In recent years, colorectal cancer has become one of the most significant diseases that endanger human health. Deep learning methods are increasingly important for the classification of colorectal histopathology images. However, existing approaches focus more on end-to-end automatic classification using computers rather than human-computer interaction. In this paper, we propose an IL-MCAM framework. It is based on attention mechanisms and interactive learning. The proposed IL-MCAM framework includes two stages: automatic learning (AL) and interactivity learning (IL). In the AL stage, a multi-channel attention mechanism model containing three different attention mechanism channels and convolutional neural networks is used to extract multi-channel features for classification. In the IL stage, the proposed IL-MCAM framework continuously adds misclassified images to the training set in an interactive approach, which improves the classification ability of the MCAM model. We carried out a comparison experiment on our dataset and an extended experiment on the HE-NCT-CRC-100K dataset to verify the performance of the proposed IL-MCAM framework, achieving classification accuracies of 98.98% and 99.77%, respectively. In addition, we conducted an ablation experiment and an interchangeability experiment to verify the ability and interchangeability of the three channels. The experimental results show that the proposed IL-MCAM framework has excellent performance in the colorectal histopathological image classification tasks.
Existing works often focus on reducing the architecture redundancy for accelerating image classification but ignore the spatial redundancy of the input image. This paper proposes an efficient image classification pipeline to solve this problem. We first pinpoint task-aware regions over the input image by a lightweight patch proposal network called AnchorNet. We then feed these localized semantic patches with much smaller spatial redundancy into a general classification network. Unlike the popular design of deep CNN, we aim to carefully design the Receptive Field of AnchorNet without intermediate convolutional paddings. This ensures the exact mapping from a high-level spatial location to the specific input image patch. The contribution of each patch is interpretable. Moreover, AnchorNet is compatible with any downstream architecture. Experimental results on ImageNet show that our method outperforms SOTA dynamic inference methods with fewer inference costs. Our code is available at https://github.com/winycg/AnchorNet.
State-of-the-art computer vision models are mostly trained with supervised learning using human-labeled images, which limits their scalability due to the expensive annotation cost. While self-supervised representation learning has achieved impressive progress, it still requires a second stage of finetuning on labeled data. On the other hand, models pre-trained with large-scale text-image supervision (e.g., CLIP) have enabled zero-shot transfer to downstream image classification tasks. However, the zero-shot performance of CLIP-like models are often insufficient for real-world adoption. In this paper, we aim to leverage the abundant unlabeled data from a target domain to improve the performance of a pre-trained zero-shot classifier, by unsupervised finetuning of the pre-trained model. We propose Masked Unsupervised Self-Training (MUST), a new unsupervised adaptation method which leverages two different and complementary sources of training signals: pseudo-labels and raw images. MUST jointly optimizes three objectives to learn both class-level global feature and pixel-level local feature and enforces a regularization between the two. We demonstrate the efficacy of MUST on a variety of downstream tasks, where it improves upon CLIP by a large margin. MUST also outperforms supervised few-shot adaptation methods. It achieves a top-1 accuracy of 77.7% on ImageNet using ViT-B, +9.4% higher than CLIP, and +6.2% higher than 16-shot CLIP adaptation. Our code is available at https://github.com/salesforce/MUST.
Bayesian neural networks (BNNs) promise improved generalization under covariate shift by providing principled probabilistic representations of epistemic uncertainty. However, weight-based BNNs often struggle with high computational complexity of large-scale architectures and datasets. Node-based BNNs have recently been introduced as scalable alternatives, which induce epistemic uncertainty by multiplying each hidden node with latent random variables, while learning a point-estimate of the weights. In this paper, we interpret these latent noise variables as implicit representations of simple and domain-agnostic data perturbations during training, producing BNNs that perform well under covariate shift due to input corruptions. We observe that the diversity of the implicit corruptions depends on the entropy of the latent variables, and propose a straightforward approach to increase the entropy of these variables during training. We evaluate the method on out-of-distribution image classification benchmarks, and show improved uncertainty estimation of node-based BNNs under covariate shift due to input perturbations. As a side effect, the method also provides robustness against noisy training labels.
