PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2007.05299

Data-Efficient Ranking Distillation for Image Retrieval

Recent advances in deep learning has lead to rapid developments in the field of image retrieval. However, the best performing architectures incur significant computational cost. Recent approaches tackle this issue using knowledge distillation to transfer knowledge from a deeper and heavier architecture to a much smaller network. In this paper we address knowledge distillation for metric learning problems. Unlike previous approaches, our proposed method jointly addresses the following constraints i) limited queries to teacher model, ii) black box teacher model with access to the final output representation, and iii) small fraction of original training data without any ground-truth labels. In addition, the distillation method does not require the student and teacher to have same dimensionality. Addressing these constraints reduces computation requirements, dependency on large-scale training datasets and addresses practical scenarios of limited or partial access to private data such as teacher models or the corresponding training data/labels. The key idea is to augment the original training set with additional samples by performing linear interpolation in the final output representation space. Distillation is then performed in the joint space of original and augmented teacher-student sample representations. Results demonstrate that our approach can match baseline models trained with full supervision. In low training sample settings, our approach outperforms the fully supervised approach on two challenging image retrieval datasets, ROxford5k and RParis6k \cite{Roxf} with the least possible teacher supervision.

arXiv:2007.05146

Optical Flow Distillation: Towards Efficient and Stable Video Style Transfer

Video style transfer techniques inspire many exciting applications on mobile devices. However, their efficiency and stability are still far from satisfactory. To boost the transfer stability across frames, optical flow is widely adopted, despite its high computational complexity, e.g. occupying over 97% inference time. This paper proposes to learn a lightweight video style transfer network via knowledge distillation paradigm. We adopt two teacher networks, one of which takes optical flow during inference while the other does not. The output difference between these two teacher networks highlights the improvements made by optical flow, which is then adopted to distill the target student network. Furthermore, a low-rank distillation loss is employed to stabilize the output of student network by mimicking the rank of input videos. Extensive experiments demonstrate that our student network without an optical flow module is still able to generate stable video and runs much faster than the teacher network.

arXiv:2007.04174공식 코드 · Papers with Code

Robust Re-Identification by Multiple Views Knowledge Distillation

To achieve robustness in Re-Identification, standard methods leverage tracking information in a Video-To-Video fashion. However, these solutions face a large drop in performance for single image queries (e.g., Image-To-Video setting). Recent works address this severe degradation by transferring temporal information from a Video-based network to an Image-based one. In this work, we devise a training strategy that allows the transfer of a superior knowledge, arising from a set of views depicting the target object. Our proposal - Views Knowledge Distillation (VKD) - pins this visual variety as a supervision signal within a teacher-student framework, where the teacher educates a student who observes fewer views. As a result, the student outperforms not only its teacher but also the current state-of-the-art in Image-To-Video by a wide margin (6.3% mAP on MARS, 8.6% on Duke-Video-ReId and 5% on VeRi-776). A thorough analysis - on Person, Vehicle and Animal Re-ID - investigates the properties of VKD from a qualitatively and quantitatively perspective. Code is available at https://github.com/aimagelab/VKD.

arXiv:2007.04108공식 코드 · Papers with Code

Tracking-by-Trackers with a Distilled and Reinforced Model

Visual object tracking was generally tackled by reasoning independently on fast processing algorithms, accurate online adaptation methods, and fusion of trackers. In this paper, we unify such goals by proposing a novel tracking methodology that takes advantage of other visual trackers, offline and online. A compact student model is trained via the marriage of knowledge distillation and reinforcement learning. The first allows to transfer and compress tracking knowledge of other trackers. The second enables the learning of evaluation measures which are then exploited online. After learning, the student can be ultimately used to build (i) a very fast single-shot tracker, (ii) a tracker with a simple and effective online adaptation mechanism, (iii) a tracker that performs fusion of other trackers. Extensive validation shows that the proposed algorithms compete with real-time state-of-the-art trackers.

arXiv:2007.01951공식 코드 · Papers with Code

Improving Weakly Supervised Visual Grounding by Contrastive Knowledge Distillation

