PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2211.09862

Knowledge distillation for fast and accurate DNA sequence correction

Accurate genome sequencing can improve our understanding of biology and the genetic basis of disease. The standard approach for generating DNA sequences from PacBio instruments relies on HMM-based models. Here, we introduce Distilled DeepConsensus - a distilled transformer-encoder model for sequence correction, which improves upon the HMM-based methods with runtime constraints in mind. Distilled DeepConsensus is 1.3x faster and 1.5x smaller than its larger counterpart while improving the yield of high quality reads (Q30) over the HMM-based method by 1.69x (vs. 1.73x for larger model). With improved accuracy of genomic sequences, Distilled DeepConsensus improves downstream applications of genomic sequence analysis such as reducing variant calling errors by 39% (34% for larger model) and improving genome assembly quality by 3.8% (4.2% for larger model). We show that the representations learned by Distilled DeepConsensus are similar between faster and slower models.

arXiv:2211.09394공식 코드 · Papers with Code

ConNER: Consistency Training for Cross-lingual Named Entity Recognition

Cross-lingual named entity recognition (NER) suffers from data scarcity in the target languages, especially under zero-shot settings. Existing translate-train or knowledge distillation methods attempt to bridge the language gap, but often introduce a high level of noise. To solve this problem, consistency training methods regularize the model to be robust towards perturbations on data or hidden states. However, such methods are likely to violate the consistency hypothesis, or mainly focus on coarse-grain consistency. We propose ConNER as a novel consistency training framework for cross-lingual NER, which comprises of: (1) translation-based consistency training on unlabeled target-language data, and (2) dropoutbased consistency training on labeled source-language data. ConNER effectively leverages unlabeled target-language data and alleviates overfitting on the source language to enhance the cross-lingual adaptability. Experimental results show our ConNER achieves consistent improvement over various baseline methods.

arXiv:2211.09768

D$^3$ETR: Decoder Distillation for Detection Transformer

While various knowledge distillation (KD) methods in CNN-based detectors show their effectiveness in improving small students, the baselines and recipes for DETR-based detectors are yet to be built. In this paper, we focus on the transformer decoder of DETR-based detectors and explore KD methods for them. The outputs of the transformer decoder lie in random order, which gives no direct correspondence between the predictions of the teacher and the student, thus posing a challenge for knowledge distillation. To this end, we propose MixMatcher to align the decoder outputs of DETR-based teachers and students, which mixes two teacher-student matching strategies, i.e., Adaptive Matching and Fixed Matching. Specifically, Adaptive Matching applies bipartite matching to adaptively match the outputs of the teacher and the student in each decoder layer, while Fixed Matching fixes the correspondence between the outputs of the teacher and the student with the same object queries, with the teacher's fixed object queries fed to the decoder of the student as an auxiliary group. Based on MixMatcher, we build \textbf{D}ecoder \textbf{D}istillation for \textbf{DE}tection \textbf{TR}ansformer (D$^3$ETR), which distills knowledge in decoder predictions and attention maps from the teachers to students. D$^3$ETR shows superior performance on various DETR-based detectors with different backbones. For example, D$^3$ETR improves Conditional DETR-R50-C5 by $\textbf{7.8}/\textbf{2.4}$ mAP under $12/50$ epochs training settings with Conditional DETR-R101-C5 as the teacher.

arXiv:2211.09740

Sub-Graph Learning for Spatiotemporal Forecasting via Knowledge Distillation

One of the challenges in studying the interactions in large graphs is to learn their diverse pattern and various interaction types. Hence, considering only one distribution and model to study all nodes and ignoring their diversity and local features in their neighborhoods, might severely affect the overall performance. Based on the structural information of the nodes in the graph and the interactions between them, the main graph can be divided into multiple sub-graphs. This graph partitioning can tremendously affect the learning process, however the overall performance is highly dependent on the clustering method to avoid misleading the model. In this work, we present a new framework called KD-SGL to effectively learn the sub-graphs, where we define one global model to learn the overall structure of the graph and multiple local models for each sub-graph. We assess the performance of the proposed framework and evaluate it on public datasets. Based on the achieved results, it can improve the performance of the state-of-the-arts spatiotemporal models with comparable results compared to ensemble of models with less complexity.

