PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2004.10171

Knowledge Distillation for Multilingual Unsupervised Neural Machine Translation

Unsupervised neural machine translation (UNMT) has recently achieved remarkable results for several language pairs. However, it can only translate between a single language pair and cannot produce translation results for multiple language pairs at the same time. That is, research on multilingual UNMT has been limited. In this paper, we empirically introduce a simple method to translate between thirteen languages using a single encoder and a single decoder, making use of multilingual data to improve UNMT for all language pairs. On the basis of the empirical findings, we propose two knowledge distillation methods to further enhance multilingual UNMT performance. Our experiments on a dataset with English translated to and from twelve other languages (including three language families and six language branches) show remarkable results, surpassing strong unsupervised individual baselines while achieving promising performance between non-English language pairs in zero-shot translation scenarios and alleviating poor performance in low-resource language pairs.

arXiv:2004.08861공식 코드 · Papers with Code

Role-Wise Data Augmentation for Knowledge Distillation

Knowledge Distillation (KD) is a common method for transferring the ``knowledge'' learned by one machine learning model (the \textit{teacher}) into another model (the \textit{student}), where typically, the teacher has a greater capacity (e.g., more parameters or higher bit-widths). To our knowledge, existing methods overlook the fact that although the student absorbs extra knowledge from the teacher, both models share the same input data -- and this data is the only medium by which the teacher's knowledge can be demonstrated. Due to the difference in model capacities, the student may not benefit fully from the same data points on which the teacher is trained. On the other hand, a human teacher may demonstrate a piece of knowledge with individualized examples adapted to a particular student, for instance, in terms of her cultural background and interests. Inspired by this behavior, we design data augmentation agents with distinct roles to facilitate knowledge distillation. Our data augmentation agents generate distinct training data for the teacher and student, respectively. We find empirically that specially tailored data points enable the teacher's knowledge to be demonstrated more effectively to the student. We compare our approach with existing KD methods on training popular neural architectures and demonstrate that role-wise data augmentation improves the effectiveness of KD over strong prior approaches. The code for reproducing our results can be found at https://github.com/bigaidream-projects/role-kd

arXiv:2004.08116

Triplet Loss for Knowledge Distillation

In recent years, deep learning has spread rapidly, and deeper, larger models have been proposed. However, the calculation cost becomes enormous as the size of the models becomes larger. Various techniques for compressing the size of the models have been proposed to improve performance while reducing computational costs. One of the methods to compress the size of the models is knowledge distillation (KD). Knowledge distillation is a technique for transferring knowledge of deep or ensemble models with many parameters (teacher model) to smaller shallow models (student model). Since the purpose of knowledge distillation is to increase the similarity between the teacher model and the student model, we propose to introduce the concept of metric learning into knowledge distillation to make the student model closer to the teacher model using pairs or triplets of the training samples. In metric learning, the researchers are developing the methods to build a model that can increase the similarity of outputs for similar samples. Metric learning aims at reducing the distance between similar and increasing the distance between dissimilar. The functionality of the metric learning to reduce the differences between similar outputs can be used for the knowledge distillation to reduce the differences between the outputs of the teacher model and the student model. Since the outputs of the teacher model for different objects are usually different, the student model needs to distinguish them. We think that metric learning can clarify the difference between the different outputs, and the performance of the student model could be improved. We have performed experiments to compare the proposed method with state-of-the-art knowledge distillation methods.

arXiv:2004.07711

Knowledge Distillation for Action Anticipation via Label Smoothing

Human capability to anticipate near future from visual observations and non-verbal cues is essential for developing intelligent systems that need to interact with people. Several research areas, such as human-robot interaction (HRI), assisted living or autonomous driving need to foresee future events to avoid crashes or help people. Egocentric scenarios are classic examples where action anticipation is applied due to their numerous applications. Such challenging task demands to capture and model domain's hidden structure to reduce prediction uncertainty. Since multiple actions may equally occur in the future, we treat action anticipation as a multi-label problem with missing labels extending the concept of label smoothing. This idea resembles the knowledge distillation process since useful information is injected into the model during training. We implement a multi-modal framework based on long short-term memory (LSTM) networks to summarize past observations and make predictions at different time steps. We perform extensive experiments on EPIC-Kitchens and EGTEA Gaze+ datasets including more than 2500 and 100 action classes, respectively. The experiments show that label smoothing systematically improves performance of state-of-the-art models for action anticipation.

