PyTorchKR Papers

최신 논문

총 9,676편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2309.10313

Investigating the Catastrophic Forgetting in Multimodal Large Language Models

Following the success of GPT4, there has been a surge in interest in multimodal large language model (MLLM) research. This line of research focuses on developing general-purpose LLMs through fine-tuning pre-trained LLMs and vision models. However, catastrophic forgetting, a notorious phenomenon where the fine-tuned model fails to retain similar performance compared to the pre-trained model, still remains an inherent problem in multimodal LLMs (MLLM). In this paper, we introduce EMT: Evaluating MulTimodality for evaluating the catastrophic forgetting in MLLMs, by treating each MLLM as an image classifier. We first apply EMT to evaluate several open-source fine-tuned MLLMs and we discover that almost all evaluated MLLMs fail to retain the same performance levels as their vision encoders on standard image classification tasks. Moreover, we continue fine-tuning LLaVA, an MLLM and utilize EMT to assess performance throughout the fine-tuning. Interestingly, our results suggest that early-stage fine-tuning on an image dataset improves performance across other image datasets, by enhancing the alignment of text and visual features. However, as fine-tuning proceeds, the MLLMs begin to hallucinate, resulting in a significant loss of generalizability, even when the image encoder remains frozen. Our results suggest that MLLMs have yet to demonstrate performance on par with their vision models on standard image classification tasks and the current MLLM fine-tuning procedure still has room for improvement.

arXiv:2309.08610

Do the Frankenstein, or how to achieve better out-of-distribution performance with manifold mixing model soup

The standard recipe applied in transfer learning is to finetune a pretrained model on the task-specific dataset with different hyperparameter settings and pick the model with the highest accuracy on the validation dataset. Unfortunately, this leads to models which do not perform well under distribution shifts, e.g. when the model is given graphical sketches of the object as input instead of photos. In order to address this, we propose the manifold mixing model soup, an algorithm which mixes together the latent space manifolds of multiple finetuned models in an optimal way in order to generate a fused model. We show that the fused model gives significantly better out-of-distribution performance (+3.5 % compared to best individual model) when finetuning a CLIP model for image classification. In addition, it provides also better accuracy on the original dataset where the finetuning has been done.

arXiv:2309.09276

MVP: Meta Visual Prompt Tuning for Few-Shot Remote Sensing Image Scene Classification

Vision Transformer (ViT) models have recently emerged as powerful and versatile models for various visual tasks. Recently, a work called PMF has achieved promising results in few-shot image classification by utilizing pre-trained vision transformer models. However, PMF employs full fine-tuning for learning the downstream tasks, leading to significant overfitting and storage issues, especially in the remote sensing domain. In order to tackle these issues, we turn to the recently proposed parameter-efficient tuning methods, such as VPT, which updates only the newly added prompt parameters while keeping the pre-trained backbone frozen. Inspired by VPT, we propose the Meta Visual Prompt Tuning (MVP) method. Specifically, we integrate the VPT method into the meta-learning framework and tailor it to the remote sensing domain, resulting in an efficient framework for Few-Shot Remote Sensing Scene Classification (FS-RSSC). Furthermore, we introduce a novel data augmentation strategy based on patch embedding recombination to enhance the representation and diversity of scenes for classification purposes. Experiment results on the FS-RSSC benchmark demonstrate the superior performance of the proposed MVP over existing methods in various settings, such as various-way-various-shot, various-way-one-shot, and cross-domain adaptation.

arXiv:2309.08744

Personalized Food Image Classification: Benchmark Datasets and New Baseline

Food image classification is a fundamental step of image-based dietary assessment, enabling automated nutrient analysis from food images. Many current methods employ deep neural networks to train on generic food image datasets that do not reflect the dynamism of real-life food consumption patterns, in which food images appear sequentially over time, reflecting the progression of what an individual consumes. Personalized food classification aims to address this problem by training a deep neural network using food images that reflect the consumption pattern of each individual. However, this problem is under-explored and there is a lack of benchmark datasets with individualized food consumption patterns due to the difficulty in data collection. In this work, we first introduce two benchmark personalized datasets including the Food101-Personal, which is created based on surveys of daily dietary patterns from participants in the real world, and the VFNPersonal, which is developed based on a dietary study. In addition, we propose a new framework for personalized food image classification by leveraging self-supervised learning and temporal image feature information. Our method is evaluated on both benchmark datasets and shows improved performance compared to existing works. The dataset has been made available at: https://skynet.ecn.purdue.edu/~pan161/dataset_personal.html

