, we propose parameter purification as a new paradigm for model performance optimization. Parameter purification attributes performance degradation to the contamination of model parameters and aims to recover clean parameters from corrupted ones in a manner analogous to image denoising. To purify parameters with massive scale and complex structure, we further introduce a novel parameter manifold purification method. In this framework, high-dimensional and complex parameters are first viewed as manifolds embedded in a high-dimensional space, and are then partitioned into nested local parameter-cluster manifolds via a proposed parameter clustering strategy. Meanwhile, to remove parameter redundancy while preserving global parameter information, we propose an implicit manifold auto-encoder along with a parameter-cluster discrepancy loss to learn low-dimensional representations of parameter-cluster manifolds. Finally, an implicit conditional diffusion model is applied to denoise the low-dimensional parameter manifolds, progressively restoring clean parameters. Extensive experiments under three representative scenarios that cause model performance degradation demonstrate that parameter manifold purification can accurately and completely purify corrupted parameters of unseen models, analogous to denoising unseen images, and rapidly improve model performance.

Model reuse detection aims to determine whether a deep neural network is derived from a pre-trained model via transfer learning, fine-tuning and pruning. Traditional methods embed extra information into the model that often degrade performance on various downstream tasks. Recent neuron functionality-based approaches have achieved promising results. However, they either compute distances directly on neuron outputs (vulnerable to noise and lacking robustness) or require training auxiliary modules (incurring significant overhead) to accomplish the task of model reuse detection. To address these challenges, we propose the Signature Feature Sequence (SFS), which fingerprints a model by measuring the stability of its intermediate convolutional responses. Specifically, SFS generates a compact binary sequence by statistically analyzing the stability of feature maps from unified convolutional kernels, and detects provenance by comparing sequence similarity. Extensive experiments show that SFS matches State-of-the-Art accuracy.

Background and Aims: Postoperative recurrence remains a major cause of mortality after HCC resection, but current tumor-centric stratification tools overlook stromal biology. The tumor–stroma ratio (TSR) captures this dimension, yet its clinical application is limited by subjective scoring and an unclear biological basis. We aimed to automate whole-slide TSR quantification, validate its prognostic value across multicenter cohorts, and characterize the stromal–immune ecosystem it represents. Methods: A Laplace-Diffusion segmentation framework quantified TSR from routine H&E whole-slide images in multicenter HCC cohorts (n = 760). A TSR cutoff was derived in the training cohort and fixed for validation. Bulk transcriptomics, single-cell RNA sequencing, spatial transcriptomics, and multiplex immunofluorescence were integrated with exploratory functional validation in orthotopic HCC models treated with the FAK inhibitor defactinib. Results: The segmentation framework achieved a mean intersection over union of 0.759 and a Dice coefficient of 0.793. High TSR, defined using a training-derived cutoff of 23.31%, independently predicted shorter recurrence-free survival across cohorts (multivariable HR, 1.53–1.79; all P < 0.05). Multiomic profiling linked high TSR to extracellular matrix remodeling, stromal activation, and an APOE-positive tumor-associated macrophage-enriched immunosuppressive niche with reduced cytotoxic T-cell features. In orthotopic models, high-stroma tumors showed preferential growth attenuation with defactinib, accompanied by reduced collagen deposition. Conclusions: Automated TSR quantification provides a scalable, H&E-derived surrogate of a recurrence-associated stromal–immune ecosystem in HCC and may support perioperative risk stratification and future investigation of stromal-targeted therapies.

Human mobility prediction is crucial for applications ranging from location-based recommendations to urban planning, which aims to forecast users' next location visits based on historical trajectories. While existing mobility prediction models excel at capturing sequential patterns through diverse architectures for different scenarios, they are hindered by the long-tailed distribution of location visits, leading to biased predictions and limited applicability. This highlights the need for a solution that enhances the long-tailed prediction capabilities of these models with broad compatibility and efficiency across diverse architectures. To address this need, we propose the first architecture-agnostic plugin for long-tailed human mobility prediction, named Adaptive LOcation HierArchy learning (ALOHA). Inspired by Maslow's theory of human motivation, we exploit and explore common mobility knowledge of head and tail locations derived from human mobility trajectories to effectively mitigate long-tailed bias. Specifically, we introduce an automatic pipeline to construct city-tailored location hierarchies based on Large Language Models (LLMs) and Chain-of-Thought (CoT) prompts, capturing high-level mobility semantics with minimal human verification. We further design an Adaptive Hierarchical Loss (AHL) that rebalances learning through Gumbel disturbance and node-wise adaptive weighting, enabling both exploitation of multi-level signals and exploration within semantically related groups. Extensive experiments across multiple state-of-the-art models demonstrate that ALOHA consistently improves long-tailed mobility prediction performance by up to 16.59% while maintaining efficiency and robustness. Our code is at https://github.com/Star607/ALOHA.

