PyTorchKR Papers

최신 논문

총 10,165편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2303.06296공식 코드 · Papers with Code

Stabilizing Transformer Training by Preventing Attention Entropy Collapse

Training stability is of great importance to Transformers. In this work, we investigate the training dynamics of Transformers by examining the evolution of the attention layers. In particular, we track the attention entropy for each attention head during the course of training, which is a proxy for model sharpness. We identify a common pattern across different architectures and tasks, where low attention entropy is accompanied by high training instability, which can take the form of oscillating loss or divergence. We denote the pathologically low attention entropy, corresponding to highly concentrated attention scores, as $\textit{entropy collapse}$. As a remedy, we propose $\sigma$Reparam, a simple and efficient solution where we reparametrize all linear layers with spectral normalization and an additional learned scalar. We demonstrate that $\sigma$Reparam successfully prevents entropy collapse in the attention layers, promoting more stable training. Additionally, we prove a tight lower bound of the attention entropy, which decreases exponentially fast with the spectral norm of the attention logits, providing additional motivation for our approach. We conduct experiments with $\sigma$Reparam on image classification, image self-supervised learning, machine translation, speech recognition, and language modeling tasks. We show that $\sigma$Reparam provides stability and robustness with respect to the choice of hyperparameters, going so far as enabling training (a) a Vision Transformer {to competitive performance} without warmup, weight decay, layer normalization or adaptive optimizers; (b) deep architectures in machine translation and (c) speech recognition to competitive performance without warmup and adaptive optimizers. Code is available at \url{https://github.com/apple/ml-sigma-reparam}.

arXiv:2303.05809공식 코드 · Papers with Code

Distributionally Robust Optimization with Probabilistic Group

Modern machine learning models may be susceptible to learning spurious correlations that hold on average but not for the atypical group of samples. To address the problem, previous approaches minimize the empirical worst-group risk. Despite the promise, they often assume that each sample belongs to one and only one group, which does not allow expressing the uncertainty in group labeling. In this paper, we propose a novel framework PG-DRO, which explores the idea of probabilistic group membership for distributionally robust optimization. Key to our framework, we consider soft group membership instead of hard group annotations. The group probabilities can be flexibly generated using either supervised learning or zero-shot approaches. Our framework accommodates samples with group membership ambiguity, offering stronger flexibility and generality than the prior art. We comprehensively evaluate PG-DRO on both image classification and natural language processing benchmarks, establishing superior performance

arXiv:2303.05860

Variational Quantum Neural Networks (VQNNS) in Image Classification

Quantum machine learning has established as an interdisciplinary field to overcome limitations of classical machine learning and neural networks. This is a field of research which can prove that quantum computers are able to solve problems with complex correlations between inputs that can be hard for classical computers. This suggests that learning models made on quantum computers may be more powerful for applications, potentially faster computation and better generalization on less data. The objective of this paper is to investigate how training of quantum neural network (QNNs) can be done using quantum optimization algorithms for improving the performance and time complexity of QNNs. A classical neural network can be partially quantized to create a hybrid quantum-classical neural network which is used mainly in classification and image recognition. In this paper, a QNN structure is made where a variational parameterized circuit is incorporated as an input layer named as Variational Quantum Neural Network (VQNNs). We encode the cost function of QNNs onto relative phases of a superposition state in the Hilbert space of the network parameters. The parameters are tuned with an iterative quantum approximate optimisation (QAOA) mixer and problem hamiltonians. VQNNs is experimented with MNIST digit recognition (less complex) and crack image classification datasets (more complex) which converges the computation in lesser time than QNN with decent training accuracy.

arXiv:2303.05789

AnoMalNet: Outlier Detection based Malaria Cell Image Classification Method Leveraging Deep Autoencoder