Weakly supervised phrase grounding aims at learning region-phrase correspondences using only image-sentence pairs. A major challenge thus lies in the missing links between image regions and sentence phrases during training. To address this challenge, we leverage a generic object detector at training time, and propose a contrastive learning framework that accounts for both region-phrase and image-sentence matching. Our core innovation is the learning of a region-phrase score function, based on which an image-sentence score function is further constructed. Importantly, our region-phrase score function is learned by distilling from soft matching scores between the detected object names and candidate phrases within an image-sentence pair, while the image-sentence score function is supervised by ground-truth image-sentence pairs. The design of such score functions removes the need of object detection at test time, thereby significantly reducing the inference cost. Without bells and whistles, our approach achieves state-of-the-art results on visual phrase grounding, surpassing previous methods that require expensive object detectors at test time.

arXiv:2007.01922

Knowledge Distillation Beyond Model Compression

Knowledge distillation (KD) is commonly deemed as an effective model compression technique in which a compact model (student) is trained under the supervision of a larger pretrained model or an ensemble of models (teacher). Various techniques have been proposed since the original formulation, which mimic different aspects of the teacher such as the representation space, decision boundary, or intra-data relationship. Some methods replace the one-way knowledge distillation from a static teacher with collaborative learning between a cohort of students. Despite the recent advances, a clear understanding of where knowledge resides in a deep neural network and an optimal method for capturing knowledge from teacher and transferring it to student remains an open question. In this study, we provide an extensive study on nine different KD methods which covers a broad spectrum of approaches to capture and transfer knowledge. We demonstrate the versatility of the KD framework on different datasets and network architectures under varying capacity gaps between the teacher and student. The study provides intuition for the effects of mimicking different aspects of the teacher and derives insights from the performance of the different distillation approaches to guide the design of more effective KD methods. Furthermore, our study shows the effectiveness of the KD framework in learning efficiently under varying severity levels of label noise and class imbalance, consistently providing generalization gains over standard training. We emphasize that the efficacy of KD goes much beyond a model compression technique and it should be considered as a general-purpose training paradigm which offers more robustness to common challenges in the real-world datasets compared to the standard training procedure.

arXiv:2007.01476

Interactive Knowledge Distillation

Knowledge distillation is a standard teacher-student learning framework to train a light-weight student network under the guidance of a well-trained large teacher network. As an effective teaching strategy, interactive teaching has been widely employed at school to motivate students, in which teachers not only provide knowledge but also give constructive feedback to students upon their responses, to improve their learning performance. In this work, we propose an InterActive Knowledge Distillation (IAKD) scheme to leverage the interactive teaching strategy for efficient knowledge distillation. In the distillation process, the interaction between teacher and student networks is implemented by a swapping-in operation: randomly replacing the blocks in the student network with the corresponding blocks in the teacher network. In the way, we directly involve the teacher's powerful feature transformation ability to largely boost the student's performance. Experiments with typical settings of teacher-student networks demonstrate that the student networks trained by our IAKD achieve better performance than those trained by conventional knowledge distillation methods on diverse image classification datasets.

arXiv:2007.00051

Extracurricular Learning: Knowledge Transfer Beyond Empirical Distribution

Knowledge distillation has been used to transfer knowledge learned by a sophisticated model (teacher) to a simpler model (student). This technique is widely used to compress model complexity. However, in most applications the compressed student model suffers from an accuracy gap with its teacher. We propose extracurricular learning, a novel knowledge distillation method, that bridges this gap by (1) modeling student and teacher output distributions; (2) sampling examples from an approximation to the underlying data distribution; and (3) matching student and teacher output distributions over this extended set including uncertain samples. We conduct rigorous evaluations on regression and classification tasks and show that compared to the standard knowledge distillation, extracurricular learning reduces the gap by 46% to 68%. This leads to major accuracy improvements compared to the empirical risk minimization-based training for various recent neural network architectures: 16% regression error reduction on the MPIIGaze dataset, +3.4% to +9.1% improvement in top-1 classification accuracy on the CIFAR100 dataset, and +2.9% top-1 improvement on the ImageNet dataset.

