Text-based visual descriptors--ranging from simple class names to more descriptive phrases--are widely used in visual concept discovery and image classification with vision-language models (VLMs). Their effectiveness, however, depends on a complex interplay of factors, including semantic clarity, presence in the VLM's pre-training data, and how well the descriptors serve as a meaningful representation space. In this work, we systematically analyze descriptor quality along two key dimensions: (1) representational capacity, and (2) relationship with VLM pre-training data. We evaluate a spectrum of descriptor generation methods, from zero-shot LLM-generated prompts to iteratively refined descriptors. Motivated by ideas from representation alignment and language understanding, we introduce two alignment-based metrics--Global Alignment and CLIP Similarity--that move beyond accuracy. These metrics shed light on how different descriptor generation strategies interact with foundation model properties, offering new ways to study descriptor effectiveness beyond accuracy evaluations.
Large Vision-Language Models (VLMs) have revolutionized computer vision, enabling tasks such as image classification, captioning, and visual question answering. However, they remain highly vulnerable to adversarial attacks, particularly in scenarios where both visual and textual modalities can be manipulated. In this study, we conduct a comprehensive reproducibility study of "An Image is Worth 1000 Lies: Adversarial Transferability Across Prompts on Vision-Language Models" validating the Cross-Prompt Attack (CroPA) and confirming its superior cross-prompt transferability compared to existing baselines. Beyond replication we propose several key improvements: (1) A novel initialization strategy that significantly improves Attack Success Rate (ASR). (2) Investigate cross-image transferability by learning universal perturbations. (3) A novel loss function targeting vision encoder attention mechanisms to improve generalization. Our evaluation across prominent VLMs -- including Flamingo, BLIP-2, and InstructBLIP as well as extended experiments on LLaVA validates the original results and demonstrates that our improvements consistently boost adversarial effectiveness. Our work reinforces the importance of studying adversarial vulnerabilities in VLMs and provides a more robust framework for generating transferable adversarial examples, with significant implications for understanding the security of VLMs in real-world applications.
To advance biomedical vison-language model capabilities through scaling up, fine-tuning, and instruction tuning, develop vision-language models with improved performance in handling long text, explore strategies to efficiently adopt vision language models for diverse multi-modal biomedical tasks, and examine the zero-shot learning performance. We developed two biomedical vision language models, BiomedGPT-Large and BiomedGPT-XLarge, based on an encoder-decoder-based transformer architecture. We fine-tuned the two models on 23 benchmark datasets from 6 multi-modal biomedical tasks including one image-only task (image classification), three language-only tasks (text understanding, text summarization and question answering), and two vision-language tasks (visual question answering and image captioning). We compared the developed scaled models with our previous BiomedGPT-Base model and existing prestigious models reported in the literature. We instruction-tuned the two models using a large-scale multi-modal biomedical instruction-tuning dataset and assessed the zero-shot learning performance and alignment accuracy.
Concept Bottleneck Models (CBMs) decompose image classification into a process governed by interpretable, human-readable concepts. Recent advances in CBMs have used Large Language Models (LLMs) to generate candidate concepts. However, a critical question remains: What is the optimal number of concepts to use? Current concept banks suffer from redundancy or insufficient coverage. To address this issue, we introduce a dynamic, agent-based approach that adjusts the concept bank in response to environmental feedback, optimizing the number of concepts for sufficiency yet concise coverage. Moreover, we propose Conditional Concept Bottleneck Models (CoCoBMs) to overcome the limitations in traditional CBMs' concept scoring mechanisms. It enhances the accuracy of assessing each concept's contribution to classification tasks and feature an editable matrix that allows LLMs to correct concept scores that conflict with their internal knowledge. Our evaluations across 6 datasets show that our method not only improves classification accuracy by 6% but also enhances interpretability assessments by 30%.
