PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2305.08076

Improving Defensive Distillation using Teacher Assistant

Adversarial attacks pose a significant threat to the security and safety of deep neural networks being applied to modern applications. More specifically, in computer vision-based tasks, experts can use the knowledge of model architecture to create adversarial samples imperceptible to the human eye. These attacks can lead to security problems in popular applications such as self-driving cars, face recognition, etc. Hence, building networks which are robust to such attacks is highly desirable and essential. Among the various methods present in literature, defensive distillation has shown promise in recent years. Using knowledge distillation, researchers have been able to create models robust against some of those attacks. However, more attacks have been developed exposing weakness in defensive distillation. In this project, we derive inspiration from teacher assistant knowledge distillation and propose that introducing an assistant network can improve the robustness of the distilled model. Through a series of experiments, we evaluate the distilled models for different distillation temperatures in terms of accuracy, sensitivity, and robustness. Our experiments demonstrate that the proposed hypothesis can improve robustness in most cases. Additionally, we show that multi-step distillation can further improve robustness with very little impact on model accuracy.

arXiv:2305.08075

Analyzing Compression Techniques for Computer Vision

Compressing deep networks is highly desirable for practical use-cases in computer vision applications. Several techniques have been explored in the literature, and research has been done in finding efficient strategies for combining them. For this project, we aimed to explore three different basic compression techniques - knowledge distillation, pruning, and quantization for small-scale recognition tasks. Along with the basic methods, we also test the efficacy of combining them in a sequential manner. We analyze them using MNIST and CIFAR-10 datasets and present the results along with few observations inferred from them.

arXiv:2305.07931공식 코드 · Papers with Code

GSB: Group Superposition Binarization for Vision Transformer with Limited Training Samples

Vision Transformer (ViT) has performed remarkably in various computer vision tasks. Nonetheless, affected by the massive amount of parameters, ViT usually suffers from serious overfitting problems with a relatively limited number of training samples. In addition, ViT generally demands heavy computing resources, which limit its deployment on resource-constrained devices. As a type of model-compression method, model binarization is potentially a good choice to solve the above problems. Compared with the full-precision one, the model with the binarization method replaces complex tensor multiplication with simple bit-wise binary operations and represents full-precision model parameters and activations with only 1-bit ones, which potentially solves the problem of model size and computational complexity, respectively. In this paper, we investigate a binarized ViT model. Empirically, we observe that the existing binarization technology designed for Convolutional Neural Networks (CNN) cannot migrate well to a ViT's binarization task. We also find that the decline of the accuracy of the binary ViT model is mainly due to the information loss of the Attention module and the Value vector. Therefore, we propose a novel model binarization technique, called Group Superposition Binarization (GSB), to deal with these issues. Furthermore, in order to further improve the performance of the binarization model, we have investigated the gradient calculation procedure in the binarization process and derived more proper gradient calculation equations for GSB to reduce the influence of gradient mismatch. Then, the knowledge distillation technique is introduced to alleviate the performance degradation caused by model binarization. Analytically, model binarization can limit the parameters search space during parameter updates while training a model....

arXiv:2305.07881

Black-box Source-free Domain Adaptation via Two-stage Knowledge Distillation

Source-free domain adaptation aims to adapt deep neural networks using only pre-trained source models and target data. However, accessing the source model still has a potential concern about leaking the source data, which reveals the patient's privacy. In this paper, we study the challenging but practical problem: black-box source-free domain adaptation where only the outputs of the source model and target data are available. We propose a simple but effective two-stage knowledge distillation method. In Stage \uppercase\expandafter{\romannumeral1}, we train the target model from scratch with soft pseudo-labels generated by the source model in a knowledge distillation manner. In Stage \uppercase\expandafter{\romannumeral2}, we initialize another model as the new student model to avoid the error accumulation caused by noisy pseudo-labels. We feed the images with weak augmentation to the teacher model to guide the learning of the student model. Our method is simple and flexible, and achieves surprising results on three cross-domain segmentation tasks.

arXiv:2305.08096공식 코드 · Papers with Code

Towards Understanding and Improving Knowledge Distillation for Neural Machine Translation