Class imbalance is a pervasive issue in the field of disease classification from medical images. It is necessary to balance out the class distribution while training a model for decent results. However, in the case of rare medical diseases, images from affected patients are much harder to come by compared to images from non-affected patients, resulting in unwanted class imbalance. Various processes of tackling class imbalance issues have been explored so far, each having its fair share of drawbacks. In this research, we propose an outlier detection based binary medical image classification technique which can handle even the most extreme case of class imbalance. We have utilized a dataset of malaria parasitized and uninfected cells. An autoencoder model titled AnoMalNet is trained with only the uninfected cell images at the beginning and then used to classify both the affected and non-affected cell images by thresholding a loss value. We have achieved an accuracy, precision, recall, and F1 score of 98.49%, 97.07%, 100%, and 98.52% respectively, performing better than large deep learning models and other published works. As our proposed approach can provide competitive results without needing the disease-positive samples during training, it should prove to be useful in binary disease classification on imbalanced datasets.

arXiv:2303.05952

Understanding and Constructing Latent Modality Structures in Multi-modal Representation Learning

Contrastive loss has been increasingly used in learning representations from multiple modalities. In the limit, the nature of the contrastive loss encourages modalities to exactly match each other in the latent space. Yet it remains an open question how the modality alignment affects the downstream task performance. In this paper, based on an information-theoretic argument, we first prove that exact modality alignment is sub-optimal in general for downstream prediction tasks. Hence we advocate that the key of better performance lies in meaningful latent modality structures instead of perfect modality alignment. To this end, we propose three general approaches to construct latent modality structures. Specifically, we design 1) a deep feature separation loss for intra-modality regularization; 2) a Brownian-bridge loss for inter-modality regularization; and 3) a geometric consistency loss for both intra- and inter-modality regularization. Extensive experiments are conducted on two popular multi-modal representation learning frameworks: the CLIP-based two-tower model and the ALBEF-based fusion model. We test our model on a variety of tasks including zero/few-shot image classification, image-text retrieval, visual question answering, visual reasoning, and visual entailment. Our method achieves consistent improvements over existing methods, demonstrating the effectiveness and generalizability of our proposed approach on latent modality structure regularization.

arXiv:2303.05780공식 코드 · Papers with Code

TAKT: Target-Aware Knowledge Transfer for Whole Slide Image Classification

Transferring knowledge from a source domain to a target domain can be crucial for whole slide image classification, since the number of samples in a dataset is often limited due to high annotation costs. However, domain shift and task discrepancy between datasets can hinder effective knowledge transfer. In this paper, we propose a Target-Aware Knowledge Transfer framework, employing a teacher-student paradigm. Our framework enables the teacher model to learn common knowledge from the source and target domains by actively incorporating unlabelled target images into the training of the teacher model. The teacher bag features are subsequently adapted to supervise the training of the student model on the target domain. Despite incorporating the target features during training, the teacher model tends to overlook them under the inherent domain shift and task discrepancy. To alleviate this, we introduce a target-aware feature alignment module to establish a transferable latent relationship between the source and target features by solving the optimal transport problem. Experimental results show that models employing knowledge transfer outperform those trained from scratch, and our method achieves state-of-the-art performance among other knowledge transfer methods on various datasets, including TCGA-RCC, TCGA-NSCLC, and Camelyon16.

arXiv:2303.05470공식 코드 · Papers with Code

Spawrious: A Benchmark for Fine Control of Spurious Correlation Biases

The problem of spurious correlations (SCs) arises when a classifier relies on non-predictive features that happen to be correlated with the labels in the training data. For example, a classifier may misclassify dog breeds based on the background of dog images. This happens when the backgrounds are correlated with other breeds in the training data, leading to misclassifications during test time. Previous SC benchmark datasets suffer from varying issues, e.g., over-saturation or only containing one-to-one (O2O) SCs, but no many-to-many (M2M) SCs arising between groups of spurious attributes and classes. In this paper, we present \benchmark-\{O2O, M2M\}-\{Easy, Medium, Hard\}, an image classification benchmark suite containing spurious correlations between classes and backgrounds. To create this dataset, we employ a text-to-image model to generate photo-realistic images and an image captioning model to filter out unsuitable ones. The resulting dataset is of high quality and contains approximately 152k images. Our experimental results demonstrate that state-of-the-art group robustness methods struggle with \benchmark, most notably on the Hard-splits with none of them getting over $70\%$ accuracy on the hardest split using a ResNet50 pretrained on ImageNet. By examining model misclassifications, we detect reliances on spurious backgrounds, demonstrating that our dataset provides a significant challenge.

