This paper presents a novel knowledge distillation neural architecture leveraging efficient transformer networks for effective image classification. Natural images display intricate arrangements encompassing numerous extraneous elements. Vision transformers utilize localized patches to compute attention. However, exclusive dependence on patch segmentation proves inadequate in sufficiently encompassing the comprehensive nature of the image. To address this issue, we have proposed an inner-outer transformer-based architecture, which gives attention to the global and local aspects of the image. Moreover, The training of transformer models poses significant challenges due to their demanding resource, time, and data requirements. To tackle this, we integrate knowledge distillation into the architecture, enabling efficient learning. Leveraging insights from a larger teacher model, our approach enhances learning efficiency and effectiveness. Significantly, the transformer-in-transformer network acquires lightweight characteristics by means of distillation conducted within the feature extraction layer. Our featured network's robustness is established through substantial experimentation on the MNIST, CIFAR10, and CIFAR100 datasets, demonstrating commendable top-1 and top-5 accuracy. The conducted ablative analysis comprehensively validates the effectiveness of the chosen parameters and settings, showcasing their superiority against contemporary methodologies. Remarkably, the proposed Transformer-in-Transformer Network (TITN) model achieves impressive performance milestones across various datasets: securing the highest top-1 accuracy of 74.71% and a top-5 accuracy of 92.28% for the CIFAR100 dataset, attaining an unparalleled top-1 accuracy of 92.03% and top-5 accuracy of 99.80% for the CIFAR-10 dataset, and registering an exceptional top-1 accuracy of 99.56% for the MNIST dataset.
While Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning for Large Language Models (LLMs), its performance often falls short of Full Fine-Tuning (Full FT). Current methods optimize LoRA by initializing with static singular value decomposition (SVD) subsets, leading to suboptimal leveraging of pre-trained knowledge. Another path for improving LoRA is incorporating a Mixture-of-Experts (MoE) architecture. However, weight misalignment and complex gradient dynamics make it challenging to adopt SVD prior to the LoRA MoE architecture. To mitigate these issues, we propose \underline{G}reat L\underline{o}R\underline{A} Mixture-of-Exper\underline{t} (GOAT), a framework that (1) adaptively integrates relevant priors using an SVD-structured MoE, and (2) aligns optimization with full fine-tuned MoE by deriving a theoretical scaling factor. We demonstrate that proper scaling, without modifying the architecture or training algorithms, boosts LoRA MoE's efficiency and performance. Experiments across 25 datasets, including natural language understanding, commonsense reasoning, image classification, and natural language generation, demonstrate GOAT's state-of-the-art performance, closing the gap with Full FT.
This paper introduces a novel multiscale object-based graph neural network called MOB-GCN for hyperspectral image (HSI) classification. The central aim of this study is to enhance feature extraction and classification performance by utilizing multiscale object-based image analysis (OBIA). Traditional pixel-based methods often suffer from low accuracy and speckle noise, while single-scale OBIA approaches may overlook crucial information of image objects at different levels of detail. MOB-GCN overcomes these challenges by extracting and integrating features from multiple segmentation scales, leveraging the Multiresolution Graph Network (MGN) architecture to capture both fine-grained and global spatial patterns. MOB-GCN addresses this issue by extracting and integrating features from multiple segmentation scales to improve classification results using the Multiresolution Graph Network (MGN) architecture that can model fine-grained and global spatial patterns. By constructing a dynamic multiscale graph hierarchy, MOB-GCN offers a more comprehensive understanding of the intricate details and global context of HSIs. Experimental results demonstrate that MOB-GCN consistently outperforms single-scale graph convolutional networks (GCNs) in terms of classification accuracy, computational efficiency, and noise reduction, particularly when labeled data is limited. The implementation of MOB-GCN is publicly available at https://github.com/HySonLab/MultiscaleHSI
Deep Neural Networks (DNNs) have demonstrated remarkable success across a wide range of tasks, particularly in fields such as image classification. However, DNNs are highly susceptible to adversarial attacks, where subtle perturbations are introduced to input images, leading to erroneous model outputs. In today's digital era, ensuring the security and integrity of images processed by DNNs is of critical importance. One of the most prominent adversarial attack methods is the Fast Gradient Sign Method (FGSM), which perturbs images in the direction of the loss gradient to deceive the model. This paper presents a novel approach for detecting and filtering FGSM adversarial attacks in image processing tasks. Our proposed method evaluates 10,000 images, each subjected to five different levels of perturbation, characterized by $\epsilon$ values of 0.01, 0.02, 0.05, 0.1, and 0.2. These perturbations are applied in the direction of the loss gradient. We demonstrate that our approach effectively filters adversarially perturbed images, mitigating the impact of FGSM attacks. The method is implemented in Python, and the source code is publicly available on GitHub for reproducibility and further research.