Pathology images are crucial for cancer diagnosis and treatment. Although artificial intelligence has driven rapid advancements in pathology image analysis, the interpretation of ultra-large and multi-scale pathology images in clinical practice still heavily relies on physicians' experience. Clinicians need to repeatedly zoom in and out on individual slides to compare and assess pathological details - a process that is both time-consuming and prone to visual fatigue. The system first employs a diffusion model to perform tissue segmentation on pathology images, then calculates pathological tissue proportions and morphological metrics. Finally, through multi-scale dynamic comparison and multi-level visual evaluation, the system facilitates comprehensive and precise analysis of pathology images. The system provides clinicians with an intelligent and interactive tool for pathology image interpretation, enabling efficient visualization and precise analysis of pathological details, thereby reducing the effort require for detailed analysis.

Deep learning has shown strong potential for massive multiple-input multiple-output (Massive MIMO) physical-layer tasks, including channel state information (CSI) feedback and channel estimation. However, environmental heterogeneity can severely degrade CSI models in unseen scenarios, while conventional adaptation requires target-domain data and substantial computation. This paper proposes Channel Conditional Parameter Generation (CCPG), an end-to-end pipeline for rapid deployment of CSI models in dynamic wireless environments. CCPG identifies scene-sensitive adaptation bottlenecks through component-freezing experiments and generates only lightweight LoRA weights instead of full model parameters. It compresses high-dimensional channel features into compact latent conditions using cascaded SVD and a Perceiver Resampler. An energy-based canonicalization mechanism mitigates permutation and sign ambiguities in LoRA weights, while a diffusion-based generator incorporates structural information and an asymmetric size-aware loss for topology-aware parameter generation. Experiments on DeepMIMO and WAIR-D for CSI feedback and channel estimation show that CCPG adapts to new scenarios in about 3 seconds with a single forward pass, without target-scenario training or fine-tuning, and achieves cross-domain recovery performance comparable to costly online adaptation. These results demonstrate that CCPG enables efficient deployment of CSI models in large-scale dynamic wireless scenarios for intelligent 6G communications.

Real-time, high-fidelity monocular depth estimation from remote sensing imagery is crucial for numerous applications, yet existing methods face a stark trade-off between accuracy and efficiency. Although using Vision Transformer (ViT) backbones for dense prediction is fast, they often exhibit poor perceptual quality. Conversely, diffusion models offer high fidelity but at a prohibitive computational cost. To overcome these limitations, we propose Depth Detail Diffusion for Remote Sensing Monocular Depth Estimation (D³-RSMDE), an efficient framework designed to achieve an optimal balance between speed and quality. Our framework first leverages a ViT-based module to rapidly generate a high-quality preliminary depth map construction, which serves as a structural prior, effectively replacing the time-consuming initial structure generation stage of diffusion models. Based on this prior, we propose a Progressive Linear Blending Refinement (PLBR) strategy, which uses a lightweight U-Net to refine the details in only a few iterations. The entire refinement step operates efficiently in a compact latent space supported by a Variational Autoencoder (VAE). Extensive experiments demonstrate that D³-RSMDE achieves a notable 11.85% reduction in the Learned Perceptual Image Patch Similarity (LPIPS) perceptual metric over leading models like Marigold, while also achieving over a 40× speedup in inference and maintaining VRAM usage comparable to lightweight ViT models.

