RENAS: Reinforced Evolutionary Neural Architecture Search

Chen, Yukang; Meng, Guang; Zhang, Qian; Xiang, Shiming; Huang, Chang; Mu, Lisen; Wang, Xinggang

doi:10.1109/cvpr.2019.00492

Cited by 117 publications

(73 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All the deep learning related frameworks are implemented with the PyTorch package 5 . Every single search experiment is run on a single GPU (NVIDIA GeForce RTX 2080 Ti) with three architectures parallelly trained on it.…”

Section: C3 Software and Hardware Descriptionsmentioning

confidence: 99%

Towards Automated Neural Interaction Discovery for Click-Through Rate Prediction

Song

Cheng

Zhou

et al. 2020

Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

View full text Add to dashboard Cite

Click-Through Rate (CTR) prediction is one of the most important machine learning tasks in recommender systems, driving personalized experience for billions of consumers. Neural architecture search (NAS), as an emerging field, has demonstrated its capabilities in discovering powerful neural network architectures, which motivates us to explore its potential for CTR predictions. Due to 1) diverse unstructured feature interactions, 2) heterogeneous feature space, and 3) high data volume and intrinsic data randomness, it is challenging to construct, search, and compare different architectures effectively for recommendation models. To address these challenges, we propose an automated interaction architecture discovering framework for CTR prediction named AutoCTR. Via modularizing simple yet representative interactions as virtual building blocks and wiring them into a space of direct acyclic graphs, AutoCTR performs evolutionary architecture exploration with learning-to-rank guidance at the architecture level and achieves acceleration using low-fidelity model. Empirical analysis demonstrates the effectiveness of AutoCTR on different datasets comparing to human-crafted architectures. The discovered architecture also enjoys generalizability and transferability among different datasets. CCS CONCEPTS • Information systems → Recommender systems; • Theory of computation → Evolutionary algorithms; • Computing methodologies → Neural networks.

show abstract

Section: C3 Software and Hardware Descriptionsmentioning

confidence: 99%

Towards Automated Neural Interaction Discovery for Click-Through Rate Prediction

Song

Cheng

Zhou

et al. 2020

Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

View full text Add to dashboard Cite

show abstract

“…Furthermore, in this model, the author benefitted from a new regularization method called ScheduledDropPath. Moreover, from the advantage of using Controller Recurrent Neural Network (CRNN), CNN, and reinforced evolutionary algorithm, this model is able to choose the best cell candidate to form the blocks and end up building the best architecture depending on the database [72]- [73]. In this model, the controller RNN generates sample architecture with a sample probability by using a set of operations.…”

Section: ) Neural Architecture Search Network (Nasnet)mentioning

confidence: 99%

Breast Cancer Classification Using Deep Learning Approaches and Histopathology Image: A Comparison Study

et al. 2020

View full text Add to dashboard Cite

Convolutional Neural Network (CNN) models are a type of deep learning architecture introduced to achieve the correct classification of breast cancer. This paper has a twofold purpose. The first aim is to investigate the various deep learning models in classifying breast cancer histopathology images. This study identified the most accurate models in terms of the binary, four, and eight classifications of breast cancer histopathology image databases. The different accuracy scores obtained for the deep learning models on the same database showed that other factors such as pre-processing, data augmentation, and transfer learning methods can impact the ability of the models to achieve higher accuracy. The second purpose of our manuscript is to investigate the latest models that have no or limited examination done in previous studies. The models like ResNeXt, Dual Path Net, SENet, and NASNet had been identified with the most cuttingedge results for the ImageNet database. These models were examined for the binary, and eight classifications on BreakHis, a breast cancer histopathology image database. Furthermore, the BACH database was used to investigate these models for four classifications. Then, these models were compared with the previous studies to find and propose the most state-of-the-art models for each classification. Since the Inception-ResNet-V2 architecture achieved the best results for binary and eight classifications, we have examined this model in our study as well to provide a better comparison result. In short, this paper provides an extensive evaluation and discussion about the experimental settings for each study that had been conducted on the breast cancer histopathology images. INDEX TERMS Breast cancer, histopathology medical images, deep learning, transfer learning, data augmentation, pre-processing, classification.

show abstract

“…Early genetic algorithm-based methods still require independent training of each network architecture from scratch, so these methods also require high computational costs, such as the search of AmoebaNet [14] which took 3150 GPU days. However, some of the recent methods have achieved state-of-the-art performance with fewer iterations and computation time [27,28].…”

Section: Genetic Algorithmmentioning

confidence: 99%

Genetic Feature Fusion for Object Skeleton Detection

Qiao

Tian

Liu

et al. 2021

Security and Communication Networks

View full text Add to dashboard Cite

Object skeleton detection requires the convolutional neural networks to recognize objects and their parts in the cluttered background, overcome the image definition degradation brought by the pooling layers, and predict the location of skeleton pixels in different scale granularity. Most existing object skeleton detection methods take great efforts into the designing of side-output networks for multiscale feature fusion. Despite the great progress achieved by them, there are still many problems that hinder the development of object skeleton detection, such as the manually designed network is labor-intensive and the network initialization depends on models pretrained on large-scale datasets. To alleviate these issues, we propose a genetic NAS method to automatically search on a newly designed architecture search space for adaptive multiscale feature fusion. Furthermore, we introduce a symmetric encoder-decoder search space based on reversing the VGG network, in which the decoder can reuse the ImageNet pretrained model of VGG. The searched networks improve the performance of the state-of-the-art methods on commonly used skeleton detection benchmarks, which proves the efficacy of our method.

show abstract

RENAS: Reinforced Evolutionary Neural Architecture Search

Cited by 117 publications

References 24 publications

Towards Automated Neural Interaction Discovery for Click-Through Rate Prediction

Towards Automated Neural Interaction Discovery for Click-Through Rate Prediction

Breast Cancer Classification Using Deep Learning Approaches and Histopathology Image: A Comparison Study

Genetic Feature Fusion for Object Skeleton Detection

Contact Info

Product

Resources

About