arXiv:2211.09386공식 코드 · Papers with Code

BEVDistill: Cross-Modal BEV Distillation for Multi-View 3D Object Detection

3D object detection from multiple image views is a fundamental and challenging task for visual scene understanding. Owing to its low cost and high efficiency, multi-view 3D object detection has demonstrated promising application prospects. However, accurately detecting objects through perspective views is extremely difficult due to the lack of depth information. Current approaches tend to adopt heavy backbones for image encoders, making them inapplicable for real-world deployment. Different from the images, LiDAR points are superior in providing spatial cues, resulting in highly precise localization. In this paper, we explore the incorporation of LiDAR-based detectors for multi-view 3D object detection. Instead of directly training a depth prediction network, we unify the image and LiDAR features in the Bird-Eye-View (BEV) space and adaptively transfer knowledge across non-homogenous representations in a teacher-student paradigm. To this end, we propose \textbf{BEVDistill}, a cross-modal BEV knowledge distillation (KD) framework for multi-view 3D object detection. Extensive experiments demonstrate that the proposed method outperforms current KD approaches on a highly-competitive baseline, BEVFormer, without introducing any extra cost in the inference phase. Notably, our best model achieves 59.4 NDS on the nuScenes test leaderboard, achieving new state-of-the-art in comparison with various image-based detectors. Code will be available at https://github.com/zehuichen123/BEVDistill.

arXiv:2211.08743

Yield Evaluation of Citrus Fruits based on the YoloV5 compressed by Knowledge Distillation

In the field of planting fruit trees, pre-harvest estimation of fruit yield is important for fruit storage and price evaluation. However, considering the cost, the yield of each tree cannot be assessed by directly picking the immature fruit. Therefore, the problem is a very difficult task. In this paper, a fruit counting and yield assessment method based on computer vision is proposed for citrus fruit trees as an example. Firstly, images of single fruit trees from different angles are acquired and the number of fruits is detected using a deep Convolutional Neural Network model YOLOv5, and the model is compressed using a knowledge distillation method. Then, a linear regression method is used to model yield-related features and evaluate yield. Experiments show that the proposed method can accurately count fruits and approximate the yield.

arXiv:2211.07715공식 코드 · Papers with Code

Fast DistilBERT on CPUs

Transformer-based language models have become the standard approach to solving natural language processing tasks. However, industry adoption usually requires the maximum throughput to comply with certain latency constraints that prevents Transformer models from being used in production. To address this gap, model compression techniques such as quantization and pruning may be used to improve inference efficiency. However, these compression techniques require specialized software to apply and deploy at scale. In this work, we propose a new pipeline for creating and running Fast Transformer models on CPUs, utilizing hardware-aware pruning, knowledge distillation, quantization, and our own Transformer inference runtime engine with optimized kernels for sparse and quantized operators. We demonstrate the efficiency of our pipeline by creating a Fast DistilBERT model showing minimal accuracy loss on the question-answering SQuADv1.1 benchmark, and throughput results under typical production constraints and environments. Our results outperform existing state-of-the-art Neural Magic's DeepSparse runtime performance by up to 50% and up to 4.1x performance speedup over ONNX Runtime. Source code is publicly available at https://github.com/intel/intel-extension-for-transformers.

arXiv:2211.08161공식 코드 · Papers with Code

An Investigation of the Combination of Rehearsal and Knowledge Distillation in Continual Learning for Spoken Language Understanding

Continual learning refers to a dynamical framework in which a model receives a stream of non-stationary data over time and must adapt to new data while preserving previously acquired knowledge. Unluckily, neural networks fail to meet these two desiderata, incurring the so-called catastrophic forgetting phenomenon. Whereas a vast array of strategies have been proposed to attenuate forgetting in the computer vision domain, for speech-related tasks, on the other hand, there is a dearth of works. In this paper, we consider the joint use of rehearsal and knowledge distillation (KD) approaches for spoken language understanding under a class-incremental learning scenario. We report on multiple KD combinations at different levels in the network, showing that combining feature-level and predictions-level KDs leads to the best results. Finally, we provide an ablation study on the effect of the size of the rehearsal memory that corroborates the efficacy of our approach for low-resource devices.