The rapid advancement of AI-generated content has made the reliable detection of generated images an increasingly critical challenge. Existing detection methods are often dominated during training by semantically salient components with high signal-to-noise ratios (SNRs), thereby suppressing subtler forensic cues associated with the underlying generation mechanisms and embedded in low-level statistical structures. From an information-theoretic perspective, we present a key insight: effective detection in the low-level statistical space requires mitigating the dominance of semantic components while emphasizing and amplifying responses to low-SNR forgery traces. Building on this insight, we propose RippleNet, an AI-generated image detection framework based on local differential signals. RippleNet adaptively identifies forgery-sensitive regions and constructs multi-directional, multi-scale differential representations within local neighborhoods, explicitly characterizing anomalous patterns in neighborhood statistics. More importantly, we refine the attention mechanism to operate within the local differential representation space, enabling the model to establish explicit dependencies at a finer statistical granularity. This design facilitates the capture of pixel-level forgery traces that are difficult to model using conventional convolutions or image-wide patch-level attention. Extensive experiments on multiple public benchmarks and under cross-generator evaluation settings demonstrate that RippleNet achieves consistently competitive performance.

Spatiotemporal vector retrieval has emerged as a critical paradigm in modern information retrieval, enabling efficient access to massive, heterogeneous data that evolve over both time and space. However, existing spatiotemporal retrieval methods are often extensions of conventional vector search systems that rely on external filters or specialized indices to incorporate temporal and spatial constraints, leading to inefficiency, architectural complexity, and limited flexibility in handling heterogeneous modalities. To overcome these challenges, we present a unified spatiotemporal vector retrieval framework that integrates temporal, spatial, and semantic cues within a coherent similarity space while maintaining scalability and adaptability to continuous data streams. Specifically, we propose (1) a Rotary-based Unified Encoding Method that embeds time and location into rotational position vectors for consistent spatiotemporal representation; (2) a Circular Incremental Update Mechanism that supports efficient sliding-window updates without global re-encoding or index reconstruction; and (3) a Weighted Interest-based Retrieval Algorithm that adaptively balances modality weights for context-aware and personalized retrieval. Extensive experiments across multiple real-world datasets demonstrate that our framework substantially outperforms state-of-the-art baselines in both retrieval accuracy and efficiency, while maintaining robustness under dynamic data evolution. These results highlight the effectiveness and practicality of the proposed approach for scalable spatiotemporal information retrieval in intelligent systems.

With the rapid advancement of AIGC technology, developing identification methods to address the security challenges posed by deepfakes has become urgent. Face forgery identification techniques can be categorized into two types: traditional classification methods and explainable VLM approaches. The former provides classification results but lacks explanatory ability, while the latter, although capable of providing coarse-grained explanations, often suffers from hallucinations and insufficient detail. To overcome these limitations, we propose EvolveReason, which mimics the reasoning and observational processes of human auditors when identifying face forgeries. By constructing a chain-of-thought dataset, CoT-Face, tailored for advanced VLMs, our approach guides the model to think in a human-like way, prompting it to output reasoning processes and judgment results. This provides practitioners with reliable analysis and helps alleviate hallucination. Additionally, our framework incorporates a forgery latent-space distribution capture module, enabling EvolveReason to identify high-frequency forgery cues difficult to extract from the original images. To further enhance the reliability of textual explanations, we introduce a self-evolution exploration strategy, leveraging reinforcement learning to allow the model to iteratively explore and optimize its textual descriptions in a two-stage process. Experimental results show that EvolveReason not only outperforms the current state-of-the-art methods in identification performance but also accurately identifies forgery details and demonstrates generalization capabilities.

Computational pathology has made progress in diagnosis and prognosis prediction from whole slide images (WSIs), yet pipelines still rely on patch-level feature extraction and aggregation, departing from the cell-centric reasoning used by pathologists. This gap limits sensitivity to micro-lesions and subtle changes, and current methods rarely provide a unified solution that supports both local and global tasks with trustworthy evidence. We propose Cello, a universal cell-wise feature aggregation framework for reliable pathology image analysis. Cello integrates cell-level representations into WSI modeling via protein-signal–supervised cell-wise learning, preserving fine-grained cellular cues under gigapixel constraints. For local tasks, Cello introduces a flexible prototype-based contrastive module for scalable, task-adaptive representation learning. For global tasks, Cello adopts a weakly supervised gated aggregation that can widely leverage WSI labels. Finally, a cell–local–global decision-route consistency objective dynamically aggregates cellular evidence and aligns local predictions with global outcomes, improving reliability and faithfulness. Trained with only hundreds to thousands of samples, Cello achieves performance gains of 3.0%~7.6% and outperforms SOTA pathology foundation models pretrained on tens of thousands of samples. Code is available at https://anonymous.4open.science/r/Cello.