arXiv:2303.04958

NIFF: Alleviating Forgetting in Generalized Few-Shot Object Detection via Neural Instance Feature Forging

Privacy and memory are two recurring themes in a broad conversation about the societal impact of AI. These concerns arise from the need for huge amounts of data to train deep neural networks. A promise of Generalized Few-shot Object Detection (G-FSOD), a learning paradigm in AI, is to alleviate the need for collecting abundant training samples of novel classes we wish to detect by leveraging prior knowledge from old classes (i.e., base classes). G-FSOD strives to learn these novel classes while alleviating catastrophic forgetting of the base classes. However, existing approaches assume that the base images are accessible, an assumption that does not hold when sharing and storing data is problematic. In this work, we propose the first data-free knowledge distillation (DFKD) approach for G-FSOD that leverages the statistics of the region of interest (RoI) features from the base model to forge instance-level features without accessing the base images. Our contribution is three-fold: (1) we design a standalone lightweight generator with (2) class-wise heads (3) to generate and replay diverse instance-level base features to the RoI head while finetuning on the novel data. This stands in contrast to standard DFKD approaches in image classification, which invert the entire network to generate base images. Moreover, we make careful design choices in the novel finetuning pipeline to regularize the model. We show that our approach can dramatically reduce the base memory requirements, all while setting a new standard for G-FSOD on the challenging MS-COCO and PASCAL-VOC benchmarks.

arXiv:2303.04183

Robustness-preserving Lifelong Learning via Dataset Condensation

Lifelong learning (LL) aims to improve a predictive model as the data source evolves continuously. Most work in this learning paradigm has focused on resolving the problem of 'catastrophic forgetting,' which refers to a notorious dilemma between improving model accuracy over new data and retaining accuracy over previous data. Yet, it is also known that machine learning (ML) models can be vulnerable in the sense that tiny, adversarial input perturbations can deceive the models into producing erroneous predictions. This motivates the research objective of this paper - specification of a new LL framework that can salvage model robustness (against adversarial attacks) from catastrophic forgetting. Specifically, we propose a new memory-replay LL strategy that leverages modern bi-level optimization techniques to determine the 'coreset' of the current data (i.e., a small amount of data to be memorized) for ease of preserving adversarial robustness over time. We term the resulting LL framework 'Data-Efficient Robustness-Preserving LL' (DERPLL). The effectiveness of DERPLL is evaluated for class-incremental image classification using ResNet-18 over the CIFAR-10 dataset. Experimental results show that DERPLL outperforms the conventional coreset-guided LL baseline and achieves a substantial improvement in both standard accuracy and robust accuracy.

arXiv:2303.04440

HyT-NAS: Hybrid Transformers Neural Architecture Search for Edge Devices

Vision Transformers have enabled recent attention-based Deep Learning (DL) architectures to achieve remarkable results in Computer Vision (CV) tasks. However, due to the extensive computational resources required, these architectures are rarely implemented on resource-constrained platforms. Current research investigates hybrid handcrafted convolution-based and attention-based models for CV tasks such as image classification and object detection. In this paper, we propose HyT-NAS, an efficient Hardware-aware Neural Architecture Search (HW-NAS) including hybrid architectures targeting vision tasks on tiny devices. HyT-NAS improves state-of-the-art HW-NAS by enriching the search space and enhancing the search strategy as well as the performance predictors. Our experiments show that HyT-NAS achieves a similar hypervolume with less than ~5x training evaluations. Our resulting architecture outperforms MLPerf MobileNetV1 by 6.3% accuracy improvement with 3.5x less number of parameters on Visual Wake Words.