arXiv:2004.07544공식 코드 · Papers with Code

Multimodal and multiview distillation for real-time player detection on a football field

Monitoring the occupancy of public sports facilities is essential to assess their use and to motivate their construction in new places. In the case of a football field, the area to cover is large, thus several regular cameras should be used, which makes the setup expensive and complex. As an alternative, we developed a system that detects players from a unique cheap and wide-angle fisheye camera assisted by a single narrow-angle thermal camera. In this work, we train a network in a knowledge distillation approach in which the student and the teacher have different modalities and a different view of the same scene. In particular, we design a custom data augmentation combined with a motion detection algorithm to handle the training in the region of the fisheye camera not covered by the thermal one. We show that our solution is effective in detecting players on the whole field filmed by the fisheye camera. We evaluate it quantitatively and qualitatively in the case of an online distillation, where the student detects players in real time while being continuously adapted to the latest video conditions.

arXiv:2004.07324

Building a Multi-domain Neural Machine Translation Model using Knowledge Distillation

Lack of specialized data makes building a multi-domain neural machine translation tool challenging. Although emerging literature dealing with low resource languages starts to show promising results, most state-of-the-art models used millions of sentences. Today, the majority of multi-domain adaptation techniques are based on complex and sophisticated architectures that are not adapted for real-world applications. So far, no scalable method is performing better than the simple yet effective mixed-finetuning, i.e finetuning a generic model with a mix of all specialized data and generic data. In this paper, we propose a new training pipeline where knowledge distillation and multiple specialized teachers allow us to efficiently finetune a model without adding new costs at inference time. Our experiments demonstrated that our training pipeline allows improving the performance of multi-domain translation over finetuning in configurations with 2, 3, and 4 domains by up to 2 points in BLEU.

arXiv:2004.07211공식 코드 · Papers with Code

Dark Experience for General Continual Learning: a Strong, Simple Baseline

Continual Learning has inspired a plethora of approaches and evaluation settings; however, the majority of them overlooks the properties of a practical scenario, where the data stream cannot be shaped as a sequence of tasks and offline training is not viable. We work towards General Continual Learning (GCL), where task boundaries blur and the domain and class distributions shift either gradually or suddenly. We address it through mixing rehearsal with knowledge distillation and regularization; our simple baseline, Dark Experience Replay, matches the network's logits sampled throughout the optimization trajectory, thus promoting consistency with its past. By conducting an extensive analysis on both standard benchmarks and a novel GCL evaluation setting (MNIST-360), we show that such a seemingly simple baseline outperforms consolidated approaches and leverages limited resources. We further explore the generalization capabilities of our objective, showing its regularization being beneficial beyond mere performance.

arXiv:2004.06297

Smart Inference for Multidigit Convolutional Neural Network based Barcode Decoding

Barcodes are ubiquitous and have been used in most of critical daily activities for decades. However, most of traditional decoders require well-founded barcode under a relatively standard condition. While wilder conditioned barcodes such as underexposed, occluded, blurry, wrinkled and rotated are commonly captured in reality, those traditional decoders show weakness of recognizing. Several works attempted to solve those challenging barcodes, but many limitations still exist. This work aims to solve the decoding problem using deep convolutional neural network with the possibility of running on portable devices. Firstly, we proposed a special modification of inference based on the feature of having checksum and test-time augmentation, named as Smart Inference (SI) in prediction phase of a trained model. SI considerably boosts accuracy and reduces the false prediction for trained models. Secondly, we have created a large practical evaluation dataset of real captured 1D barcode under various challenging conditions to test our methods vigorously, which is publicly available for other researchers. The experiments' results demonstrated the SI effectiveness with the highest accuracy of 95.85% which outperformed many existing decoders on the evaluation set. Finally, we successfully minimized the best model by knowledge distillation to a shallow model which is shown to have high accuracy (90.85%) with good inference speed of 34.2 ms per image on a real edge device.

