PyTorchKR Papers

최신 논문

총 10,240편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2303.10594공식 코드 · Papers with Code

AdaptGuard: Defending Against Universal Attacks for Model Adaptation

Model adaptation aims at solving the domain transfer problem under the constraint of only accessing the pretrained source models. With the increasing considerations of data privacy and transmission efficiency, this paradigm has been gaining recent popularity. This paper studies the vulnerability to universal attacks transferred from the source domain during model adaptation algorithms due to the existence of malicious providers. We explore both universal adversarial perturbations and backdoor attacks as loopholes on the source side and discover that they still survive in the target models after adaptation. To address this issue, we propose a model preprocessing framework, named AdaptGuard, to improve the security of model adaptation algorithms. AdaptGuard avoids direct use of the risky source parameters through knowledge distillation and utilizes the pseudo adversarial samples under adjusted radius to enhance the robustness. AdaptGuard is a plug-and-play module that requires neither robust pretrained models nor any changes for the following model adaptation algorithms. Extensive results on three commonly used datasets and two popular adaptation methods validate that AdaptGuard can effectively defend against universal attacks and maintain clean accuracy in the target domain simultaneously. We hope this research will shed light on the safety and robustness of transfer learning. Code is available at https://github.com/TomSheng21/AdaptGuard.

arXiv:2303.10608

A model is worth tens of thousands of examples

Traditional signal processing methods relying on mathematical data generation models have been cast aside in favour of deep neural networks, which require vast amounts of data. Since the theoretical sample complexity is nearly impossible to evaluate, these amounts of examples are usually estimated with crude rules of thumb. However, these rules only suggest when the networks should work, but do not relate to the traditional methods. In particular, an interesting question is: how much data is required for neural networks to be on par or outperform, if possible, the traditional model-based methods? In this work, we empirically investigate this question in two simple examples, where the data is generated according to precisely defined mathematical models, and where well-understood optimal or state-of-the-art mathematical data-agnostic solutions are known. A first problem is deconvolving one-dimensional Gaussian signals and a second one is estimating a circle's radius and location in random grayscale images of disks. By training various networks, either naive custom designed or well-established ones, with various amounts of training data, we find that networks require tens of thousands of examples in comparison to the traditional methods, whether the networks are trained from scratch or even with transfer-learning or finetuning.

arXiv:2303.11325공식 코드 · Papers with Code

GeoMIM: Towards Better 3D Knowledge Transfer via Masked Image Modeling for Multi-view 3D Understanding

Multi-view camera-based 3D detection is a challenging problem in computer vision. Recent works leverage a pretrained LiDAR detection model to transfer knowledge to a camera-based student network. However, we argue that there is a major domain gap between the LiDAR BEV features and the camera-based BEV features, as they have different characteristics and are derived from different sources. In this paper, we propose Geometry Enhanced Masked Image Modeling (GeoMIM) to transfer the knowledge of the LiDAR model in a pretrain-finetune paradigm for improving the multi-view camera-based 3D detection. GeoMIM is a multi-camera vision transformer with Cross-View Attention (CVA) blocks that uses LiDAR BEV features encoded by the pretrained BEV model as learning targets. During pretraining, GeoMIM's decoder has a semantic branch completing dense perspective-view features and the other geometry branch reconstructing dense perspective-view depth maps. The depth branch is designed to be camera-aware by inputting the camera's parameters for better transfer capability. Extensive results demonstrate that GeoMIM outperforms existing methods on nuScenes benchmark, achieving state-of-the-art performance for camera-based 3D object detection and 3D segmentation. Code and pretrained models are available at https://github.com/Sense-X/GeoMIM.