As deep neural networks achieve unprecedented performance in various tasks, neural architecture search (NAS), a research field for designing neural network architectures with automated processes, is actively underway. More recently, differentiable NAS has a great impact by reducing the search cost to the level of training a single network. Besides, the search space that defines candidate architectures to be searched directly affects the performance of the final architecture. In this paper, we propose an adaptation scheme of the search space by introducing a search scope. The effectiveness of proposed method is demonstrated with ProxylessNAS for the image classification task. Furthermore, we visualize the trajectory of architecture parameter updates and provide insights to improve the architecture search.
Adaptive gradient methods have become popular in optimizing deep neural networks; recent examples include AdaGrad and Adam. Although Adam usually converges faster, variations of Adam, for instance, the AdaBelief algorithm, have been proposed to enhance Adam's poor generalization ability compared to the classical stochastic gradient method. This paper develops a generic framework for adaptive gradient methods that solve non-convex optimization problems. We first model the adaptive gradient methods in a state-space framework, which allows us to present simpler convergence proofs of adaptive optimizers such as AdaGrad, Adam, and AdaBelief. We then utilize the transfer function paradigm from classical control theory to propose a new variant of Adam, coined AdamSSM. We add an appropriate pole-zero pair in the transfer function from squared gradients to the second moment estimate. We prove the convergence of the proposed AdamSSM algorithm. Applications on benchmark machine learning tasks of image classification using CNN architectures and language modeling using LSTM architecture demonstrate that the AdamSSM algorithm improves the gap between generalization accuracy and faster convergence than the recent adaptive gradient methods.
We consider fine-tuning a pretrained deep neural network on a target task. We study the generalization properties of fine-tuning to understand the problem of overfitting, which has often been observed (e.g., when the target dataset is small or when the training labels are noisy). Existing generalization measures for deep networks depend on notions such as distance from the initialization (i.e., the pretrained network) of the fine-tuned model and noise stability properties of deep networks. This paper identifies a Hessian-based distance measure through PAC-Bayesian analysis, which is shown to correlate well with observed generalization gaps of fine-tuned models. Theoretically, we prove Hessian distance-based generalization bounds for fine-tuned models. We also describe an extended study of fine-tuning against label noise, where overfitting remains a critical problem. We present an algorithm and a generalization error guarantee for this algorithm under a class conditional independent noise model. Empirically, we observe that the Hessian-based distance measure can match the scale of the observed generalization gap of fine-tuned models in practice. We also test our algorithm on several image classification tasks with noisy training labels, showing gains over prior methods and decreases in the Hessian distance measure of the fine-tuned model.
This article describes Jigsaw, a convolutional neural network (CNN) used in geosciences and based on Inception but tailored for geoscientific analyses. Introduces JigsawHSI (based on Jigsaw) and uses it on the land-use land-cover (LULC) classification problem with the Indian Pines, Pavia University and Salinas hyperspectral image data sets. The network is compared against HybridSN, a spectral-spatial 3D-CNN followed by 2D-CNN that achieves state-of-the-art results on the datasets. This short article proves that JigsawHSI is able to meet or exceed HybridSN's performance in all three cases. It also introduces a generalized Jigsaw architecture in d-dimensional space for any number of multimodal inputs. Additionally, the use of jigsaw in geosciences is highlighted, while the code and toolkit are made available.