arXiv:2309.09901

The role of causality in explainable artificial intelligence

Causality and eXplainable Artificial Intelligence (XAI) have developed as separate fields in computer science, even though the underlying concepts of causation and explanation share common ancient roots. This is further enforced by the lack of review works jointly covering these two fields. In this paper, we investigate the literature to try to understand how and to what extent causality and XAI are intertwined. More precisely, we seek to uncover what kinds of relationships exist between the two concepts and how one can benefit from them, for instance, in building trust in AI systems. As a result, three main perspectives are identified. In the first one, the lack of causality is seen as one of the major limitations of current AI and XAI approaches, and the "optimal" form of explanations is investigated. The second is a pragmatic perspective and considers XAI as a tool to foster scientific exploration for causal inquiry, via the identification of pursue-worthy experimental manipulations. Finally, the third perspective supports the idea that causality is propaedeutic to XAI in three possible manners: exploiting concepts borrowed from causality to support or improve XAI, utilizing counterfactuals for explainability, and considering accessing a causal model as explaining itself. To complement our analysis, we also provide relevant software solutions used to automate causal tasks. We believe our work provides a unified view of the two fields of causality and XAI by highlighting potential domain bridges and uncovering possible limitations.

arXiv:2309.09571

Heterogeneous Generative Knowledge Distillation with Masked Image Modeling

Small CNN-based models usually require transferring knowledge from a large model before they are deployed in computationally resource-limited edge devices. Masked image modeling (MIM) methods achieve great success in various visual tasks but remain largely unexplored in knowledge distillation for heterogeneous deep models. The reason is mainly due to the significant discrepancy between the Transformer-based large model and the CNN-based small network. In this paper, we develop the first Heterogeneous Generative Knowledge Distillation (H-GKD) based on MIM, which can efficiently transfer knowledge from large Transformer models to small CNN-based models in a generative self-supervised fashion. Our method builds a bridge between Transformer-based models and CNNs by training a UNet-style student with sparse convolution, which can effectively mimic the visual representation inferred by a teacher over masked modeling. Our method is a simple yet effective learning paradigm to learn the visual representation and distribution of data from heterogeneous teacher models, which can be pre-trained using advanced generative methods. Extensive experiments show that it adapts well to various models and sizes, consistently achieving state-of-the-art performance in image classification, object detection, and semantic segmentation tasks. For example, in the Imagenet 1K dataset, H-GKD improves the accuracy of Resnet50 (sparse) from 76.98% to 80.01%.

arXiv:2309.08760공식 코드 · Papers with Code

Biased Attention: Do Vision Transformers Amplify Gender Bias More than Convolutional Neural Networks?

Deep neural networks used in computer vision have been shown to exhibit many social biases such as gender bias. Vision Transformers (ViTs) have become increasingly popular in computer vision applications, outperforming Convolutional Neural Networks (CNNs) in many tasks such as image classification. However, given that research on mitigating bias in computer vision has primarily focused on CNNs, it is important to evaluate the effect of a different network architecture on the potential for bias amplification. In this paper we therefore introduce a novel metric to measure bias in architectures, Accuracy Difference. We examine bias amplification when models belonging to these two architectures are used as a part of large multimodal models, evaluating the different image encoders of Contrastive Language Image Pretraining which is an important model used in many generative models such as DALL-E and Stable Diffusion. Our experiments demonstrate that architecture can play a role in amplifying social biases due to the different techniques employed by the models for feature extraction and embedding as well as their different learning properties. This research found that ViTs amplified gender bias to a greater extent than CNNs

arXiv:2309.08045공식 코드 · Papers with Code

Traveling Waves Encode the Recent Past and Enhance Sequence Learning

Traveling waves of neural activity have been observed throughout the brain at a diversity of regions and scales; however, their precise computational role is still debated. One physically inspired hypothesis suggests that the cortical sheet may act like a wave-propagating system capable of invertibly storing a short-term memory of sequential stimuli through induced waves traveling across the cortical surface, and indeed many experimental results from neuroscience correlate wave activity with memory tasks. To date, however, the computational implications of this idea have remained hypothetical due to the lack of a simple recurrent neural network architecture capable of exhibiting such waves. In this work, we introduce a model to fill this gap, which we denote the Wave-RNN (wRNN), and demonstrate how such an architecture indeed efficiently encodes the recent past through a suite of synthetic memory tasks where wRNNs learn faster and reach significantly lower error than wave-free counterparts. We further explore the implications of this memory storage system on more complex sequence modeling tasks such as sequential image classification and find that wave-based models not only again outperform comparable wave-free RNNs while using significantly fewer parameters, but additionally perform comparably to more complex gated architectures such as LSTMs and GRUs.