arXiv:2303.11223

Monocular Cyclist Detection with Convolutional Neural Networks

Cycling is an increasingly popular method of transportation for sustainability and health benefits. However, cyclists face growing risks, especially when encountering large vehicles on the road. This study aims to reduce the number of vehicle-cyclist collisions, which are often caused by poor driver attention to blind spots. To achieve this, we designed a state-of-the-art real-time monocular cyclist detection that can detect cyclists with object detection convolutional neural networks, such as EfficientDet Lite and SSD MobileNetV2. First, our proposed cyclist detection models achieve greater than 0.900 mAP (IoU: 0.5), fine-tuned on a newly proposed cyclist image dataset comprising over 20,000 images. Next, the models were deployed onto a Google Coral Dev Board mini-computer with a camera module and analyzed for speed, reaching inference times as low as 15 milliseconds. Lastly, the end-to-end cyclist detection device was tested in real-time to model traffic scenarios and analyzed further for performance and feasibility. We concluded that this cyclist detection device can accurately and quickly detect cyclists and has the potential to improve cyclist safety significantly. Future studies could determine the feasibility of the proposed device in the vehicle industry and improvements to cyclist safety over time.

arXiv:2303.09639

Neural Architecture Search for Effective Teacher-Student Knowledge Transfer in Language Models

Large pretrained language models have achieved state-of-the-art results on a variety of downstream tasks. Knowledge Distillation (KD) into a smaller student model addresses their inefficiency, allowing for deployment in resource-constrained environments. However, KD can be ineffective when the student is manually selected from a set of existing options, since it can be a sub-optimal choice within the space of all possible student architectures. We develop multilingual KD-NAS, the use of Neural Architecture Search (NAS) guided by KD to find the optimal student architecture for task agnostic distillation from a multilingual teacher. In each episode of the search process, a NAS controller predicts a reward based on the distillation loss and latency of inference. The top candidate architectures are then distilled from the teacher on a small proxy set. Finally the architecture(s) with the highest reward is selected, and distilled on the full training corpus. KD-NAS can automatically trade off efficiency and effectiveness, and recommends architectures suitable to various latency budgets. Using our multi-layer hidden state distillation process, our KD-NAS student model achieves a 7x speedup on CPU inference (2x on GPU) compared to a XLM-Roberta Base Teacher, while maintaining 90% performance, and has been deployed in 3 software offerings requiring large throughput, low latency and deployment on CPU.

arXiv:2303.09992

LION: Implicit Vision Prompt Tuning

Despite recent competitive performance across a range of vision tasks, vision Transformers still have an issue of heavy computational costs. Recently, vision prompt learning has provided an economic solution to this problem without fine-tuning the whole large-scale models. However, the efficiency of existing models are still far from satisfactory due to insertion of extensive prompts blocks and trick prompt designs. In this paper, we propose an efficient vision model named impLicit vIsion prOmpt tuNing (LION), which is motivated by deep implicit models with stable memory costs for various complex tasks. In particular, we merely insect two equilibrium implicit layers in two ends of the pre-trained main backbone with parameters in the backbone frozen. Moreover, we prune the parameters in these two layers according to lottery hypothesis. The performance obtained by our LION are promising on a wide range of datasets. In particular, our LION reduces up to 11.5% of training parameter numbers while obtaining higher performance compared with the state-of-the-art baseline VPT, especially under challenging scenes. Furthermore, we find that our proposed LION had a good generalization performance, making it an easy way to boost transfer learning in the future.

arXiv:2303.09769공식 코드 · Papers with Code

Denoising Diffusion Autoencoders are Unified Self-supervised Learners

Inspired by recent advances in diffusion models, which are reminiscent of denoising autoencoders, we investigate whether they can acquire discriminative representations for classification via generative pre-training. This paper shows that the networks in diffusion models, namely denoising diffusion autoencoders (DDAE), are unified self-supervised learners: by pre-training on unconditional image generation, DDAE has already learned strongly linear-separable representations within its intermediate layers without auxiliary encoders, thus making diffusion pre-training emerge as a general approach for generative-and-discriminative dual learning. To validate this, we conduct linear probe and fine-tuning evaluations. Our diffusion-based approach achieves 95.9% and 50.0% linear evaluation accuracies on CIFAR-10 and Tiny-ImageNet, respectively, and is comparable to contrastive learning and masked autoencoders for the first time. Transfer learning from ImageNet also confirms the suitability of DDAE for Vision Transformers, suggesting the potential to scale DDAEs as unified foundation models. Code is available at github.com/FutureXiang/ddae.