Microscopic assessment of histopathology images is vital for accurate cancer diagnosis and treatment. Whole Slide Image (WSI) classification and captioning have become crucial tasks in computer-aided pathology. However, microscopic WSI face challenges such as redundant patches and unknown patch positions due to subjective pathologist captures. Moreover, generating automatic pathology captions remains a significant challenge. To address these issues, we introduce a novel GNN-ViTCap framework for classification and caption generation from histopathological microscopic images. First, a visual feature extractor generates patch embeddings. Redundant patches are then removed by dynamically clustering these embeddings using deep embedded clustering and selecting representative patches via a scalar dot attention mechanism. We build a graph by connecting each node to its nearest neighbors in the similarity matrix and apply a graph neural network to capture both local and global context. The aggregated image embeddings are projected into the language model's input space through a linear layer and combined with caption tokens to fine-tune a large language model. We validate our method on the BreakHis and PatchGastric datasets. GNN-ViTCap achieves an F1 score of 0.934 and an AUC of 0.963 for classification, along with a BLEU-4 score of 0.811 and a METEOR score of 0.569 for captioning. Experimental results demonstrate that GNN-ViTCap outperforms state of the art approaches, offering a reliable and efficient solution for microscopy based patient diagnosis.
In recent years, compressed domain semantic inference has primarily relied on learned image coding models optimized for mean squared error (MSE). However, MSE-oriented optimization tends to yield latent spaces with limited semantic richness, which hinders effective semantic inference in downstream tasks. Moreover, achieving high performance with these models often requires fine-tuning the entire vision model, which is computationally intensive, especially for large models. To address these problems, we introduce Perception-Oriented Latent Coding (POLC), an approach that enriches the semantic content of latent features for high-performance compressed domain semantic inference. With the semantically rich latent space, POLC requires only a plug-and-play adapter for fine-tuning, significantly reducing the parameter count compared to previous MSE-oriented methods. Experimental results demonstrate that POLC achieves rate-perception performance comparable to state-of-the-art generative image coding methods while markedly enhancing performance in vision tasks, with minimal fine-tuning overhead. Code is available at https://github.com/NJUVISION/POLC.
Very low-resolution (VLR) images present a significant challenge for deep classification networks due to their inherent lack of fine spatial detail. While capsule networks (CapsNets), which encode spatial and pose information, are robust to resolution changes, they often struggle to perform and scale effectively on complex datasets. In this study, we introduce a novel frequency routing-based CapsNet (FR-CapsNet) to replace the conventional spatial routing in CapsNets. Though VLR images lose fine grained features, they retain high-level features captured by the low-frequency components. By computing capsule activation and pose information in the frequency domain and subsequently encoding them in the spatial domain, FR-CapsNet improves robustness to resolution degradation. Furthermore, our method utilizes a global routing framework that considerably reduces computational demands, enabling FR-CapsNet to scale effectively to larger and more diverse datasets. FR-CapsNet outperforms state-of-the-art (SOTA) convolutional neural networks (CNNs), other CapsNets, Transformers, and other advanced architectures in real-world VLR digit and image classification tasks. Specifically, on the VLR CIFAR-10 dataset, FR-CapsNet surpasses the current benchmark by 4.77% while using 4 times fewer parameters. Similarly, on the VLR SVHN and CIFAR-100 datasets, it exceeds the benchmark by 0.27% and 1.55%, respectively. Extensive experiments further demonstrate the superior generalization and robustness of FR-CapsNet compared to other SOTA methods. The codes for our models are available at https://github.com/kdhasi/FR-CapsNet.git
Vision-Language Models (VLMs) have rapidly advanced alongside Large Language Models (LLMs). This study evaluates the capabilities of prominent generative VLMs, such as GPT-4.1 and Gemini 2.5 Pro, accessed via APIs, for histopathology image classification tasks, including cell typing. Using diverse datasets from public and private sources, we apply zero-shot and one-shot prompting methods to assess VLM performance, comparing them against custom-trained Convolutional Neural Networks (CNNs). Our findings demonstrate that while one-shot prompting significantly improves VLM performance over zero-shot ($p \approx 1.005 \times 10^{-5}$ based on Kappa scores), these general-purpose VLMs currently underperform supervised CNNs on most tasks. This work underscores both the promise and limitations of applying current VLMs to specialized domains like pathology via in-context learning. All code and instructions for reproducing the study can be accessed from the repository https://www.github.com/a12dongithub/VLMCCE.
