Digital libraries often face the challenge of processing a large volume of diverse document types. The manual collection and tagging of metadata can be a time-consuming and error-prone task. To address this, we aim to develop an automatic metadata extractor for digital libraries. In this work, we introduce the Heterogeneous Learning Resources (HLR) dataset designed for document image classification. The approach involves decomposing individual learning resources into constituent document images (sheets). These images are then processed through an OCR tool to extract textual representation. State-of-the-art classifiers are employed to classify both the document image and its textual content. Subsequently, the labels of the constituent document images are utilized to predict the label of the overall document.
We present a general class of machine learning algorithms called parametric matrix models. In contrast with most existing machine learning models that imitate the biology of neurons, parametric matrix models use matrix equations that emulate physical systems. Similar to how physics problems are usually solved, parametric matrix models learn the governing equations that lead to the desired outputs. Parametric matrix models can be efficiently trained from empirical data, and the equations may use algebraic, differential, or integral relations. While originally designed for scientific computing, we prove that parametric matrix models are universal function approximators that can be applied to general machine learning problems. After introducing the underlying theory, we apply parametric matrix models to a series of different challenges that show their performance for a wide range of problems. For all the challenges tested here, parametric matrix models produce accurate results within an efficient and interpretable computational framework that allows for input feature extrapolation.
Datasets encountered in scientific and engineering applications appear in complex formats (e.g., images, multivariate time series, molecules, video, text strings, networks). Graph theory provides a unifying framework to model such datasets and enables the use of powerful tools that can help analyze, visualize, and extract value from data. In this work, we present PlasmoData.jl, an open-source, Julia framework that uses concepts of graph theory to facilitate the modeling and analysis of complex datasets. The core of our framework is a general data modeling abstraction, which we call a DataGraph. We show how the abstraction and software implementation can be used to represent diverse data objects as graphs and to enable the use of tools from topology, graph theory, and machine learning (e.g., graph neural networks) to conduct a variety of tasks. We illustrate the versatility of the framework by using real datasets: i) an image classification problem using topological data analysis to extract features from the graph model to train machine learning models; ii) a disease outbreak problem where we model multivariate time series as graphs to detect abnormal events; and iii) a technology pathway analysis problem where we highlight how we can use graphs to navigate connectivity. Our discussion also highlights how PlasmoData.jl leverages native Julia capabilities to enable compact syntax, scalable computations, and interfaces with diverse packages.
Knowledge distillation has emerged as a highly effective method for bridging the representation discrepancy between large-scale models and lightweight models. Prevalent approaches involve leveraging appropriate metrics to minimize the divergence or distance between the knowledge extracted from the teacher model and the knowledge learned by the student model. Centered Kernel Alignment (CKA) is widely used to measure representation similarity and has been applied in several knowledge distillation methods. However, these methods are complex and fail to uncover the essence of CKA, thus not answering the question of how to use CKA to achieve simple and effective distillation properly. This paper first provides a theoretical perspective to illustrate the effectiveness of CKA, which decouples CKA to the upper bound of Maximum Mean Discrepancy~(MMD) and a constant term. Drawing from this, we propose a novel Relation-Centered Kernel Alignment~(RCKA) framework, which practically establishes a connection between CKA and MMD. Furthermore, we dynamically customize the application of CKA based on the characteristics of each task, with less computational source yet comparable performance than the previous methods. The extensive experiments on the CIFAR-100, ImageNet-1k, and MS-COCO demonstrate that our method achieves state-of-the-art performance on almost all teacher-student pairs for image classification and object detection, validating the effectiveness of our approaches. Our code is available in https://github.com/Klayand/PCKA
Spiking Neural Networks (SNNs), as the third generation of neural networks, have gained prominence for their biological plausibility and computational efficiency, especially in processing diverse datasets. The integration of attention mechanisms, inspired by advancements in neural network architectures, has led to the development of Spiking Transformers. These have shown promise in enhancing SNNs' capabilities, particularly in the realms of both static and neuromorphic datasets. Despite their progress, a discernible gap exists in these systems, specifically in the Spiking Self Attention (SSA) mechanism's effectiveness in leveraging the temporal processing potential of SNNs. To address this, we introduce the Temporal Interaction Module (TIM), a novel, convolution-based enhancement designed to augment the temporal data processing abilities within SNN architectures. TIM's integration into existing SNN frameworks is seamless and efficient, requiring minimal additional parameters while significantly boosting their temporal information handling capabilities. Through rigorous experimentation, TIM has demonstrated its effectiveness in exploiting temporal information, leading to state-of-the-art performance across various neuromorphic datasets. The code is available at https://github.com/BrainCog-X/Brain-Cog/tree/main/examples/TIM.