arXiv:2303.09663공식 코드 · Papers with Code

Efficient Computation Sharing for Multi-Task Visual Scene Understanding

Solving multiple visual tasks using individual models can be resource-intensive, while multi-task learning can conserve resources by sharing knowledge across different tasks. Despite the benefits of multi-task learning, such techniques can struggle with balancing the loss for each task, leading to potential performance degradation. We present a novel computation- and parameter-sharing framework that balances efficiency and accuracy to perform multiple visual tasks utilizing individually-trained single-task transformers. Our method is motivated by transfer learning schemes to reduce computational and parameter storage costs while maintaining the desired performance. Our approach involves splitting the tasks into a base task and the other sub-tasks, and sharing a significant portion of activations and parameters/weights between the base and sub-tasks to decrease inter-task redundancies and enhance knowledge sharing. The evaluation conducted on NYUD-v2 and PASCAL-context datasets shows that our method is superior to the state-of-the-art transformer-based multi-task learning techniques with higher accuracy and reduced computational resources. Moreover, our method is extended to video stream inputs, further reducing computational costs by efficiently sharing information across the temporal domain as well as the task domain. Our codes and models will be publicly available.

arXiv:2303.09002

Imitation and Transfer Learning for LQG Control

In this paper we study an imitation and transfer learning setting for Linear Quadratic Gaussian (LQG) control, where (i) the system dynamics, noise statistics and cost function are unknown and expert data is provided (that is, sequences of optimal inputs and outputs) to learn the LQG controller, and (ii) multiple control tasks are performed for the same system but with different LQG costs. We show that the LQG controller can be learned from a set of expert trajectories of length $n(l+2)-1$, with $n$ and $l$ the dimension of the system state and output, respectively. Further, the controller can be decomposed as the product of an estimation matrix, which depends only on the system dynamics, and a control matrix, which depends on the LQG cost. This data-based separation principle allows us to transfer the estimation matrix across different LQG tasks, and to reduce the length of the expert trajectories needed to learn the LQG controller to~$2n+m-1$ with $m$ the dimension of the inputs (for single-input systems with $l=2$, this yields approximately a $50\%$ reduction of the required expert data).

arXiv:2303.08915

LRDB: LSTM Raw data DNA Base-caller based on long-short term models in an active learning environment

The first important step in extracting DNA characters is using the output data of MinION devices in the form of electrical current signals. Various cutting-edge base callers use this data to detect the DNA characters based on the input. In this paper, we discuss several shortcomings of prior base callers in the case of time-critical applications, privacy-aware design, and the problem of catastrophic forgetting. Next, we propose the LRDB model, a lightweight open-source model for private developments with a better read-identity (0.35% increase) for the target bacterial samples in the paper. We have limited the extent of training data and benefited from the transfer learning algorithm to make the active usage of the LRDB viable in critical applications. Henceforth, less training time for adapting to new DNA samples (in our case, Bacterial samples) is needed. Furthermore, LRDB can be modified concerning the user constraints as the results show a negligible accuracy loss in case of using fewer parameters. We have also assessed the noise-tolerance property, which offers about a 1.439% decline in accuracy for a 15dB noise injection, and the performance metrics show that the model executes in a medium speed range compared with current cutting-edge models.

arXiv:2303.08818

Boosting Convolutional Neural Networks' Protein Binding Site Prediction Capacity Using SE(3)-invariant transformers, Transfer Learning and Homology-based Augmentation