arXiv:2006.16589

On the Demystification of Knowledge Distillation: A Residual Network Perspective

Knowledge distillation (KD) is generally considered as a technique for performing model compression and learned-label smoothing. However, in this paper, we study and investigate the KD approach from a new perspective: we study its efficacy in training a deeper network without any residual connections. We find that in most of the cases, non-residual student networks perform equally or better than their residual versions trained on raw data without KD (baseline network). Surprisingly, in some cases, they surpass the accuracy of baseline networks even with the inferior teachers. After a certain depth of non-residual student network, the accuracy drop, coming from the removal of residual connections, is substantial, and training with KD boosts the accuracy of the student up to a great extent; however, it does not fully recover the accuracy drop. Furthermore, we observe that the conventional teacher-student view of KD is incomplete and does not adequately explain our findings. We propose a novel interpretation of KD with the Trainee-Mentor hypothesis, which provides a holistic view of KD. We also present two viewpoints, loss landscape, and feature reuse, to explain the interplay between residual connections and KD. We substantiate our claims through extensive experiments on residual networks.

arXiv:2006.15920

Interpreting and Disentangling Feature Components of Various Complexity from DNNs

This paper aims to define, quantify, and analyze the feature complexity that is learned by a DNN. We propose a generic definition for the feature complexity. Given the feature of a certain layer in the DNN, our method disentangles feature components of different complexity orders from the feature. We further design a set of metrics to evaluate the reliability, the effectiveness, and the significance of over-fitting of these feature components. Furthermore, we successfully discover a close relationship between the feature complexity and the performance of DNNs. As a generic mathematical tool, the feature complexity and the proposed metrics can also be used to analyze the success of network compression and knowledge distillation.

arXiv:2006.15710

Motion Pyramid Networks for Accurate and Efficient Cardiac Motion Estimation

Cardiac motion estimation plays a key role in MRI cardiac feature tracking and function assessment such as myocardium strain. In this paper, we propose Motion Pyramid Networks, a novel deep learning-based approach for accurate and efficient cardiac motion estimation. We predict and fuse a pyramid of motion fields from multiple scales of feature representations to generate a more refined motion field. We then use a novel cyclic teacher-student training strategy to make the inference end-to-end and further improve the tracking performance. Our teacher model provides more accurate motion estimation as supervision through progressive motion compensations. Our student model learns from the teacher model to estimate motion in a single step while maintaining accuracy. The teacher-student knowledge distillation is performed in a cyclic way for a further performance boost. Our proposed method outperforms a strong baseline model on two public available clinical datasets significantly, evaluated by a variety of metrics and the inference time. New evaluation metrics are also proposed to represent errors in a clinically meaningful manner.

arXiv:2006.15127

Diverse Knowledge Distillation (DKD): A Solution for Improving The Robustness of Ensemble Models Against Adversarial Attacks

This paper proposes an ensemble learning model that is resistant to adversarial attacks. To build resilience, we introduced a training process where each member learns a radically distinct latent space. Member models are added one at a time to the ensemble. Simultaneously, the loss function is regulated by a reverse knowledge distillation, forcing the new member to learn different features and map to a latent space safely distanced from those of existing members. We assessed the security and performance of the proposed solution on image classification tasks using CIFAR10 and MNIST datasets and showed security and performance improvement compared to the state of the art defense methods.

arXiv:2006.13108

Distilling Object Detectors with Task Adaptive Regularization

Current state-of-the-art object detectors are at the expense of high computational costs and are hard to deploy to low-end devices. Knowledge distillation, which aims at training a smaller student network by transferring knowledge from a larger teacher model, is one of the promising solutions for model miniaturization. In this paper, we investigate each module of a typical detector in depth, and propose a general distillation framework that adaptively transfers knowledge from teacher to student according to the task specific priors. The intuition is that simply distilling all information from teacher to student is not advisable, instead we should only borrow priors from the teacher model where the student cannot perform well. Towards this goal, we propose a region proposal sharing mechanism to interflow region responses between the teacher and student models. Based on this, we adaptively transfer knowledge at three levels, \emph{i.e.}, feature backbone, classification head, and bounding box regression head, according to which model performs more reasonably. Furthermore, considering that it would introduce optimization dilemma when minimizing distillation loss and detection loss simultaneously, we propose a distillation decay strategy to help improve model generalization via gradually reducing the distillation penalty. Experiments on widely used detection benchmarks demonstrate the effectiveness of our method. In particular, using Faster R-CNN with FPN as an instantiation, we achieve an accuracy of $39.0\%$ with Resnet-50 on COCO dataset, which surpasses the baseline $36.3\%$ by $2.7\%$ points, and even better than the teacher model with $38.5\%$ mAP.