Federated learning (FL) has emerged as a promising approach to collaboratively train machine learning models across multiple edge devices while preserving privacy. The success of FL hinges on the efficiency of participating models and their ability to handle the unique challenges of distributed learning. While several variants of Vision Transformer (ViT) have shown great potential as alternatives to modern convolutional neural networks (CNNs) for centralized training, the unprecedented size and higher computational demands hinder their deployment on resource-constrained edge devices, challenging their widespread application in FL. Since client devices in FL typically have limited computing resources and communication bandwidth, models intended for such devices must strike a balance between model size, computational efficiency, and the ability to adapt to the diverse and non-IID data distributions encountered in FL. To address these challenges, we propose OnDev-LCT: Lightweight Convolutional Transformers for On-Device vision tasks with limited training data and resources. Our models incorporate image-specific inductive biases through the LCT tokenizer by leveraging efficient depthwise separable convolutions in residual linear bottleneck blocks to extract local features, while the multi-head self-attention (MHSA) mechanism in the LCT encoder implicitly facilitates capturing global representations of images. Extensive experiments on benchmark image datasets indicate that our models outperform existing lightweight vision models while having fewer parameters and lower computational demands, making them suitable for FL scenarios with data heterogeneity and communication bottlenecks.
We propose the Multi-Head Density Adaptive Attention Mechanism (DAAM), a novel probabilistic attention framework that can be used for Parameter-Efficient Fine-tuning (PEFT), and the Density Adaptive Transformer (DAT), designed to enhance information aggregation across multiple modalities, including Speech, Text, and Vision. DAAM integrates learnable mean and variance into its attention mechanism, implemented in a multi-head framework, enabling it to collectively model any probability distribution for dynamic recalibration of feature significance. This method demonstrates significant improvements, especially with highly non-stationary data, surpassing the state-of-the-art attention techniques in model performance, up to approximately +20% (abs.) in accuracy. Empirically, DAAM exhibits superior adaptability and efficacy across a diverse range of tasks, including emotion recognition in speech, image classification, and text classification, thereby establishing its robustness and versatility in handling data across multiple modalities. Furthermore, we introduce the Importance Factor, a new learning-based metric that enhances the explainability of models trained with DAAM-based methods.
Histologic examination plays a crucial role in oncology research and diagnostics. The adoption of digital scanning of whole slide images (WSI) has created an opportunity to leverage deep learning-based image classification methods to enhance diagnosis and risk stratification. Technical limitations of current approaches to training deep convolutional neural networks (DCNN) result in suboptimal model performance and make training and deployment of comprehensive classification models unobtainable. In this study, we introduce a novel approach that addresses the main limitations of traditional histopathology classification model training. Our method, termed Learned Resizing with Efficient Training (LRET), couples efficient training techniques with image resizing to facilitate seamless integration of larger histology image patches into state-of-the-art classification models while preserving important structural information. We used the LRET method coupled with two distinct resizing techniques to train three diverse histology image datasets using multiple diverse DCNN architectures. Our findings demonstrate a significant enhancement in classification performance and training efficiency. Across the spectrum of experiments, LRET consistently outperforms existing methods, yielding a substantial improvement of 15-28% in accuracy for a large-scale, multiclass tumor classification task consisting of 74 distinct brain tumor types. LRET not only elevates classification accuracy but also substantially reduces training times, unlocking the potential for faster model development and iteration. The implications of this work extend to broader applications within medical imaging and beyond, where efficient integration of high-resolution images into deep learning pipelines is paramount for driving advancements in research and clinical practice.
