Acute lymphoblastic leukemia (ALL) is the most malignant form of leukemia and the most common cancer in adults and children. Traditionally, leukemia is diagnosed by analyzing blood and bone marrow smears under a microscope, with additional cytochemical tests for confirmation. However, these methods are expensive, time consuming, and highly dependent on expert knowledge. In recent years, deep learning, particularly Convolutional Neural Networks (CNNs), has provided advanced methods for classifying microscopic smear images, aiding in the detection of leukemic cells. These approaches are quick, cost effective, and not subject to human bias. However, most methods lack the ability to quantify uncertainty, which could lead to critical misdiagnoses. In this research, hybrid deep learning models (InceptionV3-GRU, EfficientNetB3-GRU, MobileNetV2-GRU) were implemented to classify ALL. Bayesian optimization was used to fine tune the model's hyperparameters and improve its performance. Additionally, Deep Ensemble uncertainty quantification was applied to address uncertainty during leukemia image classification. The proposed models were trained on the publicly available datasets ALL-IDB1 and ALL-IDB2. Their results were then aggregated at the score level using the sum rule. The parallel architecture used in these models offers a high level of confidence in differentiating between ALL and non-ALL cases. The proposed method achieved a remarkable detection accuracy rate of 100% on the ALL-IDB1 dataset, 98.07% on the ALL-IDB2 dataset, and 98.64% on the combined dataset, demonstrating its potential for accurate and reliable leukemia diagnosis.
The growing reproducibility crisis in machine learning has brought forward a need for careful examination of research findings. This paper investigates the claims made by Lei et al. (2023) regarding their proposed method, LICO, for enhancing post-hoc interpretability techniques and improving image classification performance. LICO leverages natural language supervision from a vision-language model to enrich feature representations and guide the learning process. We conduct a comprehensive reproducibility study, employing (Wide) ResNets and established interpretability methods like Grad-CAM and RISE. We were mostly unable to reproduce the authors' results. In particular, we did not find that LICO consistently led to improved classification performance or improvements in quantitative and qualitative measures of interpretability. Thus, our findings highlight the importance of rigorous evaluation and transparent reporting in interpretability research.
Automated data augmentation methods have significantly improved the performance and generalization capability of deep learning models in image classification. Yet, most state-of-the-art methods are optimized on common benchmark datasets, limiting their applicability to more diverse or domain-specific data, such as medical datasets. In this paper, we propose a strategy that uses large language models to automatically generate efficient augmentation policies, customized to fit the specific characteristics of any dataset and model architecture. The proposed method iteratively interacts with an LLM to obtain and refine the augmentation policies on model performance feedback, creating a dataset-agnostic data augmentation pipeline. The proposed method was evaluated on medical imaging datasets, showing a clear improvement over state-of-the-art methods. The proposed approach offers an adaptive and scalable solution. Although it increases computational cost, it significantly boosts model robustness, automates the process, and minimizes the need for human involvement during model development.
Contrastive Language-Image Pretraining (CLIP) performs zero-shot image classification by mapping images and textual class representation into a shared embedding space, then retrieving the class closest to the image. This work provides a new approach for interpreting CLIP models for image classification from the lens of mutual knowledge between the two modalities. Specifically, we ask: what concepts do both vision and language CLIP encoders learn in common that influence the joint embedding space, causing points to be closer or further apart? We answer this question via an approach of textual concept-based explanations, showing their effectiveness, and perform an analysis encompassing a pool of 13 CLIP models varying in architecture, size and pretraining datasets. We explore those different aspects in relation to mutual knowledge, and analyze zero-shot predictions. Our approach demonstrates an effective and human-friendly way of understanding zero-shot classification decisions with CLIP.