arXiv:2309.08035공식 코드 · Papers with Code

Interpretability-Aware Vision Transformer

Vision Transformers (ViTs) have become prominent models for solving various vision tasks. However, the interpretability of ViTs has not kept pace with their promising performance. While there has been a surge of interest in developing {\it post hoc} solutions to explain ViTs' outputs, these methods do not generalize to different downstream tasks and various transformer architectures. Furthermore, if ViTs are not properly trained with the given data and do not prioritize the region of interest, the {\it post hoc} methods would be less effective. Instead of developing another {\it post hoc} approach, we introduce a novel training procedure that inherently enhances model interpretability. Our interpretability-aware ViT (IA-ViT) draws inspiration from a fresh insight: both the class patch and image patches consistently generate predicted distributions and attention maps. IA-ViT is composed of a feature extractor, a predictor, and an interpreter, which are trained jointly with an interpretability-aware training objective. Consequently, the interpreter simulates the behavior of the predictor and provides a faithful explanation through its single-head self-attention mechanism. Our comprehensive experimental results demonstrate the effectiveness of IA-ViT in several image classification tasks, with both qualitative and quantitative evaluations of model performance and interpretability. Source code is available from: https://github.com/qiangyao1988/IA-ViT.

arXiv:2309.08395공식 코드 · Papers with Code

Learning by Self-Explaining

Much of explainable AI research treats explanations as a means for model inspection. Yet, this neglects findings from human psychology that describe the benefit of self-explanations in an agent's learning process. Motivated by this, we introduce a novel workflow in the context of image classification, termed Learning by Self-Explaining (LSX). LSX utilizes aspects of self-refining AI and human-guided explanatory machine learning. The underlying idea is that a learner model, in addition to optimizing for the original predictive task, is further optimized based on explanatory feedback from an internal critic model. Intuitively, a learner's explanations are considered "useful" if the internal critic can perform the same task given these explanations. We provide an overview of important components of LSX and, based on this, perform extensive experimental evaluations via three different example instantiations. Our results indicate improvements via Learning by Self-Explaining on several levels: in terms of model generalization, reducing the influence of confounding factors, and providing more task-relevant and faithful model explanations. Overall, our work provides evidence for the potential of self-explaining within the learning phase of an AI model.

arXiv:2309.08353공식 코드 · Papers with Code

Continual Learning with Deep Streaming Regularized Discriminant Analysis

Continual learning is increasingly sought after in real world machine learning applications, as it enables learning in a more human-like manner. Conventional machine learning approaches fail to achieve this, as incrementally updating the model with non-identically distributed data leads to catastrophic forgetting, where existing representations are overwritten. Although traditional continual learning methods have mostly focused on batch learning, which involves learning from large collections of labeled data sequentially, this approach is not well-suited for real-world applications where we would like new data to be integrated directly. This necessitates a paradigm shift towards streaming learning. In this paper, we propose a streaming version of regularized discriminant analysis as a solution to this challenge. We combine our algorithm with a convolutional neural network and demonstrate that it outperforms both batch learning and existing streaming learning algorithms on the ImageNet ILSVRC-2012 dataset.

arXiv:2309.07909공식 코드 · Papers with Code

DiffAug: Enhance Unsupervised Contrastive Learning with Domain-Knowledge-Free Diffusion-based Data Augmentation

Unsupervised Contrastive learning has gained prominence in fields such as vision, and biology, leveraging predefined positive/negative samples for representation learning. Data augmentation, categorized into hand-designed and model-based methods, has been identified as a crucial component for enhancing contrastive learning. However, hand-designed methods require human expertise in domain-specific data while sometimes distorting the meaning of the data. In contrast, generative model-based approaches usually require supervised or large-scale external data, which has become a bottleneck constraining model training in many domains. To address the problems presented above, this paper proposes DiffAug, a novel unsupervised contrastive learning technique with diffusion mode-based positive data generation. DiffAug consists of a semantic encoder and a conditional diffusion model; the conditional diffusion model generates new positive samples conditioned on the semantic encoding to serve the training of unsupervised contrast learning. With the help of iterative training of the semantic encoder and diffusion model, DiffAug improves the representation ability in an uninterrupted and unsupervised manner. Experimental evaluations show that DiffAug outperforms hand-designed and SOTA model-based augmentation methods on DNA sequence, visual, and bio-feature datasets. The code for review is released at \url{https://github.com/zangzelin/code_diffaug}.