Visual fine-tuning has garnered significant attention with the rise of pre-trained vision models. The current prevailing method, full fine-tuning, suffers from the issue of knowledge forgetting as it focuses solely on fitting the downstream training set. In this paper, we propose a novel weight rollback-based fine-tuning method called OLOR (One step Learning, One step Review). OLOR combines fine-tuning with optimizers, incorporating a weight rollback term into the weight update term at each step. This ensures consistency in the weight range of upstream and downstream models, effectively mitigating knowledge forgetting and enhancing fine-tuning performance. In addition, a layer-wise penalty is presented to employ penalty decay and the diversified decay rate to adjust the weight rollback levels of layers for adapting varying downstream tasks. Through extensive experiments on various tasks such as image classification, object detection, semantic segmentation, and instance segmentation, we demonstrate the general applicability and state-of-the-art performance of our proposed OLOR. Code is available at https://github.com/rainbow-xiao/OLOR-AAAI-2024.
Few-shot hyperspectral image classification aims to identify the classes of each pixel in the images by only marking few of these pixels. And in order to obtain the spatial-spectral joint features of each pixel, the fixed-size patches centering around each pixel are often used for classification. However, observing the classification results of existing methods, we found that boundary patches corresponding to the pixels which are located at the boundary of the objects in the hyperspectral images, are hard to classify. These boundary patchs are mixed with multi-class spectral information. Inspired by this, we propose to augment the prototype network with TransMix for few-shot hyperspectrial image classification(APNT). While taking the prototype network as the backbone, it adopts the transformer as feature extractor to learn the pixel-to-pixel relation and pay different attentions to different pixels. At the same time, instead of directly using the patches which are cut from the hyperspectral images for training, it randomly mixs up two patches to imitate the boundary patches and uses the synthetic patches to train the model, with the aim to enlarge the number of hard training samples and enhance their diversity. And by following the data agumentation technique TransMix, the attention returned by the transformer is also used to mix up the labels of two patches to generate better labels for synthetic patches. Compared with existing methods, the proposed method has demonstrated sate of the art performance and better robustness for few-shot hyperspectral image classification in our experiments.
Deep learning could be prone to learning shortcuts raised by dataset bias and result in inaccurate, unreliable, and unfair models, which impedes its adoption in real-world clinical applications. Despite its significance, there is a dearth of research in the medical image classification domain to address dataset bias. Furthermore, the bias labels are often agnostic, as identifying biases can be laborious and depend on post-hoc interpretation. This paper proposes learning Adaptive Agreement from a Biased Council (Ada-ABC), a debiasing framework that does not rely on explicit bias labels to tackle dataset bias in medical images. Ada-ABC develops a biased council consisting of multiple classifiers optimized with generalized cross entropy loss to learn the dataset bias. A debiasing model is then simultaneously trained under the guidance of the biased council. Specifically, the debiasing model is required to learn adaptive agreement with the biased council by agreeing on the correctly predicted samples and disagreeing on the wrongly predicted samples by the biased council. In this way, the debiasing model could learn the target attribute on the samples without spurious correlations while also avoiding ignoring the rich information in samples with spurious correlations. We theoretically demonstrated that the debiasing model could learn the target features when the biased model successfully captures dataset bias. Moreover, to our best knowledge, we constructed the first medical debiasing benchmark from four datasets containing seven different bias scenarios. Our extensive experiments practically showed that our proposed Ada-ABC outperformed competitive approaches, verifying its effectiveness in mitigating dataset bias for medical image classification. The codes and organized benchmark datasets will be made publicly available.