Contrastive Language-Image Pre-training (CLIP) formulates image classification as an image-to-text matching task, i.e., matching images to the corresponding natural language descriptions instead of discrete category IDs. This allows for open-vocabulary visual recognition, where the model can recognize images from an open class set (also known as an open vocabulary) in a zero-shot manner. However, evaluating the openness of CLIP-like models is challenging, as the models are open to arbitrary vocabulary in theory, but their accuracy varies in practice. To address this, we resort to an incremental perspective to assess the openness through vocabulary expansions, and define extensibility to measure a model's ability to handle novel classes. Our evaluation shows that CLIP-like models are not truly open, and their performance deteriorates as the vocabulary expands. We further dissect the feature space of CLIP from the perspectives of representation alignment and uniformity. Our investigation reveals that the overestimation of openness is due to confusion among competing text features, rather than a failure to capture the similarity between image features and text features of novel classes. We hope that our investigation and analysis will facilitate future research on the CLIP openness issue.
In Continual Learning (CL), a neural network is trained on a stream of data whose distribution changes over time. In this context, the main problem is how to learn new information without forgetting old knowledge (i.e., Catastrophic Forgetting). Most existing CL approaches focus on finding solutions to preserve acquired knowledge, so working on the past of the model. However, we argue that as the model has to continually learn new tasks, it is also important to put focus on the present knowledge that could improve following tasks learning. In this paper we propose a new, simple, CL algorithm that focuses on solving the current task in a way that might facilitate the learning of the next ones. More specifically, our approach combines the main data stream with a secondary, diverse and uncorrelated stream, from which the network can draw auxiliary knowledge. This helps the model from different perspectives, since auxiliary data may contain useful features for the current and the next tasks and incoming task classes can be mapped onto auxiliary classes. Furthermore, the addition of data to the current task is implicitly making the classifier more robust as we are forcing the extraction of more discriminative features. Our method can outperform existing state-of-the-art models on the most common CL Image Classification benchmarks.
Curating a large scale medical imaging dataset for machine learning applications is both time consuming and expensive. Balancing the workload between model development, data collection and annotations is difficult for machine learning practitioners, especially under time constraints. Causal analysis is often used in medicine and economics to gain insights about the effects of actions and policies. In this paper we explore the effect of dataset interventions on the output of image classification models. Through a causal approach we investigate the effects of the quantity and type of data we need to incorporate in a dataset to achieve better performance for specific subtasks. The main goal of this paper is to highlight the potential of causal analysis as a tool for resource optimization for developing medical imaging ML applications. We explore this concept with a synthetic dataset and an exemplary use-case for Diabetic Retinopathy image analysis.
Mobile vision transformers (MobileViT) can achieve state-of-the-art performance across several mobile vision tasks, including classification and detection. Though these models have fewer parameters, they have high latency as compared to convolutional neural network-based models. The main efficiency bottleneck in MobileViT is the multi-headed self-attention (MHA) in transformers, which requires $O(k^2)$ time complexity with respect to the number of tokens (or patches) $k$. Moreover, MHA requires costly operations (e.g., batch-wise matrix multiplication) for computing self-attention, impacting latency on resource-constrained devices. This paper introduces a separable self-attention method with linear complexity, i.e. $O(k)$. A simple yet effective characteristic of the proposed method is that it uses element-wise operations for computing self-attention, making it a good choice for resource-constrained devices. The improved model, MobileViTv2, is state-of-the-art on several mobile vision tasks, including ImageNet object classification and MS-COCO object detection. With about three million parameters, MobileViTv2 achieves a top-1 accuracy of 75.6% on the ImageNet dataset, outperforming MobileViT by about 1% while running $3.2\times$ faster on a mobile device. Our source code is available at: \url{https://github.com/apple/ml-cvnets}
Convolutional Neural Network (CNN) have been widely used in image classification. Over the years, they have also benefited from various enhancements and they are now considered as state of the art techniques for image like data. However, when they are used for regression to estimate some function value from images, fewer recommendations are available. In this study, a novel CNN regression model is proposed. It combines convolutional neural layers to extract high level features representations from images with a soft labelling technique. More specifically, as the deep regression task is challenging, the idea is to account for some uncertainty in the targets that are seen as distributions around their mean. The estimations are carried out by the model in the form of distributions. Building from earlier work, a specific histogram loss function based on the Kullback-Leibler (KL) divergence is applied during training. The model takes advantage of the CNN feature representation and is able to carry out estimation from multi-channel input images. To assess and illustrate the technique, the model is applied to Global Navigation Satellite System (GNSS) multi-path estimation where multi-path signal parameters have to be estimated from correlator output images from the I and Q channels. The multi-path signal delay, magnitude, Doppler shift frequency and phase parameters are estimated from synthetically generated datasets of satellite signals. Experiments are conducted under various receiving conditions and various input images resolutions to test the estimation performances quality and robustness. The results show that the proposed soft labelling CNN technique using distributional loss outperforms classical CNN regression under all conditions. Furthermore, the extra learning performance achieved by the model allows the reduction of input image resolution from 80x80 down to 40x40 or sometimes 20x20.