Knowledge distillation (KD) is a promising technique for model compression in neural machine translation. However, where the knowledge hides in KD is still not clear, which may hinder the development of KD. In this work, we first unravel this mystery from an empirical perspective and show that the knowledge comes from the top-1 predictions of teachers, which also helps us build a potential connection between word- and sequence-level KD. Further, we point out two inherent issues in vanilla word-level KD based on this finding. Firstly, the current objective of KD spreads its focus to whole distributions to learn the knowledge, yet lacks special treatment on the most crucial top-1 information. Secondly, the knowledge is largely covered by the golden information due to the fact that most top-1 predictions of teachers overlap with ground-truth tokens, which further restricts the potential of KD. To address these issues, we propose a novel method named \textbf{T}op-1 \textbf{I}nformation \textbf{E}nhanced \textbf{K}nowledge \textbf{D}istillation (TIE-KD). Specifically, we design a hierarchical ranking loss to enforce the learning of the top-1 information from the teacher. Additionally, we develop an iterative KD procedure to infuse more additional knowledge by distilling on the data without ground-truth targets. Experiments on WMT'14 English-German, WMT'14 English-French and WMT'16 English-Romanian demonstrate that our method can respectively boost Transformer$_{base}$ students by +1.04, +0.60 and +1.11 BLEU scores and significantly outperform the vanilla word-level KD baseline. Besides, our method shows higher generalizability on different teacher-student capacity gaps than existing KD techniques.

arXiv:2305.08031공식 코드 · Papers with Code

On enhancing the robustness of Vision Transformers: Defensive Diffusion

Privacy and confidentiality of medical data are of utmost importance in healthcare settings. ViTs, the SOTA vision model, rely on large amounts of patient data for training, which raises concerns about data security and the potential for unauthorized access. Adversaries may exploit vulnerabilities in ViTs to extract sensitive patient information and compromising patient privacy. This work address these vulnerabilities to ensure the trustworthiness and reliability of ViTs in medical applications. In this work, we introduced a defensive diffusion technique as an adversarial purifier to eliminate adversarial noise introduced by attackers in the original image. By utilizing the denoising capabilities of the diffusion model, we employ a reverse diffusion process to effectively eliminate the adversarial noise from the attack sample, resulting in a cleaner image that is then fed into the ViT blocks. Our findings demonstrate the effectiveness of the diffusion model in eliminating attack-agnostic adversarial noise from images. Additionally, we propose combining knowledge distillation with our framework to obtain a lightweight student model that is both computationally efficient and robust against gray box attacks. Comparison of our method with a SOTA baseline method, SEViT, shows that our work is able to outperform the baseline. Extensive experiments conducted on a publicly available Tuberculosis X-ray dataset validate the computational efficiency and improved robustness achieved by our proposed architecture.

arXiv:2305.07928

AMTSS: An Adaptive Multi-Teacher Single-Student Knowledge Distillation Framework For Multilingual Language Inference

Knowledge distillation is of key importance to launching multilingual pre-trained language models for real applications. To support cost-effective language inference in multilingual settings, we propose AMTSS, an adaptive multi-teacher single-student distillation framework, which allows distilling knowledge from multiple teachers to a single student. We first introduce an adaptive learning strategy and teacher importance weight, which enables a student to effectively learn from max-margin teachers and easily adapt to new languages. Moreover, we present a shared student encoder with different projection layers in support of multiple languages, which contributes to largely reducing development and machine cost. Experimental results show that AMTSS gains competitive results on the public XNLI dataset and the realistic industrial dataset AliExpress (AE) in the E-commerce scenario.

arXiv:2305.08553

Distilling Knowledge for Short-to-Long Term Trajectory Prediction

Long-term trajectory forecasting is an important and challenging problem in the fields of computer vision, machine learning, and robotics. One fundamental difficulty stands in the evolution of the trajectory that becomes more and more uncertain and unpredictable as the time horizon grows, subsequently increasing the complexity of the problem. To overcome this issue, in this paper, we propose Di-Long, a new method that employs the distillation of a short-term trajectory model forecaster that guides a student network for long-term trajectory prediction during the training process. Given a total sequence length that comprehends the allowed observation for the student network and the complementary target sequence, we let the student and the teacher solve two different related tasks defined over the same full trajectory: the student observes a short sequence and predicts a long trajectory, whereas the teacher observes a longer sequence and predicts the remaining short target trajectory. The teacher's task is less uncertain, and we use its accurate predictions to guide the student through our knowledge distillation framework, reducing long-term future uncertainty. Our experiments show that our proposed Di-Long method is effective for long-term forecasting and achieves state-of-the-art performance on the Intersection Drone Dataset (inD) and the Stanford Drone Dataset (SDD).

arXiv:2305.07586

Knowledge distillation with Segment Anything (SAM) model for Planetary Geological Mapping

