The self-attention mechanism utilizes large implicit weight matrices, programmed through dot product-based activations with very few trainable parameters, to enable long sequence modeling. In this paper, we investigate the possibility of discarding residual learning by employing large implicit kernels to achieve full context interaction at each layer of the network. To accomplish it, we introduce coordinate-based implicit MLPs as a slow network to generate hyper-kernels for another fast convolutional network. To get context-varying weights for fast dynamic encoding, we propose a $\mathrm{Hyper}\mathcal{Z{\cdot}Z{\cdot}W}$ operator that connects hyper-kernels ($\mathcal{W}$) and hidden activations ($\mathcal{Z}$) through simple elementwise multiplication, followed by convolution of $\mathcal{Z}$ using the context-dependent $\mathcal{W}$. Based on this design, we present a novel Terminator architecture that integrates hyper-kernels of different sizes to produce multi-branch hidden representations for enhancing the feature extraction capability of each layer. Additionally, a bottleneck layer is employed to compress the concatenated channels, allowing only valuable information to propagate to the subsequent layers. Notably, our model incorporates several innovative components and exhibits excellent properties, such as introducing local feedback error for updating the slow network, stable zero-mean features, faster training convergence, and fewer model parameters. Extensive experimental results on pixel-level 1D and 2D image classification benchmarks demonstrate the superior performance of our architecture.
The increasing usage of Artificial Intelligence (AI) models, especially Deep Neural Networks (DNNs), is increasing the power consumption during training and inference, posing environmental concerns and driving the need for more energy-efficient algorithms and hardware solutions. This work addresses the growing energy consumption problem in Machine Learning (ML), particularly during the inference phase. Even a slight reduction in power usage can lead to significant energy savings, benefiting users, companies, and the environment. Our approach focuses on maximizing the accuracy of Artificial Neural Network (ANN) models using a neuroevolutionary framework whilst minimizing their power consumption. To do so, power consumption is considered in the fitness function. We introduce a new mutation strategy that stochastically reintroduces modules of layers, with power-efficient modules having a higher chance of being chosen. We introduce a novel technique that allows training two separate models in a single training step whilst promoting one of them to be more power efficient than the other while maintaining similar accuracy. The results demonstrate a reduction in power consumption of ANN models by up to 29.2% without a significant decrease in predictive performance.
Convolutional neural networks and vision transformers have achieved outstanding performance in machine perception, particularly for image classification. Although these image classifiers excel at predicting image-level class labels, they may not discriminate missing or shifted parts within an object. As a result, they may fail to detect corrupted images that involve missing or disarrayed semantic information in the object composition. On the contrary, human perception easily distinguishes such corruptions. To mitigate this gap, we introduce the concept of "image grammar", consisting of "image semantics" and "image syntax", to denote the semantics of parts or patches of an image and the order in which these parts are arranged to create a meaningful object. To learn the image grammar relative to a class of visual objects/scenes, we propose a weakly supervised two-stage approach. In the first stage, we use a deep clustering framework that relies on iterative clustering and feature refinement to produce part-semantic segmentation. In the second stage, we incorporate a recurrent bi-LSTM module to process a sequence of semantic segmentation patches to capture the image syntax. Our framework is trained to reason over patch semantics and detect faulty syntax. We benchmark the performance of several grammar learning models in detecting patch corruptions. Finally, we verify the capabilities of our framework in Celeb and SUNRGBD datasets and demonstrate that it can achieve a grammar validation accuracy of 70 to 90% in a wide variety of semantic and syntactical corruption scenarios.
Imaging Atmospheric Cherenkov Telescopes (IACTs) of gamma ray observatory TAIGA detect the Extesnive Air Showers (EASs) originating from the cosmic or gamma rays interactions with the atmosphere. Thereby, telescopes obtain images of the EASs. The ability to segregate gamma rays images from the hadronic cosmic ray background is one of the main features of this type of detectors. However, in actual IACT observations simultaneous observation of the background and the source of gamma ray is needed. This observation mode (called wobbling) modifies images of events, which affects the quality of selection by neural networks. Thus, in this work, the results of the application of neural networks (NN) for image classification task on Monte Carlo (MC) images of TAIGA-IACTs are presented. The wobbling mode is considered together with the image adaptation for adequate analysis by NNs. Simultaneously, we explore several neural network structures that classify events both directly from images or through Hillas parameters extracted from images. In addition, by employing NNs, MC simulation data are used to evaluate the quality of the segregation of rare gamma events with the account of all necessary image modifications.