By replacing the lens with a thin optical element, lensless imaging enables new applications and solutions beyond those supported by traditional camera design and post-processing, e.g. compact and lightweight form factors and visual privacy. The latter arises from the highly multiplexed measurements of lensless cameras, which require knowledge of the imaging system to recover a recognizable image. In this work, we exploit this unique multiplexing property: casting the optics as an encoder that produces learned embeddings directly at the camera sensor. We do so in the context of image classification, where we jointly optimize the encoder's parameters and those of an image classifier in an end-to-end fashion. Our experiments show that jointly learning the lensless optical encoder and the digital processing allows for lower resolution embeddings at the sensor, and hence better privacy as it is much harder to recover meaningful images from these measurements. Additional experiments show that such an optimization allows for lensless measurements that are more robust to typical real-world image transformations. While this work focuses on classification, the proposed programmable lensless camera and end-to-end optimization can be applied to other computational imaging tasks.
Efficient deployment of deep neural networks across many devices and resource constraints, particularly on edge devices, is one of the most challenging problems in the presence of data-privacy preservation issues. Conventional approaches have evolved to either improve a single global model while keeping each local heterogeneous training data decentralized (i.e. data heterogeneity; Federated Learning (FL)) or to train an overarching network that supports diverse architectural settings to address heterogeneous systems equipped with different computational capabilities (i.e. system heterogeneity; Neural Architecture Search). However, few studies have considered both directions simultaneously. This paper proposes the federation of supernet training (FedSup) framework to consider both scenarios simultaneously, i.e., where clients send and receive a supernet that contains all possible architectures sampled from itself. The approach is inspired by observing that averaging parameters during model aggregation for FL is similar to weight-sharing in supernet training. Thus, the proposed FedSup framework combines a weight-sharing approach widely used for training single shot models with FL averaging (FedAvg). Furthermore, we develop an efficient algorithm (E-FedSup) by sending the sub-model to clients on the broadcast stage to reduce communication costs and training overhead, including several strategies to enhance supernet training in the FL environment. We verify the proposed approach with extensive empirical evaluations. The resulting framework also ensures data and model heterogeneity robustness on several standard benchmarks.
Neural ordinary differential equations (neural ODEs) have emerged as a novel network architecture that bridges dynamical systems and deep learning. However, the gradient obtained with the continuous adjoint method in the vanilla neural ODE is not reverse-accurate. Other approaches suffer either from an excessive memory requirement due to deep computational graphs or from limited choices for the time integration scheme, hampering their application to large-scale complex dynamical systems. To achieve accurate gradients without compromising memory efficiency and flexibility, we present a new neural ODE framework, PNODE, based on high-level discrete adjoint algorithmic differentiation. By leveraging discrete adjoint time integrators and advanced checkpointing strategies tailored for these integrators, PNODE can provide a balance between memory and computational costs, while computing the gradients consistently and accurately. We provide an open-source implementation based on PyTorch and PETSc, one of the most commonly used portable, scalable scientific computing libraries. We demonstrate the performance through extensive numerical experiments on image classification and continuous normalizing flow problems. We show that PNODE achieves the highest memory efficiency when compared with other reverse-accurate methods. On the image classification problems, PNODE is up to two times faster than the vanilla neural ODE and up to 2.3 times faster than the best existing reverse-accurate method. We also show that PNODE enables the use of the implicit time integration methods that are needed for stiff dynamical systems.