arXiv:2309.07186

LCReg: Long-Tailed Image Classification with Latent Categories based Recognition

In this work, we tackle the challenging problem of long-tailed image recognition. Previous long-tailed recognition approaches mainly focus on data augmentation or re-balancing strategies for the tail classes to give them more attention during model training. However, these methods are limited by the small number of training images for the tail classes, which results in poor feature representations. To address this issue, we propose the Latent Categories based long-tail Recognition (LCReg) method. Our hypothesis is that common latent features shared by head and tail classes can be used to improve feature representation. Specifically, we learn a set of class-agnostic latent features shared by both head and tail classes, and then use semantic data augmentation on the latent features to implicitly increase the diversity of the training sample. We conduct extensive experiments on five long-tailed image recognition datasets, and the results show that our proposed method significantly improves the baselines.

arXiv:2309.07149공식 코드 · Papers with Code

Decoding visual brain representations from electroencephalography through Knowledge Distillation and latent diffusion models

Decoding visual representations from human brain activity has emerged as a thriving research domain, particularly in the context of brain-computer interfaces. Our study presents an innovative method that employs to classify and reconstruct images from the ImageNet dataset using electroencephalography (EEG) data from subjects that had viewed the images themselves (i.e. "brain decoding"). We analyzed EEG recordings from 6 participants, each exposed to 50 images spanning 40 unique semantic categories. These EEG readings were converted into spectrograms, which were then used to train a convolutional neural network (CNN), integrated with a knowledge distillation procedure based on a pre-trained Contrastive Language-Image Pre-Training (CLIP)-based image classification teacher network. This strategy allowed our model to attain a top-5 accuracy of 80%, significantly outperforming a standard CNN and various RNN-based benchmarks. Additionally, we incorporated an image reconstruction mechanism based on pre-trained latent diffusion models, which allowed us to generate an estimate of the images which had elicited EEG activity. Therefore, our architecture not only decodes images from neural activity but also offers a credible image reconstruction from EEG only, paving the way for e.g. swift, individualized feedback experiments. Our research represents a significant step forward in connecting neural signals with visual cognition.

arXiv:2309.07113

Contrastive Deep Encoding Enables Uncertainty-aware Machine-learning-assisted Histopathology

Deep neural network models can learn clinically relevant features from millions of histopathology images. However generating high-quality annotations to train such models for each hospital, each cancer type, and each diagnostic task is prohibitively laborious. On the other hand, terabytes of training data -- while lacking reliable annotations -- are readily available in the public domain in some cases. In this work, we explore how these large datasets can be consciously utilized to pre-train deep networks to encode informative representations. We then fine-tune our pre-trained models on a fraction of annotated training data to perform specific downstream tasks. We show that our approach can reach the state-of-the-art (SOTA) for patch-level classification with only 1-10% randomly selected annotations compared to other SOTA approaches. Moreover, we propose an uncertainty-aware loss function, to quantify the model confidence during inference. Quantified uncertainty helps experts select the best instances to label for further training. Our uncertainty-aware labeling reaches the SOTA with significantly fewer annotations compared to random labeling. Last, we demonstrate how our pre-trained encoders can surpass current SOTA for whole-slide image classification with weak supervision. Our work lays the foundation for data and task-agnostic pre-trained deep networks with quantified uncertainty.

arXiv:2309.06891공식 코드 · Papers with Code

Keep It SimPool: Who Said Supervised Transformers Suffer from Attention Deficit?

Convolutional networks and vision transformers have different forms of pairwise interactions, pooling across layers and pooling at the end of the network. Does the latter really need to be different? As a by-product of pooling, vision transformers provide spatial attention for free, but this is most often of low quality unless self-supervised, which is not well studied. Is supervision really the problem? In this work, we develop a generic pooling framework and then we formulate a number of existing methods as instantiations. By discussing the properties of each group of methods, we derive SimPool, a simple attention-based pooling mechanism as a replacement of the default one for both convolutional and transformer encoders. We find that, whether supervised or self-supervised, this improves performance on pre-training and downstream tasks and provides attention maps delineating object boundaries in all cases. One could thus call SimPool universal. To our knowledge, we are the first to obtain attention maps in supervised transformers of at least as good quality as self-supervised, without explicit losses or modifying the architecture. Code at: https://github.com/billpsomas/simpool.

arXiv:2309.06724

Deep Nonparametric Convexified Filtering for Computational Photography, Image Synthesis and Adversarial Defense