Visual transformers have achieved remarkable performance in image classification tasks, but this performance gain has come at the cost of interpretability. One of the main obstacles to the interpretation of transformers is the self-attention mechanism, which mixes visual information across the whole image in a complex way. In this paper, we propose Hindered Transformer (HiT), a novel interpretable by design architecture inspired by visual transformers. Our proposed architecture rethinks the design of transformers to better disentangle patch influences at the classification stage. Ultimately, HiT can be interpreted as a linear combination of patch-level information. We show that the advantages of our approach in terms of explicability come with a reasonable trade-off in performance, making it an attractive alternative for applications where interpretability is paramount.
Knowledge transfer between teacher and student models has proven effective across various machine learning applications. However, challenges arise when the teacher's predictions are noisy, or the data domain during student training shifts from the teacher's pretraining data. In such scenarios, blindly relying on the teacher's predictions can lead to suboptimal knowledge transfer. To address these challenges, we propose a novel and universal framework, Adaptive Uncertainty-guided Knowledge Transfer ($\textbf{AUKT}$), which leverages Conformal Prediction (CP) to dynamically adjust the student's reliance on the teacher's guidance based on the teacher's prediction uncertainty. CP is a distribution-free, model-agnostic approach that provides reliable prediction sets with statistical coverage guarantees and minimal computational overhead. This adaptive mechanism mitigates the risk of learning undesirable or incorrect knowledge. We validate the proposed framework across diverse applications, including image classification, imitation-guided reinforcement learning, and autonomous driving. Experimental results consistently demonstrate that our approach improves performance, robustness and transferability, offering a promising direction for enhanced knowledge transfer in real-world applications.
Robust fine-tuning aims to adapt large foundation models to downstream tasks while preserving their robustness to distribution shifts. Existing methods primarily focus on constraining and projecting current model towards the pre-trained initialization based on the magnitudes between fine-tuned and pre-trained weights, which often require extensive hyper-parameter tuning and can sometimes result in underfitting. In this work, we propose Directional Gradient Projection (DiGraP), a novel layer-wise trainable method that incorporates directional information from gradients to bridge regularization and multi-objective optimization. Besides demonstrating our method on image classification, as another contribution we generalize this area to the multi-modal evaluation settings for robust fine-tuning. Specifically, we first bridge the uni-modal and multi-modal gap by performing analysis on Image Classification reformulated Visual Question Answering (VQA) benchmarks and further categorize ten out-of-distribution (OOD) VQA datasets by distribution shift types and degree (i.e. near versus far OOD). Experimental results show that DiGraP consistently outperforms existing baselines across Image Classfication and VQA tasks with discriminative and generative backbones, improving both in-distribution (ID) generalization and OOD robustness.
Label Smoothing (LS) is widely adopted to curb overconfidence in neural network predictions and enhance generalization. However, previous research shows that LS can force feature representations into excessively tight clusters, eroding intra-class distinctions. More recent findings suggest that LS also induces overconfidence in misclassifications, yet the precise mechanism remained unclear. In this work, we decompose the loss term introduced by LS, revealing two key components: (i) a regularization term that functions only when the prediction is correct, and (ii) an error-enhancement term that emerges under misclassifications. This latter term compels the model to reinforce incorrect predictions with exaggerated certainty, further collapsing the feature space. To address these issues, we propose Max Suppression (MaxSup), which uniformly applies the intended regularization to both correct and incorrect predictions by penalizing the top-1 logit instead of the ground-truth logit. Through feature analyses, we show that MaxSup restores intra-class variation and sharpens inter-class boundaries. Extensive experiments on image classification and downstream tasks confirm that MaxSup is a more robust alternative to LS. Code is available at: https://github.com/ZhouYuxuanYX/Maximum-Suppression-Regularization.