Large-scale image datasets are often partially labeled, where only a few categories' labels are known for each image. Assigning pseudo-labels to unknown labels to gain additional training signals has become prevalent for training deep classification models. However, some pseudo-labels are inevitably incorrect, leading to a notable decline in the model classification performance. In this paper, we propose a novel method called Category-wise Fine-Tuning (CFT), aiming to reduce model inaccuracies caused by the wrong pseudo-labels. In particular, CFT employs known labels without pseudo-labels to fine-tune the logistic regressions of trained models individually to calibrate each category's model predictions. Genetic Algorithm, seldom used for training deep models, is also utilized in CFT to maximize the classification performance directly. CFT is applied to well-trained models, unlike most existing methods that train models from scratch. Hence, CFT is general and compatible with models trained with different methods and schemes, as demonstrated through extensive experiments. CFT requires only a few seconds for each category for calibration with consumer-grade GPUs. We achieve state-of-the-art results on three benchmarking datasets, including the CheXpert chest X-ray competition dataset (ensemble mAUC 93.33%, single model 91.82%), partially labeled MS-COCO (average mAP 83.69%), and Open Image V3 (mAP 85.31%), outperforming the previous bests by 0.28%, 2.21%, 2.50%, and 0.91%, respectively. The single model on CheXpert has been officially evaluated by the competition server, endorsing the correctness of the result. The outstanding results and generalizability indicate that CFT could be substantial and prevalent for classification model development. Code is available at: https://github.com/maxium0526/category-wise-fine-tuning.
Recently, efficient Vision Transformers have shown great performance with low latency on resource-constrained devices. Conventionally, they use 4x4 patch embeddings and a 4-stage structure at the macro level, while utilizing sophisticated attention with multi-head configuration at the micro level. This paper aims to address computational redundancy at all design levels in a memory-efficient manner. We discover that using larger-stride patchify stem not only reduces memory access costs but also achieves competitive performance by leveraging token representations with reduced spatial redundancy from the early stages. Furthermore, our preliminary analyses suggest that attention layers in the early stages can be substituted with convolutions, and several attention heads in the latter stages are computationally redundant. To handle this, we introduce a single-head attention module that inherently prevents head redundancy and simultaneously boosts accuracy by parallelly combining global and local information. Building upon our solutions, we introduce SHViT, a Single-Head Vision Transformer that obtains the state-of-the-art speed-accuracy tradeoff. For example, on ImageNet-1k, our SHViT-S4 is 3.3x, 8.1x, and 2.4x faster than MobileViTv2 x1.0 on GPU, CPU, and iPhone12 mobile device, respectively, while being 1.3% more accurate. For object detection and instance segmentation on MS COCO using Mask-RCNN head, our model achieves performance comparable to FastViT-SA12 while exhibiting 3.8x and 2.0x lower backbone latency on GPU and mobile device, respectively.
GNNs are widely used to solve various tasks including node classification and link prediction. Most of the GNN architectures assume the initial embedding to be random or generated from popular distributions. These initial embeddings require multiple layers of transformation to converge into a meaningful latent representation. While number of layers allow accumulation of larger neighbourhood of a node it also introduce the problem of over-smoothing. In addition, GNNs are inept at representing structural information. For example, the output embedding of a node does not capture its triangles participation. In this paper, we presented a novel feature extraction methodology GraphViz2Vec that can capture the structural information of a node's local neighbourhood to create meaningful initial embeddings for a GNN model. These initial embeddings helps existing models achieve state-of-the-art results in various classification tasks. Further, these initial embeddings help the model to produce desired results with only two layers which in turn reduce the problem of over-smoothing. The initial encoding of a node is obtained from an image classification model trained on multiple energy diagrams of its local neighbourhood. These energy diagrams are generated with the induced sub-graph of the nodes traversed by multiple random walks. The generated encodings increase the performance of existing models on classification tasks (with a mean increase of $4.65\%$ and $2.58\%$ for the node and link classification tasks, respectively), with some models achieving state-of-the-art results.