arXiv:2303.03932공식 코드 · Papers with Code

FFT-based Dynamic Token Mixer for Vision

Multi-head-self-attention (MHSA)-equipped models have achieved notable performance in computer vision. Their computational complexity is proportional to quadratic numbers of pixels in input feature maps, resulting in slow processing, especially when dealing with high-resolution images. New types of token-mixer are proposed as an alternative to MHSA to circumvent this problem: an FFT-based token-mixer involves global operations similar to MHSA but with lower computational complexity. However, despite its attractive properties, the FFT-based token-mixer has not been carefully examined in terms of its compatibility with the rapidly evolving MetaFormer architecture. Here, we propose a novel token-mixer called Dynamic Filter and novel image recognition models, DFFormer and CDFFormer, to close the gaps above. The results of image classification and downstream tasks, analysis, and visualization show that our models are helpful. Notably, their throughput and memory efficiency when dealing with high-resolution image recognition is remarkable. Our results indicate that Dynamic Filter is one of the token-mixer options that should be seriously considered. The code is available at https://github.com/okojoalg/dfformer

arXiv:2303.03707

Hybrid quantum-classical convolutional neural network for phytoplankton classification

The taxonomic composition and abundance of phytoplankton, having direct impact on marine ecosystem dynamic and global environment change, are listed as essential ocean variables. Phytoplankton classification is very crucial for Phytoplankton analysis, but it is very difficult because of the huge amount and tiny volume of Phytoplankton. Machine learning is the principle way of performing phytoplankton image classification automatically. When carrying out large-scale research on the marine phytoplankton, the volume of data increases overwhelmingly and more powerful computational resources are required for the success of machine learning algorithms. Recently, quantum machine learning has emerged as the potential solution for large-scale data processing by harnessing the exponentially computational power of quantum computer. Here, for the first time, we demonstrate the feasibility of quantum deep neural networks for phytoplankton classification. Hybrid quantum-classical convolutional and residual neural networks are developed based on the classical architectures. These models make a proper balance between the limited function of the current quantum devices and the large size of phytoplankton images, which make it possible to perform phytoplankton classification on the near-term quantum computers. Better performance is obtained by the quantum-enhanced models against the classical counterparts. In particular, quantum models converge much faster than classical ones. The present quantum models are versatile, and can be applied for various tasks of image classification in the field of marine science.

arXiv:2303.03922공식 코드 · Papers with Code

Structure Pretraining and Prompt Tuning for Knowledge Graph Transfer

Knowledge graphs (KG) are essential background knowledge providers in many tasks. When designing models for KG-related tasks, one of the key tasks is to devise the Knowledge Representation and Fusion (KRF) module that learns the representation of elements from KGs and fuses them with task representations. While due to the difference of KGs and perspectives to be considered during fusion across tasks, duplicate and ad hoc KRF modules design are conducted among tasks. In this paper, we propose a novel knowledge graph pretraining model KGTransformer that could serve as a uniform KRF module in diverse KG-related tasks. We pretrain KGTransformer with three self-supervised tasks with sampled sub-graphs as input. For utilization, we propose a general prompt-tuning mechanism regarding task data as a triple prompt to allow flexible interactions between task KGs and task data. We evaluate pretrained KGTransformer on three tasks, triple classification, zero-shot image classification, and question answering. KGTransformer consistently achieves better results than specifically designed task models. Through experiments, we justify that the pretrained KGTransformer could be used off the shelf as a general and effective KRF module across KG-related tasks. The code and datasets are available at https://github.com/zjukg/KGTransformer.