Planetary science research involves analysing vast amounts of remote sensing data, which are often costly and time-consuming to annotate and process. One of the essential tasks in this field is geological mapping, which requires identifying and outlining regions of interest in planetary images, including geological features and landforms. However, manually labelling these images is a complex and challenging task that requires significant domain expertise and effort. To expedite this endeavour, we propose the use of knowledge distillation using the recently introduced cutting-edge Segment Anything (SAM) model. We demonstrate the effectiveness of this prompt-based foundation model for rapid annotation and quick adaptability to a prime use case of mapping planetary skylights. Our work reveals that with a small set of annotations obtained with the right prompts from the model and subsequently training a specialised domain decoder, we can achieve satisfactory semantic segmentation on this task. Key results indicate that the use of knowledge distillation can significantly reduce the effort required by domain experts for manual annotation and improve the efficiency of image segmentation tasks. This approach has the potential to accelerate extra-terrestrial discovery by automatically detecting and segmenting Martian landforms.

arXiv:2305.07446

A Lightweight Domain Adversarial Neural Network Based on Knowledge Distillation for EEG-based Cross-subject Emotion Recognition

Individual differences of Electroencephalogram (EEG) could cause the domain shift which would significantly degrade the performance of cross-subject strategy. The domain adversarial neural networks (DANN), where the classification loss and domain loss jointly update the parameters of feature extractor, are adopted to deal with the domain shift. However, limited EEG data quantity and strong individual difference are challenges for the DANN with cumbersome feature extractor. In this work, we propose knowledge distillation (KD) based lightweight DANN to enhance cross-subject EEG-based emotion recognition. Specifically, the teacher model with strong context learning ability is utilized to learn complex temporal dynamics and spatial correlations of EEG, and robust lightweight student model is guided by the teacher model to learn more difficult domain-invariant features. In the feature-based KD framework, a transformer-based hierarchical temporalspatial learning model is served as the teacher model. The student model, which is composed of Bi-LSTM units, is a lightweight version of the teacher model. Hence, the student model could be supervised to mimic the robust feature representations of teacher model by leveraging complementary latent temporal features and spatial features. In the DANN-based cross-subject emotion recognition, we combine the obtained student model and a lightweight temporal-spatial feature interaction module as the feature extractor. And the feature aggregation is fed to the emotion classifier and domain classifier for domain-invariant feature learning. To verify the effectiveness of the proposed method, we conduct the subject-independent experiments on the public dataset DEAP with arousal and valence classification. The outstanding performance and t-SNE visualization of latent features verify the advantage and effectiveness of the proposed method.

arXiv:2305.06620공식 코드 · Papers with Code

Improving Continual Relation Extraction by Distinguishing Analogous Semantics

Continual relation extraction (RE) aims to learn constantly emerging relations while avoiding forgetting the learned relations. Existing works store a small number of typical samples to re-train the model for alleviating forgetting. However, repeatedly replaying these samples may cause the overfitting problem. We conduct an empirical study on existing works and observe that their performance is severely affected by analogous relations. To address this issue, we propose a novel continual extraction model for analogous relations. Specifically, we design memory-insensitive relation prototypes and memory augmentation to overcome the overfitting problem. We also introduce integrated training and focal knowledge distillation to enhance the performance on analogous relations. Experimental results show the superiority of our model and demonstrate its effectiveness in distinguishing analogous relations and overcoming overfitting.

arXiv:2305.06616공식 코드 · Papers with Code

Serial Contrastive Knowledge Distillation for Continual Few-shot Relation Extraction

Continual few-shot relation extraction (RE) aims to continuously train a model for new relations with few labeled training data, of which the major challenges are the catastrophic forgetting of old relations and the overfitting caused by data sparsity. In this paper, we propose a new model, namely SCKD, to accomplish the continual few-shot RE task. Specifically, we design serial knowledge distillation to preserve the prior knowledge from previous models and conduct contrastive learning with pseudo samples to keep the representations of samples in different relations sufficiently distinguishable. Our experiments on two benchmark datasets validate the effectiveness of SCKD for continual few-shot RE and its superiority in knowledge transfer and memory utilization over state-of-the-art models.