The rapid growth of model scale has necessitated substantial computational resources for fine-tuning. Existing approach such as Low-Rank Adaptation (LoRA) has sought to address the problem of handling the large updated parameters in full fine-tuning. However, LoRA utilize random initialization and optimization of low-rank matrices to approximate updated weights, which can result in suboptimal convergence and an accuracy gap compared to full fine-tuning. To address these issues, we propose LoLDU, a Parameter-Efficient Fine-Tuning (PEFT) approach that significantly reduces trainable parameters by 2600 times compared to regular PEFT methods while maintaining comparable performance. LoLDU leverages Lower-Diag-Upper Decomposition (LDU) to initialize low-rank matrices for faster convergence and orthogonality. We focus on optimizing the diagonal matrix for scaling transformations. To the best of our knowledge, LoLDU has the fewest parameters among all PEFT approaches. We conducted extensive experiments across 4 instruction-following datasets, 6 natural language understanding (NLU) datasets, 8 image classification datasets, and image generation datasets with multiple model types (LLaMA2, RoBERTa, ViT, and Stable Diffusion), providing a comprehensive and detailed analysis. Our open-source code can be accessed at \href{https://github.com/SKDDJ/LoLDU}{https://github.com/SKDDJ/LoLDU}.
Crop yield production could be enhanced for agricultural growth if various plant nutrition deficiencies, and diseases are identified and detected at early stages. The deep learning methods have proven its superior performances in the automated detection of plant diseases and nutrition deficiencies from visual symptoms in leaves. This article proposes a new deep learning method for plant nutrition deficiencies and disease classification using a graph convolutional network (GNN), added upon a base convolutional neural network (CNN). Sometimes, a global feature descriptor might fail to capture the vital region of a diseased leaf, which causes inaccurate classification of disease. To address this issue, regional feature learning is crucial for a holistic feature aggregation. In this work, region-based feature summarization at multi-scales is explored using spatial pyramidal pooling for discriminative feature representation. A GCN is developed to capacitate learning of finer details for classifying plant diseases and insufficiency of nutrients. The proposed method, called Plant Nutrition Deficiency and Disease Network (PND-Net), is evaluated on two public datasets for nutrition deficiency, and two for disease classification using four CNNs. The best classification performances are: (a) 90.00% Banana and 90.54% Coffee nutrition deficiency; and (b) 96.18% Potato diseases and 84.30% on PlantDoc datasets using Xception backbone. Furthermore, additional experiments have been carried out for generalization, and the proposed method has achieved state-of-the-art performances on two public datasets, namely the Breast Cancer Histopathology Image Classification (BreakHis 40X: 95.50%, and BreakHis 100X: 96.79% accuracy) and Single cells in Pap smear images for cervical cancer classification (SIPaKMeD: 99.18% accuracy). Also, PND-Net achieves improved performances using five-fold cross validation.
Self-supervised contrastive learning heavily relies on the view variance brought by data augmentation, so that it can learn a view-invariant pre-trained representation. Beyond increasing the view variance for contrast, this work focuses on improving the diversity of training data, to improve the generalization and robustness of the pre-trained models. To this end, we propose a unified framework to conduct data augmentation in the feature space, known as feature augmentation. This strategy is domain-agnostic, which augments similar features to the original ones and thus improves the data diversity. We perform a systematic investigation of various feature augmentation architectures, the gradient-flow skill, and the relationship between feature augmentation and traditional data augmentation. Our study reveals some practical principles for feature augmentation in self-contrastive learning. By integrating feature augmentation on the instance discrimination or the instance similarity paradigm, we consistently improve the performance of pre-trained feature learning and gain better generalization over the downstream image classification and object detection task.
Transformer plays a central role in many fundamental deep learning models, e.g., the ViT in computer vision and the BERT and GPT in natural language processing, whose effectiveness is mainly attributed to its multi-head attention (MHA) mechanism. In this study, we propose a simple and novel channel-wise sample permutation (CSP) operator, achieving a new structured MHA with fewer parameters and lower complexity. Given an input matrix, CSP circularly shifts the samples of different channels with various steps and then sorts grouped samples of each channel. This operator is equivalent to implicitly implementing cross-channel attention maps as permutation matrices, which achieves linear complexity and suppresses the risk of rank collapse when representing data. We replace the MHA of some representative models with CSP and test the CSP-based models in several discriminative tasks, including image classification and long sequence analysis. Experiments show that the CSP-based models achieve comparable or better performance with fewer parameters and lower computational costs than the classic Transformer and its state-of-the-art variants. The code is available at https://github.com/DaShenZi721/CSP.