arXiv:2303.03462

Towards Composable Distributions of Latent Space Augmentations

We propose a composable framework for latent space image augmentation that allows for easy combination of multiple augmentations. Image augmentation has been shown to be an effective technique for improving the performance of a wide variety of image classification and generation tasks. Our framework is based on the Variational Autoencoder architecture and uses a novel approach for augmentation via linear transformation within the latent space itself. We explore losses and augmentation latent geometry to enforce the transformations to be composable and involuntary, thus allowing the transformations to be readily combined or inverted. Finally, we show these properties are better performing with certain pairs of augmentations, but we can transfer the latent space to other sets of augmentations to modify performance, effectively constraining the VAE's bottleneck to preserve the variance of specific augmentations and features of the image which we care about. We demonstrate the effectiveness of our approach with initial results on the MNIST dataset against both a standard VAE and a Conditional VAE. This latent augmentation method allows for much greater control and geometric interpretability of the latent space, making it a valuable tool for researchers and practitioners in the field.

arXiv:2303.01672

Self-supervised Learning for Gastrointestinal Pathologies Endoscopy Image Classification with Triplet Loss

Recently, the amount of GI tract datasets is introduced more and more by gathering from contests and challenges. The most common task needs to solve that is to classify images from the GI tract into various classes. However, the contributions of the existing approaches exhibit lots of limitations. In this paper, we aim to develop a computer-aided diagnosis system to classify the pathological findings in endoscopy images, the system can classify some common pathologies including polyps, esophagitis, and ulcerative -- colitis. To evaluate the proposed work, we use the public dataset which is Hyper--Kvasir instead of gathering the data. The key idea of our system is to develop self-supervised learning based on the Barlow Twins framework with a downstream task which is an endoscopy image classification integrated with triplet loss and focal loss functions. The self-supervision framework and focal loss function are used to overcome class-imbalanced data, while the triplet loss function is to tackle the domain-specific properties in endoscopy images which are inter/intra-class problems. An extensive experimental study on the pathological finding images in the Hyper--Kvasir dataset has shown that our proposed system is in general better than the compared methods, whereas using a simple neural network model. This means the proposed system can be used efficiently and capable of accurately for the classification of pathology images in the GI tract.

arXiv:2303.03169공식 코드 · Papers with Code

A Unified Algebraic Perspective on Lipschitz Neural Networks

Important research efforts have focused on the design and training of neural networks with a controlled Lipschitz constant. The goal is to increase and sometimes guarantee the robustness against adversarial attacks. Recent promising techniques draw inspirations from different backgrounds to design 1-Lipschitz neural networks, just to name a few: convex potential layers derive from the discretization of continuous dynamical systems, Almost-Orthogonal-Layer proposes a tailored method for matrix rescaling. However, it is today important to consider the recent and promising contributions in the field under a common theoretical lens to better design new and improved layers. This paper introduces a novel algebraic perspective unifying various types of 1-Lipschitz neural networks, including the ones previously mentioned, along with methods based on orthogonality and spectral methods. Interestingly, we show that many existing techniques can be derived and generalized via finding analytical solutions of a common semidefinite programming (SDP) condition. We also prove that AOL biases the scaled weight to the ones which are close to the set of orthogonal matrices in a certain mathematical manner. Moreover, our algebraic condition, combined with the Gershgorin circle theorem, readily leads to new and diverse parameterizations for 1-Lipschitz network layers. Our approach, called SDP-based Lipschitz Layers (SLL), allows us to design non-trivial yet efficient generalization of convex potential layers. Finally, the comprehensive set of experiments on image classification shows that SLLs outperform previous approaches on certified robust accuracy. Code is available at https://github.com/araujoalexandre/Lipschitz-SLL-Networks.

arXiv:2303.02388

Graph-based Representation for Image based on Granular-ball

Current image processing methods usually operate on the finest-granularity unit; that is, the pixel, which leads to challenges in terms of efficiency, robustness, and understandability in deep learning models. We present an improved granular-ball computing method to represent the image as a graph, in which each node expresses a structural block in the image and each edge represents the association between two nodes. Specifically:(1) We design a gradient-based strategy for the adaptive reorganization of all pixels in the image into numerous rectangular regions, each of which can be regarded as one node. (2) Each node has a connection edge with the nodes with which it shares regions. (3) We design a low-dimensional vector as the attribute of each node. All nodes and their corresponding edges form a graphical representation of a digital image. In the experiments, our proposed graph representation is applied to benchmark datasets for image classification tasks, and the efficiency and good understandability demonstrate that our proposed method offers significant potential in artificial intelligence theory and application.