Over the last decade, the use of machine learning (ML) approaches in medicinal applications has increased manifold. Most of these approaches are based on deep learning, which aims to learn representations from grid data (like medical images). However, reinforcement learning (RL) applications in medicine are relatively less explored. Medical applications often involve a sequence of subtasks that form a diagnostic pipeline, and RL is uniquely suited to optimize over such sequential decision-making tasks. Ultrasound (US) image analysis is a quintessential example of such a sequential decision-making task, where the raw signal captured by the US transducer undergoes a series of signal processing and image post-processing steps, generally leading to a diagnostic suggestion. The application of RL in US remains limited. Deep Reinforcement Learning (DRL), that combines deep learning and RL, holds great promise in optimizing these pipelines by enabling intelligent and sequential decision-making. This review paper surveys the applications of RL in US over the last decade. We provide a succinct overview of the theoretic framework of RL and its application in US image processing and review existing work in each aspect of the image analysis pipeline. A comprehensive search of Scopus filtered on relevance yielded 14 papers most relevant to this topic. These papers were further categorized based on their target applications image classification, image segmentation, image enhancement, video summarization, and auto navigation and path planning. We also examined the type of RL approach used in each publication. Finally, we discuss key areas in healthcare where DRL approaches in US could be used for sequential decision-making. We analyze the opportunities, challenges, and limitations, providing insights into the future potential of DRL in US image analysis.
Classical machine learning often struggles with complex, high-dimensional data. Quantum machine learning offers a potential solution, promising more efficient processing. The quantum convolutional neural network (QCNN), a hybrid algorithm, fits current noisy intermediate-scale quantum hardware. However, its training depends largely on classical computation. Future gate-based quantum computers may realize full quantum advantages. In contrast to QCNNs, quantum autoencoders (QAEs) leverage classical optimization solely for parameter tuning. Data compression and reconstruction are handled entirely within quantum circuits, enabling purely quantum-based feature extraction. This study introduces a novel image-classification approach using QAEs, achieving classification without requiring additional qubits compared with conventional QAE implementations. The quantum circuit structure significantly impacts classification accuracy. Unlike hybrid methods such as QCNN, QAE-based classification emphasizes quantum computation. Our experiments demonstrate high accuracy in a four-class classification task, evaluating various quantum-gate configurations to understand the impact of different parameterized quantum circuit structures on classification performance. Specifically, noise-free conditions are considered, and simulations are performed using a statevector simulator to model the quantum system with full amplitude precision. Our results reveal that specific ansatz structures achieve superior accuracy. Moreover, the proposed approach achieves performance comparable to that of conventional machine-learning methods while significantly reducing the number of parameters requiring optimization. These findings indicate that QAEs can serve as efficient classification models with fewer parameters and highlight the potential of utilizing quantum circuits for complete end-to-end learning.
Image Steganography is a cryptographic technique that embeds secret information into an image, ensuring the hidden data remains undetectable to the human eye while preserving the image's original visual integrity. Least Significant Bit (LSB) Steganography achieves this by replacing the k least significant bits of an image with the k most significant bits of a secret image, maintaining the appearance of the original image while simultaneously encoding the essential elements of the hidden data. In this work, we shift away from conventional applications of steganography in deep learning and explore its potential from a new angle. We present experimental results on CIFAR-10 showing that LSB Steganography, when used as a data augmentation strategy for downstream computer vision tasks such as image classification, can significantly improve the training efficiency of deep neural networks. It can also act as an implicit, uniformly discretized piecewise linear approximation of color augmentations such as (brightness, contrast, hue, and saturation), without introducing additional training overhead through a new joint image training regime that disregards the need for tuning sensitive augmentation hyperparameters.
Stochastic resonance describes the utility of noise in improving the detectability of weak signals in certain types of systems. It has been observed widely in natural and engineered settings, but its utility in image classification with rate-based neural networks has not been studied extensively. In this analysis a simple LSTM recurrent neural network is trained for digit recognition and classification. During the test phase, image contrast is reduced to a point where the model fails to recognize the presence of a stimulus. Controlled noise is added to partially recover classification performance. The results indicate the presence of stochastic resonance in rate-based recurrent neural networks.