The recent advances in Deep Neural Networks (DNNs) stem from their exceptional performance across various domains. However, their inherent large size hinders deploying these networks on resource-constrained devices like edge, mobile, and IoT platforms. Strategies have emerged, from partial cloud computation offloading (split computing) to integrating early exits within DNN layers. Our work presents an innovative unified approach merging early exits and split computing. We determine the 'splitting layer', the optimal depth in the DNN for edge device computations, and whether to infer on edge device or be offloaded to the cloud for inference considering accuracy, computational efficiency, and communication costs. Also, Image classification faces diverse environmental distortions, influenced by factors like time of day, lighting, and weather. To adapt to these distortions, we introduce I-SplitEE, an online unsupervised algorithm ideal for scenarios lacking ground truths and with sequential data. Experimental validation using Caltech-256 and Cifar-10 datasets subjected to varied distortions showcases I-SplitEE's ability to reduce costs by a minimum of 55% with marginal performance degradation of at most 5%.
In this paper, we introduce Fast&Focused-Net, a novel deep neural network architecture tailored for efficiently encoding small objects into fixed-length feature vectors. Contrary to conventional Convolutional Neural Networks (CNNs), Fast&Focused-Net employs a series of our newly proposed layer, the Volume-wise Dot Product (VDP) layer, designed to address several inherent limitations of CNNs. Specifically, CNNs often exhibit a smaller effective receptive field than their theoretical counterparts, limiting their vision span. Additionally, the initial layers in CNNs produce low-dimensional feature vectors, presenting a bottleneck for subsequent learning. Lastly, the computational overhead of CNNs, particularly in capturing diverse image regions by parameter sharing, is significantly high. The VDP layer, at the heart of Fast&Focused-Net, aims to remedy these issues by efficiently covering the entire image patch information with reduced computational demand. Experimental results demonstrate the prowess of Fast&Focused-Net in a variety of applications. For small object classification tasks, our network outperformed state-of-the-art methods on datasets such as CIFAR-10, CIFAR-100, STL-10, SVHN-Cropped, and Fashion-MNIST. In the context of larger image classification, when combined with a transformer encoder (ViT), Fast&Focused-Net produced competitive results for OpenImages V6, ImageNet-1K, and Places365 datasets. Moreover, the same combination showcased unparalleled performance in text recognition tasks across SVT, IC15, SVTP, and HOST datasets. This paper presents the architecture, the underlying motivation, and extensive empirical evidence suggesting that Fast&Focused-Net is a promising direction for efficient and focused deep learning.
Land Cover (LC) image classification has become increasingly significant in understanding environmental changes, urban planning, and disaster management. However, traditional LC methods are often labor-intensive and prone to human error. This paper explores state-of-the-art deep learning models for enhanced accuracy and efficiency in LC analysis. We compare convolutional neural networks (CNN) against transformer-based methods, showcasing their applications and advantages in LC studies. We used EuroSAT, a patch-based LC classification data set based on Sentinel-2 satellite images and achieved state-of-the-art results using current transformer models.
Understanding the encoded representation of Deep Neural Networks (DNNs) has been a fundamental yet challenging objective. In this work, we focus on two possible directions for analyzing representations of DNNs by studying simple image classification tasks. Specifically, we consider \textit{On-Off pattern} and \textit{PathCount} for investigating how information is stored in deep representations. On-off pattern of a neuron is decided as `on' or `off' depending on whether the neuron's activation after ReLU is non-zero or zero. PathCount is the number of paths that transmit non-zero energy from the input to a neuron. We investigate how neurons in the network encodes information by replacing each layer's activation with On-Off pattern or PathCount and evaluating its effect on classification performance. We also examine correlation between representation and PathCount. Finally, we show a possible way to improve an existing DNN interpretation method, Class Activation Map (CAM), by directly utilizing On-Off or PathCount.