arXiv:2006.11487공식 코드 · Papers with Code

Paying more attention to snapshots of Iterative Pruning: Improving Model Compression via Ensemble Distillation

Network pruning is one of the most dominant methods for reducing the heavy inference cost of deep neural networks. Existing methods often iteratively prune networks to attain high compression ratio without incurring significant loss in performance. However, we argue that conventional methods for retraining pruned networks (i.e., using small, fixed learning rate) are inadequate as they completely ignore the benefits from snapshots of iterative pruning. In this work, we show that strong ensembles can be constructed from snapshots of iterative pruning, which achieve competitive performance and vary in network structure. Furthermore, we present simple, general and effective pipeline that generates strong ensembles of networks during pruning with large learning rate restarting, and utilizes knowledge distillation with those ensembles to improve the predictive power of compact models. In standard image classification benchmarks such as CIFAR and Tiny-Imagenet, we advance state-of-the-art pruning ratio of structured pruning by integrating simple l1-norm filters pruning into our pipeline. Specifically, we reduce 75-80% of total parameters and 65-70% MACs of numerous variants of ResNet architectures while having comparable or better performance than that of original networks. Code associate with this paper is made publicly available at https://github.com/lehduong/kesi.

arXiv:2006.12000공식 코드 · Papers with Code

Self-Knowledge Distillation with Progressive Refinement of Targets

The generalization capability of deep neural networks has been substantially improved by applying a wide spectrum of regularization methods, e.g., restricting function space, injecting randomness during training, augmenting data, etc. In this work, we propose a simple yet effective regularization method named progressive self-knowledge distillation (PS-KD), which progressively distills a model's own knowledge to soften hard targets (i.e., one-hot vectors) during training. Hence, it can be interpreted within a framework of knowledge distillation as a student becomes a teacher itself. Specifically, targets are adjusted adaptively by combining the ground-truth and past predictions from the model itself. We show that PS-KD provides an effect of hard example mining by rescaling gradients according to difficulty in classifying examples. The proposed method is applicable to any supervised learning tasks with hard targets and can be easily combined with existing regularization methods to further enhance the generalization performance. Furthermore, it is confirmed that PS-KD achieves not only better accuracy, but also provides high quality of confidence estimates in terms of calibration as well as ordinal ranking. Extensive experimental results on three different tasks, image classification, object detection, and machine translation, demonstrate that our method consistently improves the performance of the state-of-the-art baselines. The code is available at https://github.com/lgcnsai/PS-KD-Pytorch.

arXiv:2006.10369공식 코드 · Papers with Code

Deep Encoder, Shallow Decoder: Reevaluating Non-autoregressive Machine Translation

Much recent effort has been invested in non-autoregressive neural machine translation, which appears to be an efficient alternative to state-of-the-art autoregressive machine translation on modern GPUs. In contrast to the latter, where generation is sequential, the former allows generation to be parallelized across target token positions. Some of the latest non-autoregressive models have achieved impressive translation quality-speed tradeoffs compared to autoregressive baselines. In this work, we reexamine this tradeoff and argue that autoregressive baselines can be substantially sped up without loss in accuracy. Specifically, we study autoregressive models with encoders and decoders of varied depths. Our extensive experiments show that given a sufficiently deep encoder, a single-layer autoregressive decoder can substantially outperform strong non-autoregressive models with comparable inference speed. We show that the speed disadvantage for autoregressive baselines compared to non-autoregressive methods has been overestimated in three aspects: suboptimal layer allocation, insufficient speed measurement, and lack of knowledge distillation. Our results establish a new protocol for future research toward fast, accurate machine translation. Our code is available at https://github.com/jungokasai/deep-shallow.