arXiv:2004.06288

Towards Robust Classification with Image Quality Assessment

Recent studies have shown that deep convolutional neural networks (DCNN) are vulnerable to adversarial examples and sensitive to perceptual quality as well as the acquisition condition of images. These findings raise a big concern for the adoption of DCNN-based applications for critical tasks. In the literature, various defense strategies have been introduced to increase the robustness of DCNN, including re-training an entire model with benign noise injection, adversarial examples, or adding extra layers. In this paper, we investigate the connection between adversarial manipulation and image quality, subsequently propose a protective mechanism that doesnt require re-training a DCNN. Our method combines image quality assessment with knowledge distillation to detect input images that would trigger a DCCN to produce egregiously wrong results. Using the ResNet model trained on ImageNet as an example, we demonstrate that the detector can effectively identify poor quality and adversarial images.

arXiv:2004.05319공식 코드 · Papers with Code

KD-MRI: A knowledge distillation framework for image reconstruction and image restoration in MRI workflow

Deep learning networks are being developed in every stage of the MRI workflow and have provided state-of-the-art results. However, this has come at the cost of increased computation requirement and storage. Hence, replacing the networks with compact models at various stages in the MRI workflow can significantly reduce the required storage space and provide considerable speedup. In computer vision, knowledge distillation is a commonly used method for model compression. In our work, we propose a knowledge distillation (KD) framework for the image to image problems in the MRI workflow in order to develop compact, low-parameter models without a significant drop in performance. We propose a combination of the attention-based feature distillation method and imitation loss and demonstrate its effectiveness on the popular MRI reconstruction architecture, DC-CNN. We conduct extensive experiments using Cardiac, Brain, and Knee MRI datasets for 4x, 5x and 8x accelerations. We observed that the student network trained with the assistance of the teacher using our proposed KD framework provided significant improvement over the student network trained without assistance across all the datasets and acceleration factors. Specifically, for the Knee dataset, the student network achieves $65\%$ parameter reduction, 2x faster CPU running time, and 1.5x faster GPU running time compared to the teacher. Furthermore, we compare our attention-based feature distillation method with other feature distillation methods. We also conduct an ablative study to understand the significance of attention-based distillation and imitation loss. We also extend our KD framework for MRI super-resolution and show encouraging results.

arXiv:2004.05304공식 코드 · Papers with Code

Inter-Region Affinity Distillation for Road Marking Segmentation

We study the problem of distilling knowledge from a large deep teacher network to a much smaller student network for the task of road marking segmentation. In this work, we explore a novel knowledge distillation (KD) approach that can transfer 'knowledge' on scene structure more effectively from a teacher to a student model. Our method is known as Inter-Region Affinity KD (IntRA-KD). It decomposes a given road scene image into different regions and represents each region as a node in a graph. An inter-region affinity graph is then formed by establishing pairwise relationships between nodes based on their similarity in feature distribution. To learn structural knowledge from the teacher network, the student is required to match the graph generated by the teacher. The proposed method shows promising results on three large-scale road marking segmentation benchmarks, i.e., ApolloScape, CULane and LLAMAS, by taking various lightweight models as students and ResNet-101 as the teacher. IntRA-KD consistently brings higher performance gains on all lightweight models, compared to previous distillation methods. Our code is available at https://github.com/cardwing/Codes-for-IntRA-KD.

arXiv:2004.05937공식 코드 · Papers with Code

Knowledge Distillation and Student-Teacher Learning for Visual Intelligence: A Review and New Outlooks

Deep neural models in recent years have been successful in almost every field, including extremely complex problem statements. However, these models are huge in size, with millions (and even billions) of parameters, thus demanding more heavy computation power and failing to be deployed on edge devices. Besides, the performance boost is highly dependent on redundant labeled data. To achieve faster speeds and to handle the problems caused by the lack of data, knowledge distillation (KD) has been proposed to transfer information learned from one model to another. KD is often characterized by the so-called `Student-Teacher' (S-T) learning framework and has been broadly applied in model compression and knowledge transfer. This paper is about KD and S-T learning, which are being actively studied in recent years. First, we aim to provide explanations of what KD is and how/why it works. Then, we provide a comprehensive survey on the recent progress of KD methods together with S-T frameworks typically for vision tasks. In general, we consider some fundamental questions that have been driving this research area and thoroughly generalize the research progress and technical details. Additionally, we systematically analyze the research status of KD in vision applications. Finally, we discuss the potentials and open challenges of existing methods and prospect the future directions of KD and S-T learning.