arXiv:2305.06557공식 코드 · Papers with Code

Long-Tailed Question Answering in an Open World

Real-world data often have an open long-tailed distribution, and building a unified QA model supporting various tasks is vital for practical QA applications. However, it is non-trivial to extend previous QA approaches since they either require access to seen tasks of adequate samples or do not explicitly model samples from unseen tasks. In this paper, we define Open Long-Tailed QA (OLTQA) as learning from long-tailed distributed data and optimizing performance over seen and unseen QA tasks. We propose an OLTQA model that encourages knowledge sharing between head, tail and unseen tasks, and explicitly mines knowledge from a large pre-trained language model (LM). Specifically, we organize our model through a pool of fine-grained components and dynamically combine these components for an input to facilitate knowledge sharing. A retrieve-then-rerank frame is further introduced to select in-context examples, which guild the LM to generate text that express knowledge for QA tasks. Moreover, a two-stage training approach is introduced to pre-train the framework by knowledge distillation (KD) from the LM and then jointly train the frame and a QA model through an adaptive mutual KD method. On a large-scale OLTQA dataset we curate from 43 existing QA datasets, our model consistently outperforms the state-of-the-art. We release the code and data at \url{https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/oltqa}.

arXiv:2305.06244

Explainable Knowledge Distillation for On-device Chest X-Ray Classification

Automated multi-label chest X-rays (CXR) image classification has achieved substantial progress in clinical diagnosis via utilizing sophisticated deep learning approaches. However, most deep models have high computational demands, which makes them less feasible for compact devices with low computational requirements. To overcome this problem, we propose a knowledge distillation (KD) strategy to create the compact deep learning model for the real-time multi-label CXR image classification. We study different alternatives of CNNs and Transforms as the teacher to distill the knowledge to a smaller student. Then, we employed explainable artificial intelligence (XAI) to provide the visual explanation for the model decision improved by the KD. Our results on three benchmark CXR datasets show that our KD strategy provides the improved performance on the compact student model, thus being the feasible choice for many limited hardware platforms. For instance, when using DenseNet161 as the teacher network, EEEA-Net-C2 achieved an AUC of 83.7%, 87.1%, and 88.7% on the ChestX-ray14, CheXpert, and PadChest datasets, respectively, with fewer parameters of 4.7 million and computational cost of 0.3 billion FLOPS.

arXiv:2305.06024공식 코드 · Papers with Code

A Survey on the Robustness of Computer Vision Models against Common Corruptions

The performance of computer vision models are susceptible to unexpected changes in input images caused by sensor errors or extreme imaging environments, known as common corruptions (e.g. noise, blur, illumination changes). These corruptions can significantly hinder the reliability of these models when deployed in real-world scenarios, yet they are often overlooked when testing model generalization and robustness. In this survey, we present a comprehensive overview of methods that improve the robustness of computer vision models against common corruptions. We categorize methods into three groups based on the model components and training methods they target: data augmentation, learning strategies, and network components. We release a unified benchmark framework (available at \url{https://github.com/nis-research/CorruptionBenchCV}) to compare robustness performance across several datasets, and we address the inconsistencies of evaluation practices in the literature. Our experimental analysis highlights the base corruption robustness of popular vision backbones, revealing that corruption robustness does not necessarily scale with model size and data size. Large models gain negligible robustness improvements, considering the increased computational requirements. To achieve generalizable and robust computer vision models, we foresee the need of developing new learning strategies that efficiently exploit limited data and mitigate unreliable learning behaviors.

arXiv:2305.04978공식 코드 · Papers with Code

NeuroComparatives: Neuro-Symbolic Distillation of Comparative Knowledge

Comparative knowledge (e.g., steel is stronger and heavier than styrofoam) is an essential component of our world knowledge, yet understudied in prior literature. In this paper, we harvest the dramatic improvements in knowledge capabilities of language models into a large-scale comparative knowledge base. While the ease of acquisition of such comparative knowledge is much higher from extreme-scale models like GPT-4, compared to their considerably smaller and weaker counterparts such as GPT-2, not even the most powerful models are exempt from making errors. We thus ask: to what extent are models at different scales able to generate valid and diverse comparative knowledge? We introduce NeuroComparatives, a novel framework for comparative knowledge distillation overgenerated from language models such as GPT-variants and LLaMA, followed by stringent filtering of the generated knowledge. Our framework acquires comparative knowledge between everyday objects, producing a corpus of up to 8.8M comparisons over 1.74M entity pairs - 10X larger and 30% more diverse than existing resources. Moreover, human evaluations show that NeuroComparatives outperform existing resources in terms of validity (up to 32% absolute improvement). Our acquired NeuroComparatives leads to performance improvements on five downstream tasks. We find that neuro-symbolic manipulation of smaller models offers complementary benefits to the currently dominant practice of prompting extreme-scale language models for knowledge distillation.

arXiv:2305.05233

DynamicKD: An Effective Knowledge Distillation via Dynamic Entropy Correction-Based Distillation for Gap Optimizing