Label-free image anomaly detection is difficult because anomalies must be separated from intra-normal variability. Diffusion models learn a manifold for normal data, and, under the common assumption that off-manifold anomalies are harder to generate and yield larger prediction errors, many methods build detectors from prediction residuals; yet reverse-process stochasticity and complex but normal structure also produce large residuals, so magnitude alone is non-diagnostic. To clarify what is recoverable from such noisy residuals, the theory examines how residual signals propagate through later reverse steps, showing that variability consistent with normal statistics is gradually absorbed toward stationarity, whereas anomalous regions retain an additional non-stationary signal that persists. Building on this insight, the Residual–Evolution Field (REF) isolates this persistent signal, with labeled source data calibrating the extractor and Cross-domain Field Alignment (CFA) transferring it to unlabeled targets. A theoretical framework with formal guarantees is established, and experiments across multiple benchmarks under substantial domain shifts demonstrate state-of-the-art performance, improving over strong baselines by 2.01–14 percentage points (pp).

In the field of medical image classification, the acquisition and annotation of images for training deep models, present significant challenges. Scholars have turned to transfer learning, particularly fine-tuning, as a partial solution to address data scarcity issues. However, when the volume of data in the target domain is insufficient, it leads to redundant parameters in deep models, resulting in overfitting and impacting the ultimate fine-tuning effect. This paper introduces a transfer learning approach based on task-adaptive parameter optimization from the perspective of sparse parameters, primarily applied within medical image classification with Convolutional Neural Networks (CNNs). Initially, individual fine-tuning is conducted on screened convolution kernels from each convolutional layer that are high closely associated with target domain classification, serving as an initial guide for fine-tuning model parameter updates. Subsequently, an adaptive low-rank fine-tuning method is designed, based on the varying contributions of each convolutional layer to classification, less correlated convolution kernels within each layer are adaptive fine-tuned to serve as bias parameters for classification. Experiments with three mainstream CNN models and fifteen medical datasets show that the proposed method can improve the overall fine-tuning efficiency, and the two fine-tuning strategies ensure optimal fine-tuning under the constraint of the number of samples in the target domain.

The development of diffusion models has led to remarkable performance in synthesizing high-fidelity and diverse images. However, despite their ability to generate high-quality images from varied text prompts, their core mechanism primarily relies on global semantic guidance, which makes it challenging to achieve precise, fine-grained control over the generated content. To address this insufficient controllability, existing methods mainly introduce additional conditional inputs (e.g., image guidance, replaceable text prompts, spatial constraints, and semantic masks) to regulate the generation process. Nevertheless, these methods often treat image content as a coupled whole. They focus primarily on generating images that satisfy given constraints but fail to decouple and independently manipulate internal semantic features. This content coupling limits a user's ability to perform fine-grained, decoupled editing. Targeting the limitations in latent space feature decoupling and fine-grained control, we propose PixelEdit, a diffusion-based framework for pixel-accurate image editing via latent diffusion feature decomposition. PixelEdit's core innovation lies in the unsupervised decomposition and recombination of semantic representations constructed during the diffusion process. Specifically, PixelEdit first extracts features from deep within the diffusion model and clusters them into a set of independent semantic representations. Subsequently, a comparative analysis of features generated with and without text prompt conditioning enables the synthesis of novel composite representations. Finally, by injecting these new features back into the diffusion process, PixelEdit achieves fine-grained image editing. Extensive experiments demonstrate that PixelEdit effectively decomposes images into independently manipulable semantic features, enabling highly refined image editing. Despite being unsupervised, PixelEdit achieves performance comparable or superior to state-of-the-art supervised methods in both controllability and output quality.