We aim to provide a general framework of for computational photography that recovers the real scene from imperfect images, via the Deep Nonparametric Convexified Filtering (DNCF). It is consists of a nonparametric deep network to resemble the physical equations behind the image formation, such as denoising, super-resolution, inpainting, and flash. DNCF has no parameterization dependent on training data, therefore has a strong generalization and robustness to adversarial image manipulation. During inference, we also encourage the network parameters to be nonnegative and create a bi-convex function on the input and parameters, and this adapts to second-order optimization algorithms with insufficient running time, having 10X acceleration over Deep Image Prior. With these tools, we empirically verify its capability to defend image classification deep networks against adversary attack algorithms in real-time.

arXiv:2309.06626

Accelerating Deep Neural Networks via Semi-Structured Activation Sparsity

The demand for efficient processing of deep neural networks (DNNs) on embedded devices is a significant challenge limiting their deployment. Exploiting sparsity in the network's feature maps is one of the ways to reduce its inference latency. It is known that unstructured sparsity results in lower accuracy degradation with respect to structured sparsity but the former needs extensive inference engine changes to get latency benefits. To tackle this challenge, we propose a solution to induce semi-structured activation sparsity exploitable through minor runtime modifications. To attain high speedup levels at inference time, we design a sparse training procedure with awareness of the final position of the activations while computing the General Matrix Multiplication (GEMM). We extensively evaluate the proposed solution across various models for image classification and object detection tasks. Remarkably, our approach yields a speed improvement of $1.25 \times$ with a minimal accuracy drop of $1.1\%$ for the ResNet18 model on the ImageNet dataset. Furthermore, when combined with a state-of-the-art structured pruning method, the resulting models provide a good latency-accuracy trade-off, outperforming models that solely employ structured pruning techniques.

arXiv:2309.06528

Strong-Weak Integrated Semi-supervision for Unsupervised Single and Multi Target Domain Adaptation

Unsupervised domain adaptation (UDA) focuses on transferring knowledge learned in the labeled source domain to the unlabeled target domain. Despite significant progress that has been achieved in single-target domain adaptation for image classification in recent years, the extension from single-target to multi-target domain adaptation is still a largely unexplored problem area. In general, unsupervised domain adaptation faces a major challenge when attempting to learn reliable information from a single unlabeled target domain. Increasing the number of unlabeled target domains further exacerbate the problem rather significantly. In this paper, we propose a novel strong-weak integrated semi-supervision (SWISS) learning strategy for image classification using unsupervised domain adaptation that works well for both single-target and multi-target scenarios. Under the proposed SWISS-UDA framework, a strong representative set with high confidence but low diversity target domain samples and a weak representative set with low confidence but high diversity target domain samples are updated constantly during the training process. Both sets are fused to generate an augmented strong-weak training batch with pseudo-labels to train the network during every iteration. The extension from single-target to multi-target domain adaptation is accomplished by exploring the class-wise distance relationship between domains and replacing the strong representative set with much stronger samples from peer domains via peer scaffolding. Moreover, a novel adversarial logit loss is proposed to reduce the intra-class divergence between source and target domains, which is back-propagated adversarially with a gradient reverse layer between the classifier and the rest of the network. Experimental results based on three benchmarks, Office-31, Office-Home, and DomainNet, show the effectiveness of the proposed SWISS framework.

arXiv:2309.07085

Mitigating Group Bias in Federated Learning for Heterogeneous Devices

Federated Learning is emerging as a privacy-preserving model training approach in distributed edge applications. As such, most edge deployments are heterogeneous in nature i.e., their sensing capabilities and environments vary across deployments. This edge heterogeneity violates the independence and identical distribution (IID) property of local data across clients and produces biased global models i.e. models that contribute to unfair decision-making and discrimination against a particular community or a group. Existing bias mitigation techniques only focus on bias generated from label heterogeneity in non-IID data without accounting for domain variations due to feature heterogeneity and do not address global group-fairness property. Our work proposes a group-fair FL framework that minimizes group-bias while preserving privacy and without resource utilization overhead. Our main idea is to leverage average conditional probabilities to compute a cross-domain group \textit{importance weights} derived from heterogeneous training data to optimize the performance of the worst-performing group using a modified multiplicative weights update method. Additionally, we propose regularization techniques to minimize the difference between the worst and best-performing groups while making sure through our thresholding mechanism to strike a balance between bias reduction and group performance degradation. Our evaluation of human emotion recognition and image classification benchmarks assesses the fair decision-making of our framework in real-world heterogeneous settings.