arXiv:2303.03185

Evaluation of Confidence-based Ensembling in Deep Learning Image Classification

Ensembling is a successful technique to improve the performance of machine learning (ML) models. Conf-Ensemble is an adaptation to Boosting to create ensembles based on model confidence instead of model errors to better classify difficult edge-cases. The key idea is to create successive model experts for samples that were difficult (not necessarily incorrectly classified) by the preceding model. This technique has been shown to provide better results than boosting in binary-classification with a small feature space (~80 features). In this paper, we evaluate the Conf-Ensemble approach in the much more complex task of image classification with the ImageNet dataset (224x224x3 features with 1000 classes). Image classification is an important benchmark for AI-based perception and thus it helps to assess if this method can be used in safety-critical applications using ML ensembles. Our experiments indicate that in a complex multi-label classification task, the expected benefit of specialization on complex input samples cannot be achieved with a small sample set, i.e., a good classifier seems to rely on very complex feature analysis that cannot be well trained on just a limited subset of "difficult samples". We propose an improvement to Conf-Ensemble to increase the number of samples fed to successive ensemble members, and a three-member Conf-Ensemble using this improvement was able to surpass a single model in accuracy, although the amount is not significant. Our findings shed light on the limits of the approach and the non-triviality of harnessing big data.

arXiv:2303.02165공식 코드 · Papers with Code

DeepMAD: Mathematical Architecture Design for Deep Convolutional Neural Network

The rapid advances in Vision Transformer (ViT) refresh the state-of-the-art performances in various vision tasks, overshadowing the conventional CNN-based models. This ignites a few recent striking-back research in the CNN world showing that pure CNN models can achieve as good performance as ViT models when carefully tuned. While encouraging, designing such high-performance CNN models is challenging, requiring non-trivial prior knowledge of network design. To this end, a novel framework termed Mathematical Architecture Design for Deep CNN (DeepMAD) is proposed to design high-performance CNN models in a principled way. In DeepMAD, a CNN network is modeled as an information processing system whose expressiveness and effectiveness can be analytically formulated by their structural parameters. Then a constrained mathematical programming (MP) problem is proposed to optimize these structural parameters. The MP problem can be easily solved by off-the-shelf MP solvers on CPUs with a small memory footprint. In addition, DeepMAD is a pure mathematical framework: no GPU or training data is required during network design. The superiority of DeepMAD is validated on multiple large-scale computer vision benchmark datasets. Notably on ImageNet-1k, only using conventional convolutional layers, DeepMAD achieves 0.7% and 1.5% higher top-1 accuracy than ConvNeXt and Swin on Tiny level, and 0.8% and 0.9% higher on Small level.

arXiv:2303.02400공식 코드 · Papers with Code

Fine-Grained ImageNet Classification in the Wild

Image classification has been one of the most popular tasks in Deep Learning, seeing an abundance of impressive implementations each year. However, there is a lot of criticism tied to promoting complex architectures that continuously push performance metrics higher and higher. Robustness tests can uncover several vulnerabilities and biases which go unnoticed during the typical model evaluation stage. So far, model robustness under distribution shifts has mainly been examined within carefully curated datasets. Nevertheless, such approaches do not test the real response of classifiers in the wild, e.g. when uncurated web-crawled image data of corresponding classes are provided. In our work, we perform fine-grained classification on closely related categories, which are identified with the help of hierarchical knowledge. Extensive experimentation on a variety of convolutional and transformer-based architectures reveals model robustness in this novel setting. Finally, hierarchical knowledge is again employed to evaluate and explain misclassifications, providing an information-rich evaluation scheme adaptable to any classifier.