arXiv:2006.09785공식 코드 · Papers with Code

Self-supervised Knowledge Distillation for Few-shot Learning

Real-world contains an overwhelmingly large number of object classes, learning all of which at once is infeasible. Few shot learning is a promising learning paradigm due to its ability to learn out of order distributions quickly with only a few samples. Recent works [7, 41] show that simply learning a good feature embedding can outperform more sophisticated meta-learning and metric learning algorithms for few-shot learning. In this paper, we propose a simple approach to improve the representation capacity of deep neural networks for few-shot learning tasks. We follow a two-stage learning process: First, we train a neural network to maximize the entropy of the feature embedding, thus creating an optimal output manifold using a self-supervised auxiliary loss. In the second stage, we minimize the entropy on feature embedding by bringing self-supervised twins together, while constraining the manifold with student-teacher distillation. Our experiments show that, even in the first stage, self-supervision can outperform current state-of-the-art methods, with further gains achieved by our second stage distillation process. Our codes are available at: https://github.com/brjathu/SKD.

arXiv:2006.09247

Prior knowledge distillation based on financial time series

One of the major characteristics of financial time series is that they contain a large amount of non-stationary noise, which is challenging for deep neural networks. People normally use various features to address this problem. However, the performance of these features depends on the choice of hyper-parameters. In this paper, we propose to use neural networks to represent these indicators and train a large network constructed of smaller networks as feature layers to fine-tune the prior knowledge represented by the indicators. During back propagation, prior knowledge is transferred from human logic to machine logic via gradient descent. Prior knowledge is the deep belief of neural network and teaches the network to not be affected by non-stationary noise. Moreover, co-distillation is applied to distill the structure into a much smaller size to reduce redundant features and the risk of overfitting. In addition, the decisions of the smaller networks in terms of gradient descent are more robust and cautious than those of large networks. In numerical experiments, we find that our algorithm is faster and more accurate than traditional methods on real financial datasets. We also conduct experiments to verify and comprehend the method.

arXiv:2006.08383

Pixel Invisibility: Detecting Objects Invisible in Color Images

Despite recent success of object detectors using deep neural networks, their deployment on safety-critical applications such as self-driving cars remains questionable. This is partly due to the absence of reliable estimation for detectors' failure under operational conditions such as night, fog, dusk, dawn and glare. Such unquantifiable failures could lead to safety violations. In order to solve this problem, we created an algorithm that predicts a pixel-level invisibility map for color images that does not require manual labeling - that computes the probability that a pixel/region contains objects that are invisible in color domain, during various lighting conditions such as day, night and fog. We propose a novel use of cross modal knowledge distillation from color to infra-red domain using weakly-aligned image pairs from the day and construct indicators for the pixel-level invisibility based on the distances of their intermediate-level features. Quantitative experiments show the great performance of our pixel-level invisibility mask and also the effectiveness of distilled mid-level features on object detection in infra-red imagery.

arXiv:2006.08341공식 코드 · Papers with Code

Multi-fidelity Neural Architecture Search with Knowledge Distillation

Neural architecture search (NAS) targets at finding the optimal architecture of a neural network for a problem or a family of problems. Evaluations of neural architectures are very time-consuming. One of the possible ways to mitigate this issue is to use low-fidelity evaluations, namely training on a part of a dataset, fewer epochs, with fewer channels, etc. In this paper, we propose a bayesian multi-fidelity method for neural architecture search: MF-KD. The method relies on a new approach to low-fidelity evaluations of neural architectures by training for a few epochs using a knowledge distillation. Knowledge distillation adds to a loss function a term forcing a network to mimic some teacher network. We carry out experiments on CIFAR-10, CIFAR-100, and ImageNet-16-120. We show that training for a few epochs with such a modified loss function leads to a better selection of neural architectures than training for a few epochs with a logistic loss. The proposed method outperforms several state-of-the-art baselines.