We propose a graph semi-supervised learning framework for classification tasks on data manifolds. Motivated by the manifold hypothesis, we model data as points sampled from a low-dimensional manifold $\mathcal{M} \subset \mathbb{R}^F$. The manifold is approximated in an unsupervised manner using a variational autoencoder (VAE), where the trained encoder maps data to embeddings that represent their coordinates in $\mathbb{R}^F$. A geometric graph is constructed with Gaussian-weighted edges inversely proportional to distances in the embedding space, transforming the point classification problem into a semi-supervised node classification task on the graph. This task is solved using a graph neural network (GNN). Our main contribution is a theoretical analysis of the statistical generalization properties of this data-to-manifold-to-graph pipeline. We show that, under uniform sampling from $\mathcal{M}$, the generalization gap of the semi-supervised task diminishes with increasing graph size, up to the GNN training error. Leveraging a training procedure which resamples a slightly larger graph at regular intervals during training, we then show that the generalization gap can be reduced even further, vanishing asymptotically. Finally, we validate our findings with numerical experiments on image classification benchmarks, demonstrating the empirical effectiveness of our approach.
Deep neural networks are increasingly being used to detect and diagnose medical conditions using medical imaging. Despite their utility, these models are highly vulnerable to adversarial attacks and distribution shifts, which can affect diagnostic reliability and undermine trust among healthcare professionals. In this study, we propose a robust training algorithm with data augmentation (RTDA) to mitigate these vulnerabilities in medical image classification. We benchmark classifier robustness against adversarial perturbations and natural variations of RTDA and six competing baseline techniques, including adversarial training and data augmentation approaches in isolation and combination, using experimental data sets with three different imaging technologies (mammograms, X-rays, and ultrasound). We demonstrate that RTDA achieves superior robustness against adversarial attacks and improved generalization performance in the presence of distribution shift in each image classification task while maintaining high clean accuracy.
This study investigates the integration of signal processing transformations -- Fast Fourier Transform (FFT), Walsh-Hadamard Transform (WHT), and Discrete Cosine Transform (DCT) -- within the ResNet50 convolutional neural network (CNN) model for image classification. The primary objective is to assess the trade-offs between computational efficiency, energy consumption, and classification accuracy during training and inference. Using the CIFAR-100 dataset (100 classes, 60,000 images), experiments demonstrated that incorporating WHT significantly reduced energy consumption while improving accuracy. Specifically, a baseline ResNet50 model achieved a testing accuracy of 66%, consuming an average of 25,606 kJ per model. In contrast, a modified ResNet50 incorporating WHT in the early convolutional layers achieved 74% accuracy, and an enhanced version with WHT applied to both early and late layers achieved 79% accuracy, with an average energy consumption of only 39 kJ per model. These results demonstrate the potential of WHT as a highly efficient and effective approach for energy-constrained CNN applications.
Federated learning (FL) has emerged as a promising approach for collaborative medical image analysis, enabling multiple institutions to build robust predictive models while preserving sensitive patient data. In the context of Whole Slide Image (WSI) classification, FL faces significant challenges, including heterogeneous computational resources across participating medical institutes and privacy concerns. To address these challenges, we propose FedWSIDD, a novel FL paradigm that leverages dataset distillation (DD) to learn and transmit synthetic slides. On the server side, FedWSIDD aggregates synthetic slides from participating centres and distributes them across all centres. On the client side, we introduce a novel DD algorithm tailored to histopathology datasets which incorporates stain normalisation into the distillation process to generate a compact set of highly informative synthetic slides. These synthetic slides, rather than model parameters, are transmitted to the server. After communication, the received synthetic slides are combined with original slides for local tasks. Extensive experiments on multiple WSI classification tasks, including CAMELYON16 and CAMELYON17, demonstrate that FedWSIDD offers flexibility for heterogeneous local models, enhances local WSI classification performance, and preserves patient privacy. This makes it a highly effective solution for complex WSI classification tasks. The code is available at FedWSIDD.