Federated Learning (FL) is vulnerable to model poisoning attacks, where malicious clients directly manipulate local gradient to corrupt global model optimization. Existing methods typically rely on task-relevant validation data or impose strict assumptions on data distributions, and therefore struggle to effectively distinguish malicious gradients from outlying benign gradients caused by data heterogeneity in privacy-constrained real world Non-IID scenarios. In this work, we investigate the fundamental differences in predictive behavior between malicious and benign models from the perspective of random data. Our key observation indicates that malicious models exhibit pronounced prediction bias when evaluated on random data due to tampering with local updates, whereas benign models are driven by prior knowledge and yield more diverse predictions. This provides a novel perspective for defending against model poisoning attacks beyond traditional gradient space analysis. To quantify this behavioral discrepancy, we introduce Stochastic Response Entropy, an entropy-based criterion that measures the diversity of model predictions on random data. Building on this insight, we propose FedSRE, a novel defense framework for FL. FedSRE introduces an adaptive robust aggregation strategy that maximizes global model prediction diversity on random data by dynamically adjusting client aggregation weights. Furthermore, FedSRE conducts cluster analysis on learned weights to identify potentially malicious clients and facilitate contribution-based aggregation of benign clients. Experiments across various datasets and attack settings show that FedSRE not only achieves competitive defense performance by effectively detecting malicious clients, but also quantifies the contributions of benign clients to enhance global model performance. The source code is available at this link.

Large reasoning models often reach correct answers through flawed intermediate steps, creating a gap between final accuracy and reasoning reliability. Existing alignment strategies address this with external verifiers or massive sampling, limiting scalability. In this work, we introduce CASPO (Confidence-Aware Step-wise Preference Optimization), a framework that aligns token-level confidence with step-wise logical correctness through iterative Direct Preference Optimization, without training a separate reward model. During inference, we propose Confidence-aware Thought (CaT), which leverages this calibrated confidence to dynamically prune uncertain reasoning branches with negligible O(V) latency. Experiments across ten benchmarks and multiple model families show that CASPO consistently improves reasoning reliability and inference efficiency. CASPO scales to Qwen3-8B-Base and surpasses tree-search baselines on AIME'24 and AIME'25 without using reward-model data. We also release a step-wise dataset with confidence annotations to support fine-grained analysis of reasoning reliability. Code is available at https://github.com/Thecommonirin/CASPO.

Cervical cytology screening requires automated and reliable assistance due to the labor-intensive nature and inter-observer variability of manual evaluation. Cropped tile classification models enable efficient analysis of gigapixel whole slide images but typically adopt binary or single-label formulations, which overlook coexisting lesion types and provide limited lesion localization, thus restricting clinical interpretability and diagnostic reliability. This study proposes MLCerNet, a Multi-Label Learning Cervical Screening Network that reformulates tile analysis as a multi-label classification task while producing lesion localization cues. To distinguish subtle morphological variations between lesion types, MLCerNet incorporates a Bilateral Attention Query (BiQuery) module that strengthens semantic interactions between lesion-specific and cell-wise features. To detect sparsely distributed lesions among abundant cells, it adopts a SimMat strategy that improves prediction robustness through a similarity matrix. By enabling lesion-specific CLS tokens to interact with fine-grained cell-wise tokens, the BiQuery module facilitates the modeling of subtle category-dependent morphological patterns. Building on these enhanced token representations, the SimMat strategy preserves category-aware token responses for prediction and provides localization cues under only tile-level label supervision. Extensive experiments demonstrate improved accuracy with enhanced interpretability for clinical practice. The code is available at https://github.com/imAzhou/MLCerNet .

Medical image analysis often suffers from insufficient data annotation. Pre-trained models can improve task performance when fine-tuned on medical data. Due to the domain differences, only a subset of the pre-trained parameters is supposed to be important to medical tasks. Current fine-tuning methods did not address the vital problems like “how many parameters should be fine-tuned” and “which group of parameters should be fine-tuned”. In this paper, we define the parameters whose removal incurs large loss increase of downstream task model as fundamental parameters. We observe that roughly 10% pre-trained parameters are fundamental to various medical applications, and these parameters dominate the optimal convergence of pre-trained model. Based on this, we introduce a divide-and-conquer method for optimal pre-trained model adaption to medical tasks. We apply a penalty factor to the gradients of fundamental parameters to control the parameter updates, and fine-tune the remaining parameters normally to fully adapt them to the downstream tasks. Our method leads to obvious performance improvement over full-parameter and other state-of-the-art fine-tuning methods on multiple medical tasks, providing new perspectives to pre-trained model adaptation.