Integrating hyperspectral imagery (HSI) with deep neural networks (DNNs) can strengthen the accuracy of intelligent vision systems by combining spectral and spatial information, which is useful for tasks like semantic segmentation in autonomous driving. To advance research in such safety-critical systems, determining the precise contribution of spectral information to complex DNNs' output is needed. To address this, several saliency methods, such as class activation maps (CAM), have been proposed primarily for image classification. However, recent studies have raised concerns regarding their reliability. In this paper, we address their limitations and propose an alternative approach by leveraging the data provided by activations and weights from relevant DNN layers to better capture the relationship between input features and predictions. The study aims to assess the superior performance of HSI compared to 3-channel and single-channel DNNs. We also address the influence of spectral signature normalization for enhancing DNN robustness in real-world driving conditions.
Optical Coherence Tomography (OCT) provides high-resolution cross-sectional images useful for diagnosing various diseases, but their distinct characteristics from natural images raise questions about whether large-scale pre-training on datasets like ImageNet is always beneficial. In this paper, we investigate the impact of ImageNet-based pre-training on Vision Transformer (ViT) performance for OCT image classification across different dataset sizes. Our experiments cover four-category retinal pathologies (CNV, DME, Drusen, Normal). Results suggest that while pre-training can accelerate convergence and potentially offer better performance in smaller datasets, training from scratch may achieve comparable or even superior accuracy when sufficient OCT data is available. Our findings highlight the importance of matching domain characteristics in pre-training and call for further study on large-scale OCT-specific pre-training.
We consider the problem of conformal prediction under covariate shift. Given labeled data from a source domain and unlabeled data from a covariate shifted target domain, we seek to construct prediction sets with valid marginal coverage in the target domain. Most existing methods require estimating the unknown likelihood ratio function, which can be prohibitive for high-dimensional data such as images. To address this challenge, we introduce the likelihood ratio regularized quantile regression (LR-QR) algorithm, which combines the pinball loss with a novel choice of regularization in order to construct a threshold function without directly estimating the unknown likelihood ratio. We show that the LR-QR method has coverage at the desired level in the target domain, up to a small error term that we can control. Our proofs draw on a novel analysis of coverage via stability bounds from learning theory. Our experiments demonstrate that the LR-QR algorithm outperforms existing methods on high-dimensional prediction tasks, including a regression task for the Communities and Crime dataset, and an image classification task from the WILDS repository.
State space models (SSMs) have recently garnered significant attention in computer vision. However, due to the unique characteristics of image data, adapting SSMs from natural language processing to computer vision has not outperformed the state-of-the-art convolutional neural networks (CNNs) and Vision Transformers (ViTs). Existing vision SSMs primarily leverage manually designed scans to flatten image patches into sequences locally or globally. This approach disrupts the original semantic spatial adjacency of the image and lacks flexibility, making it difficult to capture complex image structures. To address this limitation, we propose Dynamic Adaptive Scan (DAS), a data-driven method that adaptively allocates scanning orders and regions. This enables more flexible modeling capabilities while maintaining linear computational complexity and global modeling capacity. Based on DAS, we further propose the vision backbone DAMamba, which significantly outperforms current state-of-the-art vision Mamba models in vision tasks such as image classification, object detection, instance segmentation, and semantic segmentation. Notably, it surpasses some of the latest state-of-the-art CNNs and ViTs. Code will be available at https://github.com/ltzovo/DAMamba.
Hyperspectral image (HSI) classification is a cornerstone of remote sensing, enabling precise material and land-cover identification through rich spectral information. While deep learning has driven significant progress in this task, small patch-based classifiers, which account for over 90% of the progress, face limitations: (1) the small patch (e.g., 7x7, 9x9)-based sampling approach considers a limited receptive field, resulting in insufficient spatial structural information critical for object-level identification and noise-like misclassifications even within uniform regions; (2) undefined optimal patch sizes lead to coarse label predictions, which degrade performance; and (3) a lack of multi-shape awareness around objects. To address these challenges, we draw inspiration from large-scale image segmentation techniques, which excel at handling object boundaries-a capability essential for semantic labeling in HSI classification. However, their application remains under-explored in this task due to (1) the prevailing notion that larger patch sizes degrade performance, (2) the extensive unlabeled regions in HSI groundtruth, and (3) the misalignment of input shapes between HSI data and segmentation models. Thus, in this study, we propose a novel paradigm and baseline, HSIseg, for HSI classification that leverages segmentation techniques combined with a novel Dynamic Shifted Regional Transformer (DSRT) to overcome these challenges. We also introduce an intuitive progressive learning framework with adaptive pseudo-labeling to iteratively incorporate unlabeled regions into the training process, thereby advancing the application of segmentation techniques. Additionally, we incorporate auxiliary data through multi-source data collaboration, promoting better feature interaction. Validated on five public HSI datasets, our proposal outperforms state-of-the-art methods.
