Posted Content•
Neural Architecture Search with Reinforcement Learning
TL;DR: This paper uses a recurrent network to generate the model descriptions of neural networks and trains this RNN with reinforcement learning to maximize the expected accuracy of the generated architectures on a validation set.
Abstract: Neural networks are powerful and flexible models that work well for many difficult learning tasks in image, speech and natural language understanding. Despite their success, neural networks are still hard to design. In this paper, we use a recurrent network to generate the model descriptions of neural networks and train this RNN with reinforcement learning to maximize the expected accuracy of the generated architectures on a validation set. On the CIFAR-10 dataset, our method, starting from scratch, can design a novel network architecture that rivals the best human-invented architecture in terms of test set accuracy. Our CIFAR-10 model achieves a test error rate of 3.65, which is 0.09 percent better and 1.05x faster than the previous state-of-the-art model that used a similar architectural scheme. On the Penn Treebank dataset, our model can compose a novel recurrent cell that outperforms the widely-used LSTM cell, and other state-of-the-art baselines. Our cell achieves a test set perplexity of 62.4 on the Penn Treebank, which is 3.6 perplexity better than the previous state-of-the-art model. The cell can also be transferred to the character language modeling task on PTB and achieves a state-of-the-art perplexity of 1.214.
Citations
More filters
17 Sep 2019
TL;DR: In this paper, a framework to automatically construct a model based on 3DCNN by network architecture search (NAS) is proposed, and the result shows the approach is superiority against prior methods both in efficiency and accuracy.
Abstract: 3D convolutional neural network (3DCNN) is a powerful and effective model utilizing spatial-temporal features, especially for gesture recognition. Unfortunately, so many parameters are modified in 3DCNN that lots of researchers choose 2DCNN or hybrid models, but these models are designed manually. In this paper, we propose a framework to automatically construct a model based on 3DCNN by network architecture search (NAS) [1]. In our method called 3DNAS, a 3D teacher network is trained from scratch as a pre-trained model to accelerate the convergence of the child networks. Then series of child networks with various architectures are generated randomly and each is trained under the direction of converted teacher model. Finally, the controller predicts a network architecture according to the rewards of all the child networks. We evaluate our method on a video-based gesture recognition dataset 20BN-Jester dataset v1 [2] and the result shows our approach is superiority against prior methods both in efficiency and accuracy.
1 citations
TL;DR: A two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks is proposed and empirical results show that this approach outperforms native Mish using SqueezeNet backbone with an average precision of 2.51% and top-1accuracy of 1.20%.
Abstract: A very complex task in deep learning such as image classification must be solved with the help of neural networks and activation functions. The backpropagation algorithm advances backward from the output layer towards the input layer, the gradients often get smaller and smaller and approach zero which eventually leaves the weights of the initial or lower layers nearly unchanged, as a result, the gradient descent never converges to the optimum. We propose a two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks. Our method uses two factors, beta (β) and alpha (α), to normalize the area below the boundary in the Mish activation function and we regard these elements as Bea. Bea-Mish provide a clear understanding of the behaviors and conditions governing this regularization term can lead to a more principled approach for constructing better performing activation functions. We evaluate Bea-Mish results against Mish and Swish activation functions in various models and data sets. Empirical results show that our approach (Bea-Mish) outperforms native Mish using SqueezeNet backbone with an average precision (AP50val) of 2.51% in CIFAR-10 and top-1accuracy in ResNet-50 on ImageNet-1k. shows an improvement of 1.20%.
1 citations
TL;DR: The proposed LDP framework is a lightweight dense prediction neural architecture search (NAS) framework that applies the novel Assisted Tabu Search for efficient architecture exploration and yields consistent improvements on all tested dense prediction tasks.
Abstract: We present LDP, a lightweight dense prediction neural architecture search (NAS) framework. Starting from a predefined generic backbone, LDP applies the novel Assisted Tabu Search for efficient architecture exploration. LDP is fast and suitable for various dense estimation problems, unlike previous NAS methods that are either computational demanding or deployed only for a single subtask. The performance of LPD is evaluated on monocular depth estimation, semantic segmentation, and image super-resolution tasks on diverse datasets, including NYU-Depth-v2, KITTI, Cityscapes, COCO-stuff, DIV2K, Set5, Set14, BSD100, Urban100. Experiments show that the proposed framework yields consistent improvements on all tested dense prediction tasks, while being 5% − 315% more compact in terms of the number of model parameters than prior arts. Dense prediction is a class of computer vision problems aiming at mapping every pixel of the input image with some predicted values. Depending on the problem, the output values can be either continous or discrete. For instance, monocular depth estimation and image super-resolution are often formulated as regression, while semantic segmentation is a dense classification, i.e. discrete, problem. More specifically, the monocular depth estimation problem produces a dense depth map from a single image to be used in various applications including robotics, scene understanding, and augmented reality. Single image super-resolution (SISR) is a low-level vision task that generates a highresolution image from its low-resolution counterpart. SISR is widely utilized in medical and surveillance imaging, where images with more precise details can provide invaluable information. On the other hand, semantic segmentation predicts a dense annotated map of different semantic categories from a given image that is crucial for image understanding tasks. Recent deep neural networks (DNN) exhibits remarkable results on dense prediction, especially subproblems such as single image depth estimation [10, 17, 31, 47, 52, 62, 69, LDP Model Devices
1 citations
TL;DR: In this paper , the authors formulate the machine learning mechanism as a bi-level optimization problem, where the inner level optimization loop involves minimizing a properly chosen loss function evaluated on the training data.