In this paper, we develop novel techniques that can be used to alter the architecture of a neural network, while maintaining the function it represents. Such operations are known as function preserving transforms and have proven useful in transferring knowledge between networks to evaluate architectures quickly, thus having applications in efficient architectures searches. Our methods allow the integration of residual connections into function preserving transforms, so we call them R2R. We provide a derivation for R2R and show that it yields competitive performance with other function preserving transforms, thereby decreasing the restrictions on deep learning architectures that can be extended through function preserving transforms. We perform a comparative analysis with other function preserving transforms such as Net2Net and Network Morphisms, where we shed light on their differences and individual use cases. Finally, we show the effectiveness of R2R to train models quickly, as well as its ability to learn a more diverse set of filters on image classification tasks compared to Net2Net and Network Morphisms.
Despite extensive applications in surveillance and remote sensing, research on very low-resolution (VLR) image classification remains relatively unexplored in comparison to high-resolution (HR) image classification. We introduce a deep hybrid network that integrates capsule routing networks with a two-layer attention module. In the proposed architecture, the attention mechanism captures the more salient features, and the capsule network encodes these features to be robust to resolution changes. To enhance the network’s performance, a transfer learning on a custom image dataset, which is well-aligned to CIFAR-10, is utilized. The proposed model (Codes for the models are available at: https://github.com/kdhasi/Deep-CapsuleAttention.git ) is evaluated on two VLR classification tasks of ‘VLR complex image’ and ‘VLR real-world digit’. Experimental results demonstrate the superiority of the proposed model, achieving state-of-the-art (SOTA) results in both VLR complex image and VLR real-world digit domains while using fewer parameters compared to previous SOTA networks. Specifically, on the VLR CIFAR-10 dataset, the proposed model attains a 3.17% improvement in detection accuracy over the current benchmarks, and, on the VLR SVHN dataset, it achieves a 3.85% improvement by using 80% fewer parameters.
Conformal prediction provides model-agnostic and distribution-free uncertainty quantification through prediction sets that are guaranteed to include the ground truth with any user-specified probability. Yet, conformal prediction is not reliable under poisoning attacks where adversaries manipulate both training and calibration data, which can significantly alter prediction sets in practice. As a solution, we propose reliable prediction sets (RPS): the first efficient method for constructing conformal prediction sets with provable reliability guarantees under poisoning. To ensure reliability under training poisoning, we introduce smoothed score functions that reliably aggregate predictions of classifiers trained on distinct partitions of the training data. To ensure reliability under calibration poisoning, we construct multiple prediction sets, each calibrated on distinct subsets of the calibration data. We then aggregate them into a majority prediction set, which includes a class only if it appears in a majority of the individual sets. Both proposed aggregations mitigate the influence of datapoints in the training and calibration data on the final prediction set. We experimentally validate our approach on image classification tasks, achieving strong reliability while maintaining utility and preserving coverage on clean data. Overall, our approach represents an important step towards more trustworthy uncertainty quantification in the presence of data poisoning.