arXiv:2211.08112

An Efficient Active Learning Pipeline for Legal Text Classification

Active Learning (AL) is a powerful tool for learning with less labeled data, in particular, for specialized domains, like legal documents, where unlabeled data is abundant, but the annotation requires domain expertise and is thus expensive. Recent works have shown the effectiveness of AL strategies for pre-trained language models. However, most AL strategies require a set of labeled samples to start with, which is expensive to acquire. In addition, pre-trained language models have been shown unstable during fine-tuning with small datasets, and their embeddings are not semantically meaningful. In this work, we propose a pipeline for effectively using active learning with pre-trained language models in the legal domain. To this end, we leverage the available unlabeled data in three phases. First, we continue pre-training the model to adapt it to the downstream task. Second, we use knowledge distillation to guide the model's embeddings to a semantically meaningful space. Finally, we propose a simple, yet effective, strategy to find the initial set of labeled samples with fewer actions compared to existing methods. Our experiments on Contract-NLI, adapted to the classification task, and LEDGAR benchmarks show that our approach outperforms standard AL strategies, and is more efficient. Furthermore, our pipeline reaches comparable results to the fully-supervised approach with a small performance gap, and dramatically reduced annotation cost. Code and the adapted data will be made available.

arXiv:2211.08398

Structured Knowledge Distillation Towards Efficient and Compact Multi-View 3D Detection

Detecting 3D objects from multi-view images is a fundamental problem in 3D computer vision. Recently, significant breakthrough has been made in multi-view 3D detection tasks. However, the unprecedented detection performance of these vision BEV (bird's-eye-view) detection models is accompanied with enormous parameters and computation, which make them unaffordable on edge devices. To address this problem, in this paper, we propose a structured knowledge distillation framework, aiming to improve the efficiency of modern vision-only BEV detection models. The proposed framework mainly includes: (a) spatial-temporal distillation which distills teacher knowledge of information fusion from different timestamps and views, (b) BEV response distillation which distills teacher response to different pillars, and (c) weight-inheriting which solves the problem of inconsistent inputs between students and teacher in modern transformer architectures. Experimental results show that our method leads to an average improvement of 2.16 mAP and 2.27 NDS on the nuScenes benchmark, outperforming multiple baselines by a large margin.

arXiv:2211.08106

Instance-aware Model Ensemble With Distillation For Unsupervised Domain Adaptation

The linear ensemble based strategy, i.e., averaging ensemble, has been proposed to improve the performance in unsupervised domain adaptation tasks. However, a typical UDA task is usually challenged by dynamically changing factors, such as variable weather, views, and background in the unlabeled target domain. Most previous ensemble strategies ignore UDA's dynamic and uncontrollable challenge, facing limited feature representations and performance bottlenecks. To enhance the model, adaptability between domains and reduce the computational cost when deploying the ensemble model, we propose a novel framework, namely Instance aware Model Ensemble With Distillation, IMED, which fuses multiple UDA component models adaptively according to different instances and distills these components into a small model. The core idea of IMED is a dynamic instance aware ensemble strategy, where for each instance, a nonlinear fusion subnetwork is learned that fuses the extracted features and predicted labels of multiple component models. The nonlinear fusion method can help the ensemble model handle dynamically changing factors. After learning a large capacity ensemble model with good adaptability to different changing factors, we leverage the ensemble teacher model to guide the learning of a compact student model by knowledge distillation. Furthermore, we provide the theoretical analysis of the validity of IMED for UDA. Extensive experiments conducted on various UDA benchmark datasets, e.g., Office 31, Office Home, and VisDA 2017, show the superiority of the model based on IMED to the state of the art methods under the comparable computation cost.

arXiv:2211.08071

Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling

DETR is a novel end-to-end transformer architecture object detector, which significantly outperforms classic detectors when scaling up the model size. In this paper, we focus on the compression of DETR with knowledge distillation. While knowledge distillation has been well-studied in classic detectors, there is a lack of researches on how to make it work effectively on DETR. We first provide experimental and theoretical analysis to point out that the main challenge in DETR distillation is the lack of consistent distillation points. Distillation points refer to the corresponding inputs of the predictions for student to mimic, and reliable distillation requires sufficient distillation points which are consistent between teacher and student. Based on this observation, we propose a general knowledge distillation paradigm for DETR(KD-DETR) with consistent distillation points sampling. Specifically, we decouple detection and distillation tasks by introducing a set of specialized object queries to construct distillation points. In this paradigm, we further propose a general-to-specific distillation points sampling strategy to explore the extensibility of KD-DETR. Extensive experiments on different DETR architectures with various scales of backbones and transformer layers validate the effectiveness and generalization of KD-DETR. KD-DETR boosts the performance of DAB-DETR with ResNet-18 and ResNet-50 backbone to 41.4$\%$, 45.7$\%$ mAP, respectively, which are 5.2$\%$, 3.5$\%$ higher than the baseline, and ResNet-50 even surpasses the teacher model by $2.2\%$.