Figuring out small molecule binding sites in target proteins, in the resolution of either pocket or residue, is critical in many virtual and real drug-discovery scenarios. Since it is not always easy to find such binding sites based on domain knowledge or traditional methods, different deep learning methods that predict binding sites out of protein structures have been developed in recent years. Here we present a new such deep learning algorithm, that significantly outperformed all state-of-the-art baselines in terms of the both resolutions$\unicode{x2013}$pocket and residue. This good performance was also demonstrated in a case study involving the protein human serum albumin and its binding sites. Our algorithm included new ideas both in the model architecture and in the training method. For the model architecture, it incorporated SE(3)-invariant geometric self-attention layers that operate on top of residue-level CNN outputs. This residue-level processing of the model allowed a transfer learning between the two resolutions, which turned out to significantly improve the binding pocket prediction. Moreover, we developed novel augmentation method based on protein homology, which prevented our model from over-fitting. Overall, we believe that our contribution to the literature is twofold. First, we provided a new computational method for binding site prediction that is relevant to real-world applications, as shown by the good performance on different benchmarks and case study. Second, the novel ideas in our method$\unicode{x2013}$the model architecture, transfer learning and the homology augmentation$\unicode{x2013}$would serve as useful components in future works.

arXiv:2303.09536공식 코드 · Papers with Code

Deep Metric Learning for Unsupervised Remote Sensing Change Detection

Remote Sensing Change Detection (RS-CD) aims to detect relevant changes from Multi-Temporal Remote Sensing Images (MT-RSIs), which aids in various RS applications such as land cover, land use, human development analysis, and disaster response. The performance of existing RS-CD methods is attributed to training on large annotated datasets. Furthermore, most of these models are less transferable in the sense that the trained model often performs very poorly when there is a domain gap between training and test datasets. This paper proposes an unsupervised CD method based on deep metric learning that can deal with both of these issues. Given an MT-RSI, the proposed method generates corresponding change probability map by iteratively optimizing an unsupervised CD loss without training it on a large dataset. Our unsupervised CD method consists of two interconnected deep networks, namely Deep-Change Probability Generator (D-CPG) and Deep-Feature Extractor (D-FE). The D-CPG is designed to predict change and no change probability maps for a given MT-RSI, while D-FE is used to extract deep features of MT-RSI that will be further used in the proposed unsupervised CD loss. We use transfer learning capability to initialize the parameters of D-FE. We iteratively optimize the parameters of D-CPG and D-FE for a given MT-RSI by minimizing the proposed unsupervised ``similarity-dissimilarity loss''. This loss is motivated by the principle of metric learning where we simultaneously maximize the distance between change pair-wise pixels while minimizing the distance between no-change pair-wise pixels in bi-temporal image domain and their deep feature domain. The experiments conducted on three CD datasets show that our unsupervised CD method achieves significant improvements over the state-of-the-art supervised and unsupervised CD methods. Code available at https://github.com/wgcban/Metric-CD

arXiv:2303.09253

A Survey of Deep Visual Cross-Domain Few-Shot Learning

Few-Shot transfer learning has become a major focus of research as it allows recognition of new classes with limited labeled data. While it is assumed that train and test data have the same data distribution, this is often not the case in real-world applications. This leads to decreased model transfer effects when the new class distribution differs significantly from the learned classes. Research into Cross-Domain Few-Shot (CDFS) has emerged to address this issue, forming a more challenging and realistic setting. In this survey, we provide a detailed taxonomy of CDFS from the problem setting and corresponding solutions view. We summarise the existing CDFS network architectures and discuss the solution ideas for each direction the taxonomy indicates. Furthermore, we introduce various CDFS downstream applications and outline classification, detection, and segmentation benchmarks and corresponding standards for evaluation. We also discuss the challenges of CDFS research and explore potential directions for future investigation. Through this review, we aim to provide comprehensive guidance on CDFS research, enabling researchers to gain insight into the state-of-the-art while allowing them to build upon existing solutions to develop their own CDFS models.