Kernel size selection in Convolutional Neural Networks (CNNs) is a critical but often overlooked design decision that affects receptive field, feature extraction, computational cost, and model accuracy. This paper proposes the Best Kernel Size Estimation Function (BKSEF), a mathematically grounded and empirically validated framework for optimal, layer-wise kernel size determination. BKSEF balances information gain, computational efficiency, and accuracy improvements by integrating principles from information theory, signal processing, and learning theory. Extensive experiments on CIFAR-10, CIFAR-100, ImageNet-lite, ChestX-ray14, and GTSRB datasets demonstrate that BKSEF-guided architectures achieve up to 3.1 percent accuracy improvement and 42.8 percent reduction in FLOPs compared to traditional models using uniform 3x3 kernels. Two real-world case studies further validate the approach: one for medical image classification in a cloud-based setup, and another for traffic sign recognition on edge devices. The former achieved enhanced interpretability and accuracy, while the latter reduced latency and model size significantly, with minimal accuracy trade-off. These results show that kernel size can be an active, optimizable parameter rather than a fixed heuristic. BKSEF provides practical heuristics and theoretical support for researchers and developers seeking efficient and application-aware CNN designs. It is suitable for integration into neural architecture search pipelines and real-time systems, offering a new perspective on CNN optimization.
The widespread use of Magnetic Resonance Imaging (MRI) and the rise of deep learning have enabled the development of powerful predictive models for a wide range of diagnostic tasks in MRI, such as image classification or object segmentation. However, training models for specific new tasks often requires large amounts of labeled data, which is difficult to obtain due to high annotation costs and data privacy concerns. To circumvent this issue, we introduce MRI-CORE (MRI COmprehensive Representation Encoder), a vision foundation model pre-trained using more than 6 million slices from over 110,000 MRI volumes across 18 main body locations. Experiments on five diverse object segmentation tasks in MRI demonstrate that MRI-CORE can significantly improve segmentation performance in realistic scenarios with limited labeled data availability, achieving an average gain of 6.97% 3D Dice Coefficient using only 10 annotated slices per task. We further demonstrate new model capabilities in MRI such as classification of image properties including body location, sequence type and institution, and zero-shot segmentation. These results highlight the value of MRI-CORE as a generalist vision foundation model for MRI, potentially lowering the data annotation resource barriers for many applications.
Quantum machine learning models that leverage quantum circuits as quantum feature maps (QFMs) are recognized for their enhanced expressive power in learning tasks. Such models have demonstrated rigorous end-to-end quantum speedups for specific families of classification problems. However, deploying deep QFMs on real quantum hardware remains challenging due to circuit noise and hardware constraints. Additionally, variational quantum algorithms often suffer from computational bottlenecks, particularly in accurate gradient estimation, which significantly increases quantum resource demands during training. We propose Iterative Quantum Feature Maps (IQFMs), a hybrid quantum-classical framework that constructs a deep architecture by iteratively connecting shallow QFMs with classically computed augmentation weights. By incorporating contrastive learning and a layer-wise training mechanism, IQFMs effectively reduces quantum runtime and mitigates noise-induced degradation. In tasks involving noisy quantum data, numerical experiments show that IQFMs outperforms quantum convolutional neural networks, without requiring the optimization of variational quantum parameters. Even for a typical classical image classification benchmark, a carefully designed IQFMs achieves performance comparable to that of classical neural networks. This framework presents a promising path to address current limitations and harness the full potential of quantum-enhanced machine learning.
We developed a proof-of-concept method for the automatic analysis of the structure and content of incunabula pages. A custom dataset comprising 500 annotated pages from five different incunabula was created using resources from the Jagiellonian Digital Library. Each page was manually labeled with five predefined classes: Text, Title, Picture, Table, and Handwriting. Additionally, the publicly available DocLayNet dataset was utilized as supplementary training data. To perform object detection, YOLO11n and YOLO11s models were employed and trained using two strategies: a combined dataset (DocLayNet and the custom dataset) and the custom dataset alone. The highest performance (F1 = 0.94) was achieved by the YOLO11n model trained exclusively on the custom data. Optical character recognition was then conducted on regions classified as Text, using both Tesseract and Kraken OCR, with Tesseract demonstrating superior results. Subsequently, image classification was applied to the Picture class using a ResNet18 model, achieving an accuracy of 98.7% across five subclasses: Decorative_letter, Illustration, Other, Stamp, and Wrong_detection. Furthermore, the CLIP model was utilized to generate semantic descriptions of illustrations. The results confirm the potential of machine learning in the analysis of early printed books, while emphasizing the need for further advancements in OCR performance and visual content interpretation.