This paper presents a framework for deep transfer learning, which aims to leverage information from multi-domain upstream data with a large number of samples $n$ to a single-domain downstream task with a considerably smaller number of samples $m$, where $m \ll n$, in order to enhance performance on downstream task. Our framework offers several intriguing features. First, it allows the existence of both shared and domain-specific features across multi-domain data and provides a framework for automatic identification, achieving precise transfer and utilization of information. Second, the framework explicitly identifies upstream features that contribute to downstream tasks, establishing clear relationships between upstream domains and downstream tasks, thereby enhancing interpretability. Error analysis shows that our framework can significantly improve the convergence rate for learning Lipschitz functions in downstream supervised tasks, reducing it from $\tilde{O}(m^{-\frac{1}{2(d+2)}}+n^{-\frac{1}{2(d+2)}})$ ("no transfer") to $\tilde{O}(m^{-\frac{1}{2(d^*+3)}} + n^{-\frac{1}{2(d+2)}})$ ("partial transfer"), and even to $\tilde{O}(m^{-1/2}+n^{-\frac{1}{2(d+2)}})$ ("complete transfer"), where $d^* \ll d$ and $d$ is the dimension of the observed data. Our theoretical findings are supported by empirical experiments on image classification and regression datasets.
Large Language Models (LLMs) are increasingly used to assess NLP tasks due to their ability to generate human-like judgments. Single LLMs were used initially, however, recent work suggests using multiple LLMs as judges yields improved performance. An important step in exploiting multiple judgements is the combination stage, aggregation. Existing methods in NLP either assign equal weight to all LLM judgments or are designed for specific tasks such as hallucination detection. This work focuses on aggregating predictions from multiple systems where no reference labels are available. A new method called SkillAggregation is proposed, which learns to combine estimates from LLM judges without needing additional data or ground truth. It extends the Crowdlayer aggregation method, developed for image classification, to exploit the judge estimates during inference. The approach is compared to a range of standard aggregation methods on HaluEval-Dialogue, TruthfulQA and Chatbot Arena tasks. SkillAggregation outperforms Crowdlayer on all tasks, and yields the best performance over all approaches on the majority of tasks.
For the diagnosis of diabetes retinopathy (DR) images, this paper proposes a classification method based on artificial intelligence. The core lies in a new data augmentation method, GreenBen, which first extracts the green channel grayscale image from the retinal image and then performs Ben enhancement. Considering that diabetes macular edema (DME) is a complication closely related to DR, this paper constructs a joint classification framework of DR and DME based on multi task learning and attention module, and uses GreenBen to enhance its data to reduce the difference of DR images and improve the accuracy of model classification. We conducted extensive experiments on three publicly available datasets, and our method achieved the best results. For GreenBen, whether based on the ResNet50 network or the Swin Transformer network, whether for individual classification or joint DME classification, compared with other data augmentation methods, GreenBen achieved stable and significant improvements in DR classification results, with an accuracy increase of 10%.
In this study, we investigate the performance of few-shot classification models across different domains, specifically natural images and histopathological images. We first train several few-shot classification models on natural images and evaluate their performance on histopathological images. Subsequently, we train the same models on histopathological images and compare their performance. We incorporated four histopathology datasets and one natural images dataset and assessed performance across 5-way 1-shot, 5-way 5-shot, and 5-way 10-shot scenarios using a selection of state-of-the-art classification techniques. Our experimental results reveal insights into the transferability and generalization capabilities of few-shot classification models between diverse image domains. We analyze the strengths and limitations of these models in adapting to new domains and provide recommendations for optimizing their performance in cross-domain scenarios. This research contributes to advancing our understanding of few-shot learning in the context of image classification across diverse domains.
Vision mambas have demonstrated strong performance with linear complexity to the number of vision tokens. Their efficiency results from processing image tokens sequentially. However, most existing methods employ patch-based image tokenization and then flatten them into 1D sequences for causal processing, which ignore the intrinsic 2D structural correlations of images. It is also difficult to extract global information by sequential processing of local patches. In this paper, we propose a global image serialization method to transform the image into a sequence of causal tokens, which contain global information of the 2D image. We first convert the image from the spatial domain to the frequency domain using Discrete Cosine Transform (DCT) and then arrange the pixels with corresponding frequency ranges. We further transform each set within the same frequency band back to the spatial domain to obtain a series of images before tokenization. We construct a vision mamba model, GlobalMamba, with a causal input format based on the proposed global image serialization, which can better exploit the causal relations among image sequences. Extensive experiments demonstrate the effectiveness of our GlobalMamba, including image classification on ImageNet-1K, object detection on COCO, and semantic segmentation on ADE20K.