Transformers have achieved great success in effectively processing sequential data such as text. Their architecture consisting of several attention and feedforward blocks can model relations between elements of a sequence in parallel manner, which makes them very efficient to train and effective in sequence modeling. Even though they have shown strong performance in processing sequential data, the size of their parameters is considerably larger when compared to other architectures such as RNN and CNN based models. Therefore, several approaches have explored parameter sharing and recurrence in Transformer models to address their computational demands. However, such methods struggle to maintain high performance compared to the original transformer model. To address this challenge, we propose our novel approach, RingFormer, which employs one Transformer layer that processes input repeatedly in a circular, ring-like manner, while utilizing low-rank matrices to generate input-dependent level signals. This allows us to reduce the model parameters substantially while maintaining high performance in a variety of tasks such as translation and image classification, as validated in the experiments.
Quantum machine learning (QML) has emerged as a promising domain to leverage the computational capabilities of quantum systems to solve complex classification tasks. In this work, we present the first comprehensive QML study by benchmarking the MedMNIST-a diverse collection of medical imaging datasets on a 127-qubit real IBM quantum hardware, to evaluate the feasibility and performance of quantum models (without any classical neural networks) in practical applications. This study explores recent advancements in quantum computing such as device-aware quantum circuits, error suppression, and mitigation for medical image classification. Our methodology is comprised of three stages: preprocessing, generation of noise-resilient and hardware-efficient quantum circuits, optimizing/training of quantum circuits on classical hardware, and inference on real IBM quantum hardware. Firstly, we process all input images in the preprocessing stage to reduce the spatial dimension due to quantum hardware limitations. We generate hardware-efficient quantum circuits using backend properties expressible to learn complex patterns for medical image classification. After classical optimization of QML models, we perform inference on real quantum hardware. We also incorporate advanced error suppression and mitigation techniques in our QML workflow, including dynamical decoupling (DD), gate twirling, and matrix-free measurement mitigation (M3) to mitigate the effects of noise and improve classification performance. The experimental results showcase the potential of quantum computing for medical imaging and establish a benchmark for future advancements in QML applied to healthcare.
Convolutional networks, transformers, hybrid models, and Mamba-based architectures have demonstrated strong performance across various medical image classification tasks. However, these methods were primarily designed to classify clean images using labeled data. In contrast, real-world clinical data often involve image corruptions that are unique to multi-center studies and stem from variations in imaging equipment across manufacturers. In this paper, we introduce the Medical Vision Transformer (MedViTV2), a novel architecture incorporating Kolmogorov-Arnold Network (KAN) layers into the transformer architecture for the first time, aiming for generalized medical image classification. We have developed an efficient KAN block to reduce computational load while enhancing the accuracy of the original MedViT. Additionally, to counteract the fragility of our MedViT when scaled up, we propose an enhanced Dilated Neighborhood Attention (DiNA), an adaptation of the efficient fused dot-product attention kernel capable of capturing global context and expanding receptive fields to scale the model effectively and addressing feature collapse issues. Moreover, a hierarchical hybrid strategy is introduced to stack our Local Feature Perception and Global Feature Perception blocks in an efficient manner, which balances local and global feature perceptions to boost performance. Extensive experiments on 17 medical image classification datasets and 12 corrupted medical image datasets demonstrate that MedViTV2 achieved state-of-the-art results in 27 out of 29 experiments with reduced computational complexity. MedViTV2 is 44\% more computationally efficient than the previous version and significantly enhances accuracy, achieving improvements of 4.6\% on MedMNIST, 5.8\% on NonMNIST, and 13.4\% on the MedMNIST-C benchmark.