Abstract: This work formulates the machine learning mechanism as a bi-level optimization problem. The inner level optimization loop entails minimizing a properly chosen loss function evaluated on the training data. This is nothing but the well-studied training process in pursuit of optimal model parameters. The outer level optimization loop is less well-studied and involves maximizing a properly chosen performance metric evaluated on the validation data. This is what we call the “iteration process”, pursuing optimal model hyper-parameters. Among many other degrees of freedom, this process entails model engineering (e.g., neural network architecture design) and management, experiment tracking, dataset versioning and augmentation. The iteration process could be automated via Automatic Machine Learning (AutoML) or left to the intuitions of machine learning students, engineers, and researchers. Regardless of the route we take, there is a need to reduce the computational cost of the iteration step and as a direct consequence reduce the carbon footprint of developing artificial intelligence algorithms. Despite the clean and unified mathematical formulation of the iteration step as a bi-level optimization problem, its solutions are case specific and complex. This work will consider such cases while increasing the level of complexity from supervised learning to semi-supervised, self-supervised, unsupervised, few-shot, federated, reinforcement, and physics-informed learning. As a consequence of this exercise, this proposal surfaces a plethora of open problems in the field, many of which can be addressed in parallel.
1 citations
TL;DR: A meta-learning framework to warm-start Differentiable architecture search (DARTS), a NAS method that can be initialized with a transferred architecture and is able to quickly adapt to new tasks, and a simple meta-transfer architecture that was learned over multiple tasks is employed.
Abstract: Neural architecture search (NAS) has shown great promise in the field of automated machine learning (AutoML). NAS has outperformed hand-designed networks and made a significant step forward in the field of automating the design of deep neural networks, thus further reducing the need for human expertise. However, most research is done targeting a single specific task, leaving research of NAS methods over multiple tasks mostly over-looked. Generally, there exist two popular ways to find an architecture for some novel task. Either searching from scratch, which is ineffective by design, or transferring discovered architectures from other tasks, which provides no performance guarantees and is probably not optimal. In this work we present a meta-learning framework to warm-start Differentiable architecture search (DARTS). DARTS is a NAS method that can be initialized with a transferred architecture and is able to quickly adapt to new tasks. A task similarity measure is used to determine which transfer architecture is selected, as transfer architectures found on similar tasks will likely perform better. Additionally, we employ a simple meta-transfer architecture that was learned over multiple tasks. Experiments show that warm-started DARTS is able to find competitive performing architectures while reducing searching cost on average by 60%. P-DARTS. Comparison between the two warm-starting approaches, transfer architecture λ i and learned transfer architecture ˆ λ i , shows that learned transfer architecture, ˆ λ i , requires more computational resources. ˆ λ i is on average faster than P-DARTS by only 58 . 5% compared to λ i ’s 64 . 7%. This difference is especially visible on smaller datasets ( aircraft , flower , birds and dtd ), whereas on tiny imagenet there was no significant statistical difference in means at p < 0 . 05.
1 citations
References
More filters
27 Jun 2016
TL;DR: In this article, the authors proposed a residual learning framework to ease the training of networks that are substantially deeper than those used previously, which won the 1st place on the ILSVRC 2015 classification task.
Abstract: Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.
123,388 citations
Proceedings Article•
01 Jan 2015TL;DR: This work introduces Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments, and provides a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework.
Abstract: We introduce Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. The method is straightforward to implement, is computationally efficient, has little memory requirements, is invariant to diagonal rescaling of the gradients, and is well suited for problems that are large in terms of data and/or parameters. The method is also appropriate for non-stationary objectives and problems with very noisy and/or sparse gradients. The hyper-parameters have intuitive interpretations and typically require little tuning. Some connections to related algorithms, on which Adam was inspired, are discussed. We also analyze the theoretical convergence properties of the algorithm and provide a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework. Empirical results demonstrate that Adam works well in practice and compares favorably to other stochastic optimization methods. Finally, we discuss AdaMax, a variant of Adam based on the infinity norm.
111,197 citations
Proceedings Article•
04 Sep 2014TL;DR: This work investigates the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting using an architecture with very small convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers.
Abstract: In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.
55,235 citations
"Neural Architecture Search with Rei..." refers methods in this paper
...Along with this success is a paradigm shift from feature designing to architecture designing, i.e., from SIFT (Lowe, 1999), and HOG (Dalal & Triggs, 2005), to AlexNet (Krizhevsky et al., 2012), VGGNet (Simonyan & Zisserman, 2014), GoogleNet (Szegedy et al., 2015), and ResNet (He et al., 2016a)....
[...]
01 Jan 1998
TL;DR: In this article, a graph transformer network (GTN) is proposed for handwritten character recognition, which can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters.
Abstract: Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.
42,067 citations
20 Jun 2005
TL;DR: It is shown experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection, and the influence of each stage of the computation on performance is studied.
Abstract: We study the question of feature sets for robust visual object recognition; adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.
31,952 citations
"Neural Architecture Search with Rei..." refers methods in this paper
...Along with this success is a paradigm shift from feature designing to architecture designing, i.e., from SIFT (Lowe, 1999), and HOG (Dalal & Triggs, 2005), to AlexNet (Krizhevsky et al., 2012), VGGNet (Simonyan & Zisserman, 2014), GoogleNet (Szegedy et al., 2015), and ResNet (He et al., 2016a)....
[...]