In this paper, we introduce the big.LITTLE Vision Transformer, an innovative architecture aimed at achieving efficient visual recognition. This dual-transformer system is composed of two distinct blocks: the big performance block, characterized by its high capacity and substantial computational demands, and the LITTLE efficiency block, designed for speed with lower capacity. The key innovation of our approach lies in its dynamic inference mechanism. When processing an image, our system determines the importance of each token and allocates them accordingly: essential tokens are processed by the high-performance big model, while less critical tokens are handled by the more efficient little model. This selective processing significantly reduces computational load without sacrificing the overall performance of the model, as it ensures that detailed analysis is reserved for the most important information. To validate the effectiveness of our big.LITTLE Vision Transformer, we conducted comprehensive experiments on image classification and segment anything task. Our results demonstrate that the big.LITTLE architecture not only maintains high accuracy but also achieves substantial computational savings. Specifically, our approach enables the efficient handling of large-scale visual recognition tasks by dynamically balancing the trade-offs between performance and efficiency. The success of our method underscores the potential of hybrid models in optimizing both computation and performance in visual recognition tasks, paving the way for more practical and scalable deployment of advanced neural networks in real-world applications.
As the demand for high-quality training data escalates, researchers have increasingly turned to generative models to create synthetic data, addressing data scarcity and enabling continuous model improvement. However, reliance on self-generated data introduces a critical question: Will this practice amplify bias in future models? While most research has focused on overall performance, the impact on model bias, particularly subgroup bias, remains underexplored. In this work, we investigate the effects of the generated data on image classification tasks, with a specific focus on bias. We develop a practical simulation environment that integrates a self-consuming loop, where the generative model and classification model are trained synergistically. Hundreds of experiments are conducted on Colorized MNIST, CIFAR-20/100, and Hard ImageNet datasets to reveal changes in fairness metrics across generations. In addition, we provide a conjecture to explain the bias dynamics when training models on continuously augmented datasets across generations. Our findings contribute to the ongoing debate on the implications of synthetic data for fairness in real-world applications.
Parameter-efficient tuning (PET) techniques calibrate the model's predictions on downstream tasks by freezing the pre-trained models and introducing a small number of learnable parameters. However, despite the numerous PET methods proposed, their robustness has not been thoroughly investigated. In this paper, we systematically explore the robustness of four classical PET techniques (e.g., VPT, Adapter, AdaptFormer, and LoRA) under both white-box attacks and information perturbations. For white-box attack scenarios, we first analyze the performance of PET techniques using FGSM and PGD attacks. Subsequently, we further explore the transferability of adversarial samples and the impact of learnable parameter quantities on the robustness of PET methods. Under information perturbation attacks, we introduce four distinct perturbation strategies, including Patch-wise Drop, Pixel-wise Drop, Patch Shuffle, and Gaussian Noise, to comprehensively assess the robustness of these PET techniques in the presence of information loss. Via these extensive studies, we enhance the understanding of the robustness of PET methods, providing valuable insights for improving their performance in computer vision applications. The code is available at https://github.com/JCruan519/PETRobustness.
Image-based Joint-Embedding Predictive Architecture (IJEPA) offers an attractive alternative to Masked Autoencoder (MAE) for representation learning using the Masked Image Modeling framework. IJEPA drives representations to capture useful semantic information by predicting in latent rather than input space. However, IJEPA relies on carefully designed context and target windows to avoid representational collapse. The encoder modules in IJEPA cannot adaptively modulate the type of predicted and/or target features based on the feasibility of the masked prediction task as they are not given sufficient information of both context and targets. Based on the intuition that in natural images, information has a strong spatial bias with spatially local regions being highly predictive of one another compared to distant ones. We condition the target encoder and context encoder modules in IJEPA with positions of context and target windows respectively. Our "conditional" encoders show performance gains on several image classification benchmark datasets, improved robustness to context window size and sample-efficiency during pretraining.