arXiv:2303.09100

Patch-Prompt Aligned Bayesian Prompt Tuning for Vision-Language Models

For downstream applications of vision-language pre-trained models, there has been significant interest in constructing effective prompts. Existing works on prompt engineering, which either require laborious manual designs or optimize the prompt tuning as a point estimation problem, may fail to describe diverse characteristics of categories and limit their applications. We introduce a Bayesian probabilistic resolution to prompt tuning, where the label-specific stochastic prompts are generated hierarchically by first sampling a latent vector from an underlying distribution and then employing a lightweight generative model. Importantly, we semantically regularize the tuning process by minimizing the statistical distance between the visual patches and linguistic prompts, which pushes the stochastic label representations to faithfully capture diverse visual concepts, instead of overfitting the training categories. We evaluate the effectiveness of our approach on four tasks: few-shot image recognition, base-to-new generalization, dataset transfer learning, and domain shifts. Extensive results over 15 datasets show promising transferability and generalization performance of our proposed model, both quantitatively and qualitatively.

arXiv:2303.09115공식 코드 · Papers with Code

Learning for Amalgamation: A Multi-Source Transfer Learning Framework For Sentiment Classification

Transfer learning plays an essential role in Deep Learning, which can remarkably improve the performance of the target domain, whose training data is not sufficient. Our work explores beyond the common practice of transfer learning with a single pre-trained model. We focus on the task of Vietnamese sentiment classification and propose LIFA, a framework to learn a unified embedding from several pre-trained models. We further propose two more LIFA variants that encourage the pre-trained models to either cooperate or compete with one another. Studying these variants sheds light on the success of LIFA by showing that sharing knowledge among the models is more beneficial for transfer learning. Moreover, we construct the AISIA-VN-Review-F dataset, the first large-scale Vietnamese sentiment classification database. We conduct extensive experiments on the AISIA-VN-Review-F and existing benchmarks to demonstrate the efficacy of LIFA compared to other techniques. To contribute to the Vietnamese NLP research, we publish our source code and datasets to the research community upon acceptance.

arXiv:2303.09062

Knowledge Transfer for Pseudo-code Generation from Low Resource Programming Language

Generation of pseudo-code descriptions of legacy source code for software maintenance is a manually intensive task. Recent encoder-decoder language models have shown promise for automating pseudo-code generation for high resource programming languages such as C++, but are heavily reliant on the availability of a large code-pseudocode corpus. Soliciting such pseudocode annotations for codes written in legacy programming languages (PL) is a time consuming and costly affair requiring a thorough understanding of the source PL. In this paper, we focus on transferring the knowledge acquired by the code-to-pseudocode neural model trained on a high resource PL (C++) using parallel code-pseudocode data. We aim to transfer this knowledge to a legacy PL (C) with no PL-pseudocode parallel data for training. To achieve this, we utilize an Iterative Back Translation (IBT) approach with a novel test-cases based filtration strategy, to adapt the trained C++-to-pseudocode model to C-to-pseudocode model. We observe an improvement of 23.27% in the success rate of the generated C codes through back translation, over the successive IBT iteration, illustrating the efficacy of our approach.

arXiv:2303.08362

Transfer Learning Based Diagnosis and Analysis of Lung Sound Aberrations

With the development of computer -systems that can collect and analyze enormous volumes of data, the medical profession is establishing several non-invasive tools. This work attempts to develop a non-invasive technique for identifying respiratory sounds acquired by a stethoscope and voice recording software via machine learning techniques. This study suggests a trained and proven CNN-based approach for categorizing respiratory sounds. A visual representation of each audio sample is constructed, allowing resource identification for classification using methods like those used to effectively describe visuals. We used a technique called Mel Frequency Cepstral Coefficients (MFCCs). Here, features are retrieved and categorized via VGG16 (transfer learning) and prediction is accomplished using 5-fold cross-validation. Employing various data splitting techniques, Respiratory Sound Database obtained cutting-edge results, including accuracy of 95%, precision of 88%, recall score of 86%, and F1 score of 81%. The ICBHI dataset is used to train and test the model.