This study compared the classification performance of Gemini Pro and GPT-4V in educational settings. Employing visual question answering (VQA) techniques, the study examined both models' abilities to read text-based rubrics and then automatically score student-drawn models in science education. We employed both quantitative and qualitative analyses using a dataset derived from student-drawn scientific models and employing NERIF (Notation-Enhanced Rubrics for Image Feedback) prompting methods. The findings reveal that GPT-4V significantly outperforms Gemini Pro in terms of scoring accuracy and Quadratic Weighted Kappa. The qualitative analysis reveals that the differences may be due to the models' ability to process fine-grained texts in images and overall image classification performance. Even adapting the NERIF approach by further de-sizing the input images, Gemini Pro seems not able to perform as well as GPT-4V. The findings suggest GPT-4V's superior capability in handling complex multimodal educational tasks. The study concludes that while both models represent advancements in AI, GPT-4V's higher performance makes it a more suitable tool for educational applications involving multimodal data interpretation.
Recently the state space models (SSMs) with efficient hardware-aware designs, i.e., the Mamba deep learning model, have shown great potential for long sequence modeling. Meanwhile building efficient and generic vision backbones purely upon SSMs is an appealing direction. However, representing visual data is challenging for SSMs due to the position-sensitivity of visual data and the requirement of global context for visual understanding. In this paper, we show that the reliance on self-attention for visual representation learning is not necessary and propose a new generic vision backbone with bidirectional Mamba blocks (Vim), which marks the image sequences with position embeddings and compresses the visual representation with bidirectional state space models. On ImageNet classification, COCO object detection, and ADE20k semantic segmentation tasks, Vim achieves higher performance compared to well-established vision transformers like DeiT, while also demonstrating significantly improved computation & memory efficiency. For example, Vim is 2.8$\times$ faster than DeiT and saves 86.8% GPU memory when performing batch inference to extract features on images with a resolution of 1248$\times$1248. The results demonstrate that Vim is capable of overcoming the computation & memory constraints on performing Transformer-style understanding for high-resolution images and it has great potential to be the next-generation backbone for vision foundation models. Code is available at https://github.com/hustvl/Vim.
This paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.
Medical image segmentation aims to identify and locate abnormal structures in medical images, such as chest radiographs, using deep neural networks. These networks require a large number of annotated images with fine-grained masks for the regions of interest, making pre-training strategies based on classification datasets essential for sample efficiency. Based on a large-scale medical image classification dataset, our work collects explanations from well-trained classifiers to generate pseudo labels of segmentation tasks. Specifically, we offer a case study on chest radiographs and train image classifiers on the CheXpert dataset to identify 14 pathological observations in radiology. We then use Integrated Gradients (IG) method to distill and boost the explanations obtained from the classifiers, generating massive diagnosis-oriented localization labels (DoLL). These DoLL-annotated images are used for pre-training the model before fine-tuning it for downstream segmentation tasks, including COVID-19 infectious areas, lungs, heart, and clavicles. Our method outperforms other baselines, showcasing significant advantages in model performance and training efficiency across various segmentation settings.
Parameter-efficient fine-tuning (PEFT) of pre-trained language models (PLMs) has emerged as a highly successful approach, with training only a small number of parameters without sacrificing performance and becoming the de-facto learning paradigm with the increasing size of PLMs. However, existing PEFT methods are not memory-efficient, because they still require caching most of the intermediate activations for the gradient calculation, akin to fine-tuning. One effective way to reduce the activation memory is to apply a reversible model, so the intermediate activations are not necessary to be cached and can be recomputed. Nevertheless, modifying a PLM to its reversible variant is not straightforward, since the reversible model has a distinct architecture from the currently released PLMs. In this paper, we first investigate what is a key factor for the success of existing PEFT methods, and realize that it's essential to preserve the PLM's starting point when initializing a PEFT method. With this finding, we propose memory-efficient fine-tuning (MEFT) that inserts adapters into a PLM, preserving the PLM's starting point and making it reversible without additional pre-training. We evaluate MEFT on the GLUE benchmark and five question-answering tasks with various backbones, BERT, RoBERTa, BART and OPT. MEFT significantly reduces the activation memory up to 84% of full fine-tuning with a negligible amount of trainable parameters. Moreover, MEFT achieves the same score on GLUE and a comparable score on the question-answering tasks as full fine-tuning. A similar finding is also observed for the image classification task.