Handwritten character recognition (HCR) is a challenging problem for machine learning researchers. Unlike printed text data, handwritten character datasets have more variation due to human-introduced bias. With numerous unique character classes present, some data, such as Logographic Scripts or Sino-Korean character sequences, bring new complications to the HCR problem. The classification task on such datasets requires the model to learn high-complexity details of the images that share similar features. With recent advances in computational resource availability and further computer vision theory development, some research teams have effectively addressed the arising challenges. Although known for achieving high accuracy while keeping the number of parameters small, many common approaches are still not generalizable and use dataset-specific solutions to achieve better results. Due to complex structure, existing methods frequently prevent the solutions from gaining popularity. This paper proposes a highly scalable approach for detailed character image classification by introducing the model architecture, data preprocessing steps, and testing design instructions. We also perform experiments to compare the performance of our method with that of existing ones to show the improvements achieved.
The linear classifier is widely used in various image classification tasks. It works by optimizing the distance between a sample and its corresponding class center. However, in real-world data, one class can contain several local clusters, e.g., birds of different poses. To address this complexity, we propose a novel multi-center classifier. Different from the vanilla linear classifier, our proposal is established on the assumption that the deep features of the training set follow a Gaussian Mixture distribution. Specifically, we create a conditional Gaussian distribution for each class and then sample multiple sub-centers from that distribution to extend the linear classifier. This approach allows the model to capture intra-class local structures more efficiently. In addition, at test time we set the mean of the conditional Gaussian distribution as the class center of the linear classifier and follow the vanilla linear classifier outputs, thus requiring no additional parameters or computational overhead. Extensive experiments on image classification show that the proposed multi-center classifier is a powerful alternative to widely used linear classifiers. Code available at https://github.com/ZheminZhang1/MultiCenter-Classifier.
Using deep learning models pre-trained on Imagenet is the traditional solution for medical image classification to deal with data scarcity. Nevertheless, relevant literature supports that this strategy may offer limited gains due to the high dissimilarity between domains. Currently, the paradigm of adapting domain-specialized foundation models is proving to be a promising alternative. However, how to perform such knowledge transfer, and the benefits and limitations it presents, are under study. The CGI-HRDC challenge for Hypertensive Retinopathy diagnosis on fundus images introduces an appealing opportunity to evaluate the transferability of a recently released vision-language foundation model of the retina, FLAIR. In this work, we explore the potential of using FLAIR features as starting point for fundus image classification, and we compare its performance with regard to Imagenet initialization on two popular transfer learning methods: Linear Probing (LP) and Fine-Tuning (FP). Our empirical observations suggest that, in any case, the use of the traditional strategy provides performance gains. In contrast, direct transferability from FLAIR model allows gains of 2.5%. When fine-tuning the whole network, the performance gap increases up to 4%. In this case, we show that avoiding feature deterioration via LP initialization of the classifier allows the best re-use of the rich pre-trained features. Although direct transferability using LP still offers limited performance, we believe that foundation models such as FLAIR will drive the evolution of deep-learning-based fundus image analysis.
When training data is scarce, it is common to make use of a feature extractor that has been pre-trained on a large base dataset, either by fine-tuning its parameters on the ``target'' dataset or by directly adopting its representation as features for a simple classifier. Fine-tuning is ineffective for few-shot learning, since the target dataset contains only a handful of examples. However, directly adopting the features without fine-tuning relies on the base and target distributions being similar enough that these features achieve separability and generalization. This paper investigates whether better features for the target dataset can be obtained by training on fewer base classes, seeking to identify a more useful base dataset for a given task.We consider cross-domain few-shot image classification in eight different domains from Meta-Dataset and entertain multiple real-world settings (domain-informed, task-informed and uninformed) where progressively less detail is known about the target task. To our knowledge, this is the first demonstration that fine-tuning on a subset of carefully selected base classes can significantly improve few-shot learning. Our contributions are simple and intuitive methods that can be implemented in any few-shot solution. We also give insights into the conditions in which these solutions are likely to provide a boost in accuracy. We release the code to reproduce all experiments from this paper on GitHub. https://github.com/RafLaf/Few-and-Fewer.git
In recent years, large-scale pre-trained multimodal models (LMM) generally emerge to integrate the vision and language modalities, achieving considerable success in various natural language processing and computer vision tasks. The growing size of LMMs, however, results in a significant computational cost for fine-tuning these models for downstream tasks. Hence, prompt-based interaction strategy is studied to align modalities more efficiently. In this contex, we propose a novel prompt-based multimodal interaction strategy inspired by human memory strategy, namely Memory-Inspired Temporal Prompt Interaction (MITP). Our proposed method involves in two stages as in human memory strategy: the acquiring stage, and the consolidation and activation stage. We utilize temporal prompts on intermediate layers to imitate the acquiring stage, leverage similarity-based prompt interaction to imitate memory consolidation, and employ prompt generation strategy to imitate memory activation. The main strength of our paper is that we interact the prompt vectors on intermediate layers to leverage sufficient information exchange between modalities, with compressed trainable parameters and memory usage. We achieve competitive results on several datasets with relatively small memory usage and 2.0M of trainable parameters (about 1% of the pre-trained foundation model).