arXiv:2211.07647

An Interpretable Neuron Embedding for Static Knowledge Distillation

Although deep neural networks have shown well-performance in various tasks, the poor interpretability of the models is always criticized. In the paper, we propose a new interpretable neural network method, by embedding neurons into the semantic space to extract their intrinsic global semantics. In contrast to previous methods that probe latent knowledge inside the model, the proposed semantic vector externalizes the latent knowledge to static knowledge, which is easy to exploit. Specifically, we assume that neurons with similar activation are of similar semantic information. Afterwards, semantic vectors are optimized by continuously aligning activation similarity and semantic vector similarity during the training of the neural network. The visualization of semantic vectors allows for a qualitative explanation of the neural network. Moreover, we assess the static knowledge quantitatively by knowledge distillation tasks. Empirical experiments of visualization show that semantic vectors describe neuron activation semantics well. Without the sample-by-sample guidance from the teacher model, static knowledge distillation exhibit comparable or even superior performance with existing relation-based knowledge distillation methods.

arXiv:2211.07364

Feature Correlation-guided Knowledge Transfer for Federated Self-supervised Learning

To eliminate the requirement of fully-labeled data for supervised model training in traditional Federated Learning (FL), extensive attention has been paid to the application of Self-supervised Learning (SSL) approaches on FL to tackle the label scarcity problem. Previous works on Federated SSL generally fall into two categories: parameter-based model aggregation (i.e., FedAvg, applicable to homogeneous cases) or data-based feature sharing (i.e., knowledge distillation, applicable to heterogeneous cases) to achieve knowledge transfer among multiple unlabeled clients. Despite the progress, all of them inevitably rely on some assumptions, such as homogeneous models or the existence of an additional public dataset, which hinder the universality of the training frameworks for more general scenarios. Therefore, in this paper, we propose a novel and general method named Federated Self-supervised Learning with Feature-correlation based Aggregation (FedFoA) to tackle the above limitations in a communication-efficient and privacy-preserving manner. Our insight is to utilize feature correlation to align the feature mappings and calibrate the local model updates across clients during their local training process. More specifically, we design a factorization-based method to extract the cross-feature relation matrix from the local representations. Then, the relation matrix can be regarded as a carrier of semantic information to perform the aggregation phase. We prove that FedFoA is a model-agnostic training framework and can be easily compatible with state-of-the-art unsupervised FL methods. Extensive empirical experiments demonstrate that our proposed approach outperforms the state-of-the-art methods by a significant margin.

arXiv:2211.07248공식 코드 · Papers with Code

FedCL: Federated Multi-Phase Curriculum Learning to Synchronously Correlate User Heterogeneity

Federated Learning (FL) is a decentralized learning method used to train machine learning algorithms. In FL, a global model iteratively collects the parameters of local models without accessing their local data. However, a significant challenge in FL is handling the heterogeneity of local data distribution, which often results in a drifted global model that is difficult to converge. To address this issue, current methods employ different strategies such as knowledge distillation, weighted model aggregation, and multi-task learning. These approaches are referred to as asynchronous FL, as they align user models either locally or post-hoc, where model drift has already occurred or has been underestimated. In this paper, we propose an active and synchronous correlation approach to address the challenge of user heterogeneity in FL. Specifically, our approach aims to approximate FL as standard deep learning by actively and synchronously scheduling user learning pace in each round with a dynamic multi-phase curriculum. A global curriculum is formed by an auto-regressive auto-encoder that integrates all user curricula on the server. This global curriculum is then divided into multiple phases and broadcast to users to measure and align the domain-agnostic learning pace. Empirical studies demonstrate that our approach outperforms existing asynchronous approaches in terms of generalization performance, even in the presence of severe user heterogeneity.

arXiv:2211.06816

Long-Range Zero-Shot Generative Deep Network Quantization

Quantization approximates a deep network model with floating-point numbers by the one with low bit width numbers, in order to accelerate inference and reduce computation. Quantizing a model without access to the original data, zero-shot quantization can be accomplished by fitting the real data distribution by data synthesis. However, zero-shot quantization achieves inferior performance compared to the post-training quantization with real data. We find it is because: 1) a normal generator is hard to obtain high diversity of synthetic data, since it lacks long-range information to allocate attention to global features; 2) the synthetic images aim to simulate the statistics of real data, which leads to weak intra-class heterogeneity and limited feature richness. To overcome these problems, we propose a novel deep network quantizer, dubbed Long-Range Zero-Shot Generative Deep Network Quantization (LRQ). Technically, we propose a long-range generator to learn long-range information instead of simple local features. In order for the synthetic data to contain more global features, long-range attention using large kernel convolution is incorporated into the generator. In addition, we also present an Adversarial Margin Add (AMA) module to force intra-class angular enlargement between feature vector and class center. As AMA increases the convergence difficulty of the loss function, which is opposite to the training objective of the original loss function, it forms an adversarial process. Furthermore, in order to transfer knowledge from the full-precision network, we also utilize a decoupled knowledge distillation. Extensive experiments demonstrate that LRQ obtains better performance than other competitors.