arXiv:2004.05686

XtremeDistil: Multi-stage Distillation for Massive Multilingual Models

Deep and large pre-trained language models are the state-of-the-art for various natural language processing tasks. However, the huge size of these models could be a deterrent to use them in practice. Some recent and concurrent works use knowledge distillation to compress these huge models into shallow ones. In this work we study knowledge distillation with a focus on multi-lingual Named Entity Recognition (NER). In particular, we study several distillation strategies and propose a stage-wise optimization scheme leveraging teacher internal representations that is agnostic of teacher architecture and show that it outperforms strategies employed in prior works. Additionally, we investigate the role of several factors like the amount of unlabeled data, annotation resources, model architecture and inference latency to name a few. We show that our approach leads to massive compression of MBERT-like teacher models by upto 35x in terms of parameters and 51x in terms of latency for batch inference while retaining 95% of its F1-score for NER over 41 languages.

arXiv:2004.04366

Knowledge Distillation for Mobile Edge Computation Offloading

Edge computation offloading allows mobile end devices to put execution of compute-intensive task on the edge servers. End devices can decide whether offload the tasks to edge servers, cloud servers or execute locally according to current network condition and devices' profile in an online manner. In this article, we propose an edge computation offloading framework based on Deep Imitation Learning (DIL) and Knowledge Distillation (KD), which assists end devices to quickly make fine-grained decisions to optimize the delay of computation tasks online. We formalize computation offloading problem into a multi-label classification problem. Training samples for our DIL model are generated in an offline manner. After model is trained, we leverage knowledge distillation to obtain a lightweight DIL model, by which we further reduce the model's inference delay. Numerical experiment shows that the offloading decisions made by our model outperforms those made by other related policies in latency metric. Also, our model has the shortest inference delay among all policies.

arXiv:2004.03846공식 코드 · Papers with Code

Structure-Level Knowledge Distillation For Multilingual Sequence Labeling

Multilingual sequence labeling is a task of predicting label sequences using a single unified model for multiple languages. Compared with relying on multiple monolingual models, using a multilingual model has the benefit of a smaller model size, easier in online serving, and generalizability to low-resource languages. However, current multilingual models still underperform individual monolingual models significantly due to model capacity limitations. In this paper, we propose to reduce the gap between monolingual models and the unified multilingual model by distilling the structural knowledge of several monolingual models (teachers) to the unified multilingual model (student). We propose two novel KD methods based on structure-level information: (1) approximately minimizes the distance between the student's and the teachers' structure level probability distributions, (2) aggregates the structure-level knowledge to local distributions and minimizes the distance between two local probability distributions. Our experiments on 4 multilingual tasks with 25 datasets show that our approaches outperform several strong baselines and have stronger zero-shot generalizability than both the baseline model and teacher models.

arXiv:2004.03844공식 코드 · Papers with Code

On the Effect of Dropping Layers of Pre-trained Transformer Models

Transformer-based NLP models are trained using hundreds of millions or even billions of parameters, limiting their applicability in computationally constrained environments. While the number of parameters generally correlates with performance, it is not clear whether the entire network is required for a downstream task. Motivated by the recent work on pruning and distilling pre-trained models, we explore strategies to drop layers in pre-trained models, and observe the effect of pruning on downstream GLUE tasks. We were able to prune BERT, RoBERTa and XLNet models up to 40%, while maintaining up to 98% of their original performance. Additionally we show that our pruned models are on par with those built using knowledge distillation, both in terms of size and performance. Our experiments yield interesting observations such as, (i) the lower layers are most critical to maintain downstream task performance, (ii) some tasks such as paraphrase detection and sentence similarity are more robust to the dropping of layers, and (iii) models trained using a different objective function exhibit different learning patterns and w.r.t the layer dropping.