The availability of training data is one of the main limitations in deep learning applications for medical imaging. Data augmentation is a popular approach to overcome this problem. A new approach is a Machine Learning based augmentation, in particular usage of Generative Adversarial Networks (GAN). In this case, GANs generate images similar to the original dataset so that the overall training data amount is bigger, which leads to better performance of trained networks. A GAN model consists of two networks, a generator and a discriminator interconnected in a feedback loop which creates a competitive environment. This work is a continuation of the previous research where we trained StyleGAN2-ADA by Nvidia on the limited COVID-19 chest X-ray image dataset. In this paper, we study the dependence of the GAN-based augmentation performance on dataset size with a focus on small samples. Two datasets are considered, one with 1000 images per class (4000 images in total) and the second with 500 images per class (2000 images in total). We train StyleGAN2-ADA with both sets and then, after validating the quality of generated images, we use trained GANs as one of the augmentations approaches in multi-class classification problems. We compare the quality of the GAN-based augmentation approach to two different approaches (classical augmentation and no augmentation at all) by employing transfer learning-based classification of COVID-19 chest X-ray images. The results are quantified using different classification quality metrics and compared to the results from the literature. The GAN-based augmentation approach is found to be comparable with classical augmentation in the case of medium and large datasets but underperforms in the case of smaller datasets. The correlation between the size of the original dataset and the quality of classification is visible independently from the augmentation approach.
Foundation vision or vision-language models are trained on large unlabeled or noisy data and learn robust representations that can achieve impressive zero- or few-shot performance on diverse tasks. Given these properties, they are a natural fit for active learning (AL), which aims to maximize labeling efficiency. However, the full potential of foundation models has not been explored in the context of AL, specifically in the low-budget regime. In this work, we evaluate how foundation models influence three critical components of effective AL, namely, 1) initial labeled pool selection, 2) ensuring diverse sampling, and 3) the trade-off between representative and uncertainty sampling. We systematically study how the robust representations of foundation models (DINOv2, OpenCLIP) challenge existing findings in active learning. Our observations inform the principled construction of a new simple and elegant AL strategy that balances uncertainty estimated via dropout with sample diversity. We extensively test our strategy on many challenging image classification benchmarks, including natural images as well as out-of-domain biomedical images that are relatively understudied in the AL literature. We also provide a highly performant and efficient implementation of modern AL strategies (including our method) at https://github.com/sanketx/AL-foundation-models.
This paper investigates how adjustments to deep learning architectures impact model performance in image classification. Small-scale experiments generate initial insights although the trends observed are not consistent with the entire dataset. Filtering operations in the image processing pipeline are crucial, with image filtering before pre-processing yielding better results. The choice and order of layers as well as filter placement significantly impact model performance. This study provides valuable insights into optimizing deep learning models, with potential avenues for future research including collaborative platforms.
Neural networks have become popular due to their versatility and state-of-the-art results in many applications, such as image classification, natural language processing, speech recognition, forecasting, etc. These applications are also used in resource-constrained environments such as embedded devices. In this work, the susceptibility of neural network implementations to reverse engineering is explored on the NVIDIA Jetson Nano microcomputer via side-channel analysis. To this end, an architecture extraction attack is presented. In the attack, 15 popular convolutional neural network architectures (EfficientNets, MobileNets, NasNet, etc.) are implemented on the GPU of Jetson Nano and the electromagnetic radiation of the GPU is analyzed during the inference operation of the neural networks. The results of the analysis show that neural network architectures are easily distinguishable using deep learning-based side-channel analysis.