arXiv:2211.07198공식 코드 · Papers with Code

Fcaformer: Forward Cross Attention in Hybrid Vision Transformer

Currently, one main research line in designing a more efficient vision transformer is reducing the computational cost of self attention modules by adopting sparse attention or using local attention windows. In contrast, we propose a different approach that aims to improve the performance of transformer-based architectures by densifying the attention pattern. Specifically, we proposed forward cross attention for hybrid vision transformer (FcaFormer), where tokens from previous blocks in the same stage are secondary used. To achieve this, the FcaFormer leverages two innovative components: learnable scale factors (LSFs) and a token merge and enhancement module (TME). The LSFs enable efficient processing of cross tokens, while the TME generates representative cross tokens. By integrating these components, the proposed FcaFormer enhances the interactions of tokens across blocks with potentially different semantics, and encourages more information flows to the lower levels. Based on the forward cross attention (Fca), we have designed a series of FcaFormer models that achieve the best trade-off between model size, computational cost, memory cost, and accuracy. For example, without the need for knowledge distillation to strengthen training, our FcaFormer achieves 83.1% top-1 accuracy on Imagenet with only 16.3 million parameters and about 3.6 billion MACs. This saves almost half of the parameters and a few computational costs while achieving 0.7% higher accuracy compared to distilled EfficientFormer.

arXiv:2211.07171공식 코드 · Papers with Code

Cross-Modality Knowledge Distillation Network for Monocular 3D Object Detection

Leveraging LiDAR-based detectors or real LiDAR point data to guide monocular 3D detection has brought significant improvement, e.g., Pseudo-LiDAR methods. However, the existing methods usually apply non-end-to-end training strategies and insufficiently leverage the LiDAR information, where the rich potential of the LiDAR data has not been well exploited. In this paper, we propose the Cross-Modality Knowledge Distillation (CMKD) network for monocular 3D detection to efficiently and directly transfer the knowledge from LiDAR modality to image modality on both features and responses. Moreover, we further extend CMKD as a semi-supervised training framework by distilling knowledge from large-scale unlabeled data and significantly boost the performance. Until submission, CMKD ranks $1^{st}$ among the monocular 3D detectors with publications on both KITTI $test$ set and Waymo $val$ set with significant performance gains compared to previous state-of-the-art methods.

arXiv:2211.06059

PILE: Pairwise Iterative Logits Ensemble for Multi-Teacher Labeled Distillation

Pre-trained language models have become a crucial part of ranking systems and achieved very impressive effects recently. To maintain high performance while keeping efficient computations, knowledge distillation is widely used. In this paper, we focus on two key questions in knowledge distillation for ranking models: 1) how to ensemble knowledge from multi-teacher; 2) how to utilize the label information of data in the distillation process. We propose a unified algorithm called Pairwise Iterative Logits Ensemble (PILE) to tackle these two questions simultaneously. PILE ensembles multi-teacher logits supervised by label information in an iterative way and achieved competitive performance in both offline and online experiments. The proposed method has been deployed in a real-world commercial search system.

arXiv:2211.06143

FAN-Trans: Online Knowledge Distillation for Facial Action Unit Detection

Due to its importance in facial behaviour analysis, facial action unit (AU) detection has attracted increasing attention from the research community. Leveraging the online knowledge distillation framework, we propose the ``FANTrans" method for AU detection. Our model consists of a hybrid network of convolution and transformer blocks to learn per-AU features and to model AU co-occurrences. The model uses a pre-trained face alignment network as the feature extractor. After further transformation by a small learnable add-on convolutional subnet, the per-AU features are fed into transformer blocks to enhance their representation. As multiple AUs often appear together, we propose a learnable attention drop mechanism in the transformer block to learn the correlation between the features for different AUs. We also design a classifier that predicts AU presence by considering all AUs' features, to explicitly capture label dependencies. Finally, we make the attempt of adapting online knowledge distillation in the training stage for this task, further improving the model's performance. Experiments on the BP4D and DISFA datasets demonstrating the effectiveness of proposed method.