In response to the situation that the conventional bridge crack manual detection method has a large amount of human and material resources wasted, this study is aimed to propose a light-weighted, high-precision, deep learning-based bridge apparent crack recognition model that can be deployed in mobile devices' scenarios. In order to enhance the performance of YOLOv5, firstly, the data augmentation methods are supplemented, and then the YOLOv5 series algorithm is trained to select a suitable basic framework. The YOLOv5s is identified as the basic framework for the light-weighted crack detection model through experiments for comparison and validation.By replacing the traditional DarkNet backbone network of YOLOv5s with GhostNet backbone network, introducing Transformer multi-headed self-attention mechanism and bi-directional feature pyramid network (BiFPN) to replace the commonly used feature pyramid network, the improved model not only has 42% fewer parameters and faster inference response, but also significantly outperforms the original model in terms of accuracy and mAP (8.5% and 1.1% improvement, respectively). Luckily each improved part has a positive impact on the result. This paper provides a feasible idea to establish a digital operation management system in the field of highway and bridge in the future and to implement the whole life cycle structure health monitoring of civil infrastructure in China.
Federated learning (FL) is a privacy-promoting framework that enables potentially large number of clients to collaboratively train machine learning models. In a FL system, a server coordinates the collaboration by collecting and aggregating clients' model updates while the clients' data remains local and private. A major challenge in federated learning arises when the local data is heterogeneous -- the setting in which performance of the learned global model may deteriorate significantly compared to the scenario where the data is identically distributed across the clients. In this paper we propose FedDPMS (Federated Differentially Private Means Sharing), an FL algorithm in which clients deploy variational auto-encoders to augment local datasets with data synthesized using differentially private means of latent data representations communicated by a trusted server. Such augmentation ameliorates effects of data heterogeneity across the clients without compromising privacy. Our experiments on deep image classification tasks demonstrate that FedDPMS outperforms competing state-of-the-art FL methods specifically designed for heterogeneous data settings.
It has been observed that visual classification models often rely mostly on the image background, neglecting the foreground, which hurts their robustness to distribution changes. To alleviate this shortcoming, we propose to monitor the model's relevancy signal and manipulate it such that the model is focused on the foreground object. This is done as a finetuning step, involving relatively few samples consisting of pairs of images and their associated foreground masks. Specifically, we encourage the model's relevancy map (i) to assign lower relevance to background regions, (ii) to consider as much information as possible from the foreground, and (iii) we encourage the decisions to have high confidence. When applied to Vision Transformer (ViT) models, a marked improvement in robustness to domain shifts is observed. Moreover, the foreground masks can be obtained automatically, from a self-supervised variant of the ViT model itself; therefore no additional supervision is required.
We introduce a vision-language foundation model called VL-BEiT, which is a bidirectional multimodal Transformer learned by generative pretraining. Our minimalist solution conducts masked prediction on both monomodal and multimodal data with a shared Transformer. Specifically, we perform masked vision-language modeling on image-text pairs, masked language modeling on texts, and masked image modeling on images. VL-BEiT is learned from scratch with one unified pretraining task, one shared backbone, and one-stage training. Our method is conceptually simple and empirically effective. Experimental results show that VL-BEiT obtains strong results on various vision-language benchmarks, such as visual question answering, visual reasoning, and image-text retrieval. Moreover, our method learns transferable visual features, achieving competitive performance on image classification, and semantic segmentation.