arXiv:2004.04124

LadaBERT: Lightweight Adaptation of BERT through Hybrid Model Compression

BERT is a cutting-edge language representation model pre-trained by a large corpus, which achieves superior performances on various natural language understanding tasks. However, a major blocking issue of applying BERT to online services is that it is memory-intensive and leads to unsatisfactory latency of user requests, raising the necessity of model compression. Existing solutions leverage the knowledge distillation framework to learn a smaller model that imitates the behaviors of BERT. However, the training procedure of knowledge distillation is expensive itself as it requires sufficient training data to imitate the teacher model. In this paper, we address this issue by proposing a hybrid solution named LadaBERT (Lightweight adaptation of BERT through hybrid model compression), which combines the advantages of different model compression methods, including weight pruning, matrix factorization and knowledge distillation. LadaBERT achieves state-of-the-art accuracy on various public datasets while the training overheads can be reduced by an order of magnitude.

arXiv:2004.03020

Enhancing Review Comprehension with Domain-Specific Commonsense

Review comprehension has played an increasingly important role in improving the quality of online services and products and commonsense knowledge can further enhance review comprehension. However, existing general-purpose commonsense knowledge bases lack sufficient coverage and precision to meaningfully improve the comprehension of domain-specific reviews. In this paper, we introduce xSense, an effective system for review comprehension using domain-specific commonsense knowledge bases (xSense KBs). We show that xSense KBs can be constructed inexpensively and present a knowledge distillation method that enables us to use xSense KBs along with BERT to boost the performance of various review comprehension tasks. We evaluate xSense over three review comprehension tasks: aspect extraction, aspect sentiment classification, and question answering. We find that xSense outperforms the state-of-the-art models for the first two tasks and improves the baseline BERT QA model significantly, demonstrating the usefulness of incorporating commonsense into review comprehension pipelines. To facilitate future research and applications, we publicly release three domain-specific knowledge bases and a domain-specific question answering benchmark along with this paper.

arXiv:2004.03303

Towards Efficient Unconstrained Palmprint Recognition via Deep Distillation Hashing

Deep palmprint recognition has become an emerging issue with great potential for personal authentication on handheld and wearable consumer devices. Previous studies of palmprint recognition are mainly based on constrained datasets collected by dedicated devices in controlled environments, which has to reduce the flexibility and convenience. In addition, general deep palmprint recognition algorithms are often too heavy to meet the real-time requirements of embedded system. In this paper, a new palmprint benchmark is established, which consists of more than 20,000 images collected by 5 brands of smart phones in an unconstrained manner. Each image has been manually labeled with 14 key points for region of interest (ROI) extraction. Further, the approach called Deep Distillation Hashing (DDH) is proposed as benchmark for efficient deep palmprint recognition. Palmprint images are converted to binary codes to improve the efficiency of feature matching. Derived from knowledge distillation, novel distillation loss functions are constructed to compress deep model to further improve the efficiency of feature extraction on light network. Comprehensive experiments are conducted on both constrained and unconstrained palmprint databases. Using DDH, the accuracy of palmprint identification can be increased by up to 11.37%, and the Equal Error Rate (EER) of palmprint verification can be reduced by up to 3.11%. The results indicate the feasibility of our database, and DDH can outperform other baselines to achieve the state-of-the-art performance. The collected dataset and related source codes are publicly available at http://gr.xjtu.edu.cn/web/bell/resource.

arXiv:2004.03097

Towards Non-task-specific Distillation of BERT via Sentence Representation Approximation

Recently, BERT has become an essential ingredient of various NLP deep models due to its effectiveness and universal-usability. However, the online deployment of BERT is often blocked by its large-scale parameters and high computational cost. There are plenty of studies showing that the knowledge distillation is efficient in transferring the knowledge from BERT into the model with a smaller size of parameters. Nevertheless, current BERT distillation approaches mainly focus on task-specified distillation, such methodologies lead to the loss of the general semantic knowledge of BERT for universal-usability. In this paper, we propose a sentence representation approximating oriented distillation framework that can distill the pre-trained BERT into a simple LSTM based model without specifying tasks. Consistent with BERT, our distilled model is able to perform transfer learning via fine-tuning to adapt to any sentence-level downstream task. Besides, our model can further cooperate with task-specific distillation procedures. The experimental results on multiple NLP tasks from the GLUE benchmark show that our approach outperforms other task-specific distillation methods or even much larger models, i.e., ELMO, with efficiency well-improved.