The knowledge distillation uses a high-performance teacher network to guide the student network. However, the performance gap between the teacher and student networks can affect the student's training. This paper proposes a novel knowledge distillation algorithm based on dynamic entropy correction to reduce the gap by adjusting the student instead of the teacher. Firstly, the effect of changing the output entropy (short for output information entropy) in the student on the distillation loss is analyzed in theory. This paper shows that correcting the output entropy can reduce the gap. Then, a knowledge distillation algorithm based on dynamic entropy correction is created, which can correct the output entropy in real-time with an entropy controller updated dynamically by the distillation loss. The proposed algorithm is validated on the CIFAR100 and ImageNet. The comparison with various state-of-the-art distillation algorithms shows impressive results, especially in the experiment on the CIFAR100 regarding teacher-student pair resnet32x4-resnet8x4. The proposed algorithm raises 2.64 points over the traditional distillation algorithm and 0.87 points over the state-of-the-art algorithm CRD in classification accuracy, demonstrating its effectiveness and efficiency.

arXiv:2305.05189공식 코드 · Papers with Code

SUR-adapter: Enhancing Text-to-Image Pre-trained Diffusion Models with Large Language Models

Diffusion models, which have emerged to become popular text-to-image generation models, can produce high-quality and content-rich images guided by textual prompts. However, there are limitations to semantic understanding and commonsense reasoning in existing models when the input prompts are concise narrative, resulting in low-quality image generation. To improve the capacities for narrative prompts, we propose a simple-yet-effective parameter-efficient fine-tuning approach called the Semantic Understanding and Reasoning adapter (SUR-adapter) for pre-trained diffusion models. To reach this goal, we first collect and annotate a new dataset SURD which consists of more than 57,000 semantically corrected multi-modal samples. Each sample contains a simple narrative prompt, a complex keyword-based prompt, and a high-quality image. Then, we align the semantic representation of narrative prompts to the complex prompts and transfer knowledge of large language models (LLMs) to our SUR-adapter via knowledge distillation so that it can acquire the powerful semantic understanding and reasoning capabilities to build a high-quality textual semantic representation for text-to-image generation. We conduct experiments by integrating multiple LLMs and popular pre-trained diffusion models to show the effectiveness of our approach in enabling diffusion models to understand and reason concise natural language without image quality degradation. Our approach can make text-to-image diffusion models easier to use with better user experience, which demonstrates our approach has the potential for further advancing the development of user-friendly text-to-image generation models by bridging the semantic gap between simple narrative prompts and complex keyword-based prompts. The code is released at https://github.com/Qrange-group/SUR-adapter.

arXiv:2305.05175

SRIL: Selective Regularization for Class-Incremental Learning

Human intelligence gradually accepts new information and accumulates knowledge throughout the lifespan. However, deep learning models suffer from a catastrophic forgetting phenomenon, where they forget previous knowledge when acquiring new information. Class-Incremental Learning aims to create an integrated model that balances plasticity and stability to overcome this challenge. In this paper, we propose a selective regularization method that accepts new knowledge while maintaining previous knowledge. We first introduce an asymmetric feature distillation method for old and new classes inspired by cognitive science, using the gradient of classification and knowledge distillation losses to determine whether to perform pattern completion or pattern separation. We also propose a method to selectively interpolate the weight of the previous model for a balance between stability and plasticity, and we adjust whether to transfer through model confidence to ensure the performance of the previous class and enable exploratory learning. We validate the effectiveness of the proposed method, which surpasses the performance of existing methods through extensive experimental protocols using CIFAR-100, ImageNet-Subset, and ImageNet-Full.

arXiv:2305.05027

Web Content Filtering through knowledge distillation of Large Language Models

We introduce a state-of-the-art approach for URL categorization that leverages the power of Large Language Models (LLMs) to address the primary objectives of web content filtering: safeguarding organizations from legal and ethical risks, limiting access to high-risk or suspicious websites, and fostering a secure and professional work environment. Our method utilizes LLMs to generate accurate classifications and then employs established knowledge distillation techniques to create smaller, more specialized student models tailored for web content filtering. Distillation results in a student model with a 9% accuracy rate improvement in classifying websites, sourced from customer telemetry data collected by a large security vendor, into 30 distinct content categories based on their URLs, surpassing the current state-of-the-art approach. Our student model matches the performance of the teacher LLM with 175 times less parameters, allowing the model to be used for in-line scanning of large volumes of URLs, and requires 3 orders of magnitude less manually labeled training data than the current state-of-the-art approach. Depending on the specific use case, the output generated by our approach can either be directly returned or employed as a pre-filter for more resource-intensive operations involving website images or HTML.