اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدDSD: Dense-Sparse-Dense Training for Deep Neural Networks
74
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Modern deep neural networks have a large number of parameters, making them very hard to train. We propose DSD, a dense-sparse-dense training flow, for regularizing deep neural networks and achieving better optimization performance. In the first D (Dense) step, we train a dense network to learn connection weights and importance. In the S (Sparse) step, we regularize the network by pruning the unimportant connections with small weights and retraining the network given the sparsity constraint. In the final D (re-Dense) step, we increase the model capacity by removing the sparsity constraint, re-initialize the pruned parameters from zero and retrain the whole dense network. Experiments show that DSD training can improve the performance for a wide range of CNNs, RNNs and LSTMs on the tasks of image classification, caption generation and speech recognition. On ImageNet, DSD improved the Top1 accuracy of GoogLeNet by 1.1%, VGG-16 by 4.3%, ResNet-18 by 1.2% and ResNet-50 by 1.1%, respectively. On the WSJ93 dataset, DSD improved DeepSpeech and DeepSpeech2 WER by 2.0% and 1.1%. On the Flickr-8K dataset, DSD improved the NeuralTalk BLEU score by over 1.7. DSD is easy to use in practice: at training time, DSD incurs only one extra hyper-parameter: the sparsity ratio in the S step. At testing time, DSD doesnt change the network architecture or incur any inference overhead. The consistent and significant performance gain of DSD experiments shows the inadequacy of the current training methods for finding the best local optimum, while DSD effectively achieves superior optimization performance for finding a better solution. DSD models are available to download at https://songhan.github.io/DSD.
قيم البحث
اقرأ أيضاً
Semantic segmentation is pixel-wise classification which retains critical spatial information. The feature map reuse has been commonly adopted in CNN based approaches to take advantage of feature maps in the early layers for the later spatial reconst
ruction. Along this direction, we go a step further by proposing a fully dense neural network with an encoder-decoder structure that we abbreviate as FDNet. For each stage in the decoder module, feature maps of all the previous blocks are adaptively aggregated to feed-forward as input. On the one hand, it reconstructs the spatial boundaries accurately. On the other hand, it learns more efficiently with the more efficient gradient backpropagation. In addition, we propose the boundary-aware loss function to focus more attention on the pixels near the boundary, which boosts the hard examples labeling. We have demonstrated the best performance of the FDNet on the two benchmark datasets: PASCAL VOC 2012, NYUDv2 over previous works when not considering training on other datasets.
Pruning has become a promising technique used to compress and accelerate neural networks. Existing methods are mainly evaluated on spare labeling applications. However, dense labeling applications are those closer to real world problems that require
real-time processing on resource-constrained mobile devices. Pruning for dense labeling applications is still a largely unexplored field. The prevailing filter channel pruning method removes the entire filter channel. Accordingly, the interaction between each kernel in one filter channel is ignored. In this study, we proposed kernel cluster pruning (KCP) to prune dense labeling networks. We developed a clustering technique to identify the least representational kernels in each layer. By iteratively removing those kernels, the parameter that can better represent the entire network is preserved; thus, we achieve better accuracy with a decent model size and computation reduction. When evaluated on stereo matching and semantic segmentation neural networks, our method can reduce more than 70% of FLOPs with less than 1% of accuracy drop. Moreover, for ResNet-50 on ILSVRC-2012, our KCP can reduce more than 50% of FLOPs reduction with 0.13% Top-1 accuracy gain. Therefore, KCP achieves state-of-the-art pruning results.
Face image animation from a single image has achieved remarkable progress. However, it remains challenging when only sparse landmarks are available as the driving signal. Given a source face image and a sequence of sparse face landmarks, our goal is
to generate a video of the face imitating the motion of landmarks. We develop an efficient and effective method for motion transfer from sparse landmarks to the face image. We then combine global and local motion estimation in a unified model to faithfully transfer the motion. The model can learn to segment the moving foreground from the background and generate not only global motion, such as rotation and translation of the face, but also subtle local motion such as the gaze change. We further improve face landmark detection on videos. With temporally better aligned landmark sequences for training, our method can generate temporally coherent videos with higher visual quality. Experiments suggest we achieve results comparable to the state-of-the-art image driven method on the same identity testing and better results on cross identity testing.
Obtaining viewer responses from videos can be useful for creators and streaming platforms to analyze the video performance and improve the future user experience. In this report, we present our method for 2021 Evoked Expression from Videos Challenge.
In particular, our model utilizes both audio and image modalities as inputs to predict emotion changes of viewers. To model long-range emotion changes, we use a GRU-based model to predict one sparse signal with 1Hz. We observe that the emotion changes are smooth. Therefore, the final dense prediction is obtained via linear interpolating the signal, which is robust to the prediction fluctuation. Albeit simple, the proposed method has achieved pearsons correlation score of 0.04430 on the final private test set.
Dense pixelwise prediction such as semantic segmentation is an up-to-date challenge for deep convolutional neural networks (CNNs). Many state-of-the-art approaches either tackle the loss of high-resolution information due to pooling in the encoder st
age, or use dilated convolutions or high-resolution lanes to maintain detailed feature maps and predictions. Motivated by the structural analogy between multi-resolution wavelet analysis and the pooling/unpooling layers of CNNs, we introduce discrete wavelet transform (DWT) into the CNN encoder-decoder architecture and propose WCNN. The high-frequency wavelet coefficients are computed at encoder, which are later used at the decoder to unpooled jointly with coarse-resolution feature maps through the inverse DWT. The DWT/iDWT is further used to develop two wavelet pyramids to capture the global context, where the multi-resolution DWT is applied to successively reduce the spatial resolution and increase the receptive field. Experiment with the Cityscape dataset, the proposed WCNNs are computationally efficient and yield improvements the accuracy for high-resolution dense pixelwise prediction.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