Differential privacy (DP) auditing aims to provide empirical lower bounds on the privacy guarantees of DP mechanisms like DP-SGD. While some existing techniques require many training runs that are prohibitively costly, recent work introduces one-run auditing approaches that effectively audit DP-SGD in white-box settings while still being computationally efficient. However, in the more practical black-box setting where gradients cannot be manipulated during training and only the last model iterate is observed, prior work shows that there is still a large gap between the empirical lower bounds and theoretical upper bounds. Consequently, in this work, we study how incorporating approaches for stronger membership inference attacks (MIA) can improve one-run auditing in the black-box setting. Evaluating on image classification models trained on CIFAR-10 with DP-SGD, we demonstrate that our proposed approach, which utilizes quantile regression for MIA, achieves tighter bounds while crucially maintaining the computational efficiency of one-run methods.
Adversarial examples have attracted significant attention over the years, yet understanding their frequency-based characteristics remains insufficient. In this paper, we investigate the intriguing properties of adversarial examples in the frequency domain for the image classification task, with the following key findings. (1) As the high-frequency components increase, the performance gap between adversarial and natural examples becomes increasingly pronounced. (2) The model performance against filtered adversarial examples initially increases to a peak and declines to its inherent robustness. (3) In Convolutional Neural Networks, mid- and high-frequency components of adversarial examples exhibit their attack capabilities, while in Transformers, low- and mid-frequency components of adversarial examples are particularly effective. These results suggest that different network architectures have different frequency preferences and that differences in frequency components between adversarial and natural examples may directly influence model robustness. Based on our findings, we further conclude with three useful proposals that serve as a valuable reference to the AI model security community.
In this paper, we propose ProtoSolo, a novel deep neural architecture for interpretable image classification inspired by prototypical networks such as ProtoPNet. Existing prototype networks usually rely on the collaborative decision-making of multiple prototypes to achieve the classification and interpretation of a single category. In contrast, ProtoSolo only requires the activation of a single prototype to complete the classification. This allows the network to explain each category decision by only providing the features that are most similar to the prototype of that category, significantly reducing the cognitive complexity of the explanation. Secondly, we propose a feature-based comparison method, which uses feature map instead of full-channel feature vector as the object of similarity comparison and prototype learning. This design enables ProtoSolo to utilize richer global information for classification while relying on a single prototype activation. In addition, we propose a non-prototype projection learning strategy, which preserves the information association between the prototype and the training image patches while avoiding the sharp change of the network structure caused by the projection operation, thus avoiding its negative impact on the classification performance. Experiments on the CUB-200-2011 and Stanford Cars datasets show that ProtoSolo achieves superior performance in classification tasks and reaches the best level in terms of cognitive complexity of explanations compared to state-of-the-art interpretable methods. The code is available at https://github.com/pyt19/ProtoSolo.
This study investigates the design choices of parameterized quantum circuits (PQCs) within quantum and hybrid convolutional neural network (HQNN and QCNN) architectures, applied to the task of satellite image classification using the EuroSAT dataset. We systematically evaluate the performance implications of data encoding techniques, variational ans\"atze, and measurement in approx. 500 distinct model configurations. Our analysis reveals a clear hierarchy of influence on model performance. For hybrid architectures, which were benchmarked against their direct classical equivalents (e.g. the same architecture with the PQCs removed), the data encoding strategy is the dominant factor, with validation accuracy varying over 30% for distinct embeddings. In contrast, the selection of variational ans\"atze and measurement basis had a comparatively marginal effect, with validation accuracy variations remaining below 5%. For purely quantum models, restricted to amplitude encoding, performance was most dependent on the measurement protocol and the data-to-amplitude mapping. The measurement strategy varied the validation accuracy by up to 30% and the encoding mapping by around 8 percentage points.