Few-shot image classification has emerged as a key challenge in the field of computer vision, highlighting the capability to rapidly adapt to new tasks with minimal labeled data. Existing methods predominantly rely on image-level features or local descriptors, often overlooking the holistic context surrounding these descriptors. In this work, we introduce a novel approach termed "Local Descriptor with Contextual Augmentation (LDCA)". Specifically, this method bridges the gap between local and global understanding uniquely by leveraging an adaptive global contextual enhancement module. This module incorporates a visual transformer, endowing local descriptors with contextual awareness capabilities, ranging from broad global perspectives to intricate surrounding nuances. By doing so, LDCA transcends traditional descriptor-based approaches, ensuring each local feature is interpreted within its larger visual narrative. Extensive experiments underscore the efficacy of our method, showing a maximal absolute improvement of 20\% over the next-best on fine-grained classification datasets, thus demonstrating significant advancements in few-shot classification tasks.
Skin tone as a demographic bias and inconsistent human labeling poses challenges in dermatology AI. We take another angle to investigate color contrast's impact, beyond skin tones, on malignancy detection in skin disease datasets: We hypothesize that in addition to skin tones, the color difference between the lesion area and skin also plays a role in malignancy detection performance of dermatology AI models. To study this, we first propose a robust labeling method to quantify color contrast scores of each image and validate our method by showing small labeling variations. More importantly, applying our method to \textit{the only} diverse-skin tone and pathologically-confirmed skin disease dataset DDI, yields \textbf{DDI-CoCo Dataset}, and we observe a performance gap between the high and low color difference groups. This disparity remains consistent across various state-of-the-art (SoTA) image classification models, which supports our hypothesis. Furthermore, we study the interaction between skin tone and color difference effects and suggest that color difference can be an additional reason behind model performance bias between skin tones. Our work provides a complementary angle to dermatology AI for improving skin disease detection.
Deep learning has made significant advances in computer vision, particularly in image classification tasks. Despite their high accuracy on training data, deep learning models often face challenges related to complexity and overfitting. One notable concern is that the model often relies heavily on a limited subset of filters for making predictions. This dependency can result in compromised generalization and an increased vulnerability to minor variations. While regularization techniques like weight decay, dropout, and data augmentation are commonly used to address this issue, they may not directly tackle the reliance on specific filters. Our observations reveal that the heavy reliance problem gets severe when slow-learning filters are deprived of learning opportunities due to fast-learning filters. Drawing inspiration from image augmentation research that combats over-reliance on specific image regions by removing and replacing parts of images, our idea is to mitigate the problem of over-reliance on strong filters by substituting highly activated features. To this end, we present a novel method called Catch-up Mix, which provides learning opportunities to a wide range of filters during training, focusing on filters that may lag behind. By mixing activation maps with relatively lower norms, Catch-up Mix promotes the development of more diverse representations and reduces reliance on a small subset of filters. Experimental results demonstrate the superiority of our method in various vision classification datasets, providing enhanced robustness.
Latent representations are used extensively for downstream tasks, such as visualization, interpolation or feature extraction of deep learning models. Invariant and equivariant neural networks are powerful and well-established models for enforcing inductive biases. In this paper, we demonstrate that the inductive bias imposed on the by an equivariant model must also be taken into account when using latent representations. We show how not accounting for the inductive biases leads to decreased performance on downstream tasks, and vice versa, how accounting for inductive biases can be done effectively by using an invariant projection of the latent representations. We propose principles for how to choose such a projection, and show the impact of using these principles in two common examples: First, we study a permutation equivariant variational auto-encoder trained for molecule graph generation; here we show that invariant projections can be designed that incur no loss of information in the resulting invariant representation. Next, we study a rotation-equivariant representation used for image classification. Here, we illustrate how random invariant projections can be used to obtain an invariant representation with a high degree of retained information. In both cases, the analysis of invariant latent representations proves superior to their equivariant counterparts. Finally, we illustrate that the phenomena documented here for equivariant neural networks have counterparts in standard neural networks where invariance is encouraged via augmentation. Thus, while these ambiguities may be known by experienced developers of equivariant models, we make both the knowledge as well as effective tools to handle the ambiguities available to the broader community.