arXiv:2303.08262

Clinical Concept and Relation Extraction Using Prompt-based Machine Reading Comprehension

Objective: To develop a natural language processing system that solves both clinical concept extraction and relation extraction in a unified prompt-based machine reading comprehension (MRC) architecture with good generalizability for cross-institution applications. Methods: We formulate both clinical concept extraction and relation extraction using a unified prompt-based MRC architecture and explore state-of-the-art transformer models. We compare our MRC models with existing deep learning models for concept extraction and end-to-end relation extraction using two benchmark datasets developed by the 2018 National NLP Clinical Challenges (n2c2) challenge (medications and adverse drug events) and the 2022 n2c2 challenge (relations of social determinants of health [SDoH]). We also evaluate the transfer learning ability of the proposed MRC models in a cross-institution setting. We perform error analyses and examine how different prompting strategies affect the performance of MRC models. Results and Conclusion: The proposed MRC models achieve state-of-the-art performance for clinical concept and relation extraction on the two benchmark datasets, outperforming previous non-MRC transformer models. GatorTron-MRC achieves the best strict and lenient F1-scores for concept extraction, outperforming previous deep learning models on the two datasets by 1%~3% and 0.7%~1.3%, respectively. For end-to-end relation extraction, GatorTron-MRC and BERT-MIMIC-MRC achieve the best F1-scores, outperforming previous deep learning models by 0.9%~2.4% and 10%-11%, respectively. For cross-institution evaluation, GatorTron-MRC outperforms traditional GatorTron by 6.4% and 16% for the two datasets, respectively. The proposed method is better at handling nested/overlapped concepts, extracting relations, and has good portability for cross-institute applications.

arXiv:2303.08142

Performance Embeddings: A Similarity-based Approach to Automatic Performance Optimization

Performance optimization is an increasingly challenging but often repetitive task. While each platform has its quirks, the underlying code transformations rely on data movement and computational characteristics that recur across applications. This paper proposes to leverage those similarities by constructing an embedding space for subprograms. The continuous space captures both static and dynamic properties of loop nests via symbolic code analysis and performance profiling, respectively. Performance embeddings enable direct knowledge transfer of performance tuning between applications, which can result from autotuning or tailored improvements. We demonstrate this transfer tuning approach on case studies in deep neural networks, dense and sparse linear algebra compositions, and numerical weather prediction stencils. Transfer tuning reduces the search complexity by up to four orders of magnitude and outperforms the MKL library in sparse-dense matrix multiplication. The results exhibit clear correspondences between program characteristics and optimizations, outperforming prior specialized state-of-the-art approaches and generalizing beyond their capabilities.

arXiv:2303.08577

Investigating GANsformer: A Replication Study of a State-of-the-Art Image Generation Model

The field of image generation through generative modelling is abundantly discussed nowadays. It can be used for various applications, such as up-scaling existing images, creating non-existing objects, such as interior design scenes, products or even human faces, and achieving transfer-learning processes. In this context, Generative Adversarial Networks (GANs) are a class of widely studied machine learning frameworks first appearing in the paper "Generative adversarial nets" by Goodfellow et al. that achieve the goal above. In our work, we reproduce and evaluate a novel variation of the original GAN network, the GANformer, proposed in "Generative Adversarial Transformers" by Hudson and Zitnick. This project aimed to recreate the methods presented in this paper to reproduce the original results and comment on the authors' claims. Due to resources and time limitations, we had to constrain the network's training times, dataset types, and sizes. Our research successfully recreated both variations of the proposed GANformer model and found differences between the authors' and our results. Moreover, discrepancies between the publication methodology and the one implemented, made available in the code, allowed us to study two undisclosed variations of the presented procedures.