Stacked What-Where Auto-encoders

Zhao, Junbo; Mathieu, Michaël; Goroshin, Ross; LeCun, Yann

doi:10.48550/arxiv.1506.02351

Cited by 83 publications

(80 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 4 shows our results on three of these ten folds. The result of SWWAE and CC-GAN are cited from Zhao et al (2015) and Denton et al (2016) respectively. We achieve better results compared to numerous methods.…”

Section: Svhn and Stl-10mentioning

confidence: 99%

RankingMatch: Delving into Semi-Supervised Learning with Consistency Regularization and Ranking Loss

Tran,

Kang,

Kim

2021

Preprint

View full text Add to dashboard Cite

Semi-supervised learning (SSL) has played an important role in leveraging unlabeled data when labeled data is limited. One of the most successful SSL approaches is based on consistency regularization, which encourages the model to produce unchanged with perturbed input. However, there has been less attention spent on inputs that have the same label. Motivated by the observation that the inputs having the same label should have the similar model outputs, we propose a novel method, RankingMatch, that considers not only the perturbed inputs but also the similarity among the inputs having the same label. We especially introduce a new objective function, dubbed BatchMean Triplet loss, which has the advantage of computational efficiency while taking into account all input samples. Our RankingMatch achieves state-of-the-art performance across many standard SSL benchmarks with a variety of labeled data amounts, including 95.13% accuracy on CIFAR-10 with 250 labels, 77.65% accuracy on CIFAR-100 with 10000 labels, 97.76% accuracy on SVHN with 250 labels, and 97.77% accuracy on SVHN with 1000 labels. We also perform an ablation study to prove the efficacy of the proposed BatchMean Triplet loss against existing versions of Triplet loss.

show abstract

Section: Svhn and Stl-10mentioning

confidence: 99%

RankingMatch: Delving into Semi-Supervised Learning with Consistency Regularization and Ranking Loss

Tran,

Kang,

Kim

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The training sets of ImageNet-10 and ImageNet-Dogs [9], which are subsets of ImageNet [31], are used for evaluation. In the manner of the researches [16,21,24] which impose hyperbolic geometry on the activations of neural networks, we used the activations of PICA [23], one of the most recent models devel- [74] 0.274 0.151 0.076 0.111 0.038 0.013 AC [19] 0.242 0.138 0.067 0.139 0.037 0.021 NMF [6] 0.230 0.132 0.065 0.118 0.044 0.016 AE [3] 0.317 0.210 0.152 0.185 0.104 0.073 CAE [37] 0.253 0.134 0.068 0.134 0.059 0.022 SAE [43] 0.325 0.212 0.174 0.183 0.112 0.072 DAE [64] 0.304 0.206 0.138 0.190 0.104 0.078 DCGAN [50] 0.346 0.225 0.157 0.174 0.121 0.078 DeCNN [73] 0.313 0.186 0.142 0.175 0.098 0.073 SWWAE [75] 0.323 0.176 0.160 0.158 0.093 0.076 VAE [26] 0.334 0.193 0.168 0.179 0.107 0.079 JULE [71] 0.300 0.175 0.138 0.138 0.054 0.028 DEC [68] 0.381 0.282 0.203 0.195 0.122 0.079 DAC [9] 0.527 0.394 0.302 0.275 0.219 0.111 DDC [8] 0.577 0.433 0.345 ---DCCM [67] 0 oped for deep image clustering. After obtaining activations from the pre-trained networks of PICA, we built the graph by mutual k nearest neighbors between activations.…”

Section: Image Clusteringmentioning

confidence: 99%

“…After obtaining activations from the pre-trained networks of PICA, we built the graph by mutual k nearest neighbors between activations. Then, both the activations and the graph were used as inputs of HGCAE-P. Extensive baselines and state-of-the-art image clustering methods [35,74,19,6,3,37,43,64,50,73,75,26,71,68,9,8,67,23] were compared. Furthermore, we also trained two auto-encoder models, GAE [27], and hyperbolic auto-encoder (HAE) whose layers are hyperbolic feed-forward layers [16].…”

Section: Image Clusteringmentioning

confidence: 99%

“…Several traditional methods including k-means clustering (Kmeans) [35], spectral clustering (SC) [74], agglomerative clustering (AC) [19], and nonnegative matrix factorization (NMF) [6] were also compared. For the representation-based clustering methods, AE [3], CAE [37], SAE [43], DAE [64], DCGAN [50], DeCNN [73], SWWAE [75], and VAE [26] were adopted. Besides, the state-of-the-art image clustering methods including JULE [71], DEC [68], DAC [9], DDC [8], DCCM [67], and PICA [23] were employed.…”

Section: C2 Image Clusteringmentioning

confidence: 99%

See 1 more Smart Citation

Unsupervised Hyperbolic Representation Learning via Message Passing Auto-Encoders

Park¹,

Cho²,

Chang³

et al. 2021

Preprint

View full text Add to dashboard Cite

Most of the existing literature regarding hyperbolic embedding concentrate upon supervised learning, whereas the use of unsupervised hyperbolic embedding is less well explored. In this paper, we analyze how unsupervised tasks can benefit from learned representations in hyperbolic space. To explore how well the hierarchical structure of unlabeled data can be represented in hyperbolic spaces, we design a novel hyperbolic message passing auto-encoder whose overall auto-encoding is performed in hyperbolic space. The proposed model conducts auto-encoding the networks via fully utilizing hyperbolic geometry in message passing. Through extensive quantitative and qualitative analyses, we validate the properties and benefits of the unsupervised hyperbolic representations. Codes are available at https://github.com/junhocho/HGCAE.

show abstract

“…Recently, RL has been used for different path planning tasks such as navigation [33,34,35,36,37,38,39], localization [40,41], and mapping [42,43]. Rl has also been applied for coverage in different contexts.…”

Section: Introductionmentioning

confidence: 99%

Reinforcement Learning-Based Coverage Path Planning with Implicit Cellular Decomposition

Heydari,

Saha,

Ganapathy

2021

Preprint

View full text Add to dashboard Cite

Coverage path planning in a generic known environment is shown to be NP-hard. When the environment is unknown, it becomes more challenging as the robot is required to rely on its online map information built during coverage for planning its path. A significant research effort focuses on designing heuristic or approximate algorithms that achieve reasonable performance. Such algorithms have sub-optimal performance in terms of covering the area or the cost of coverage, e.g., coverage time or energy consumption. In this paper, we provide a systematic analysis of the coverage problem and formulate it as an optimal stopping time problem, where the trade-off between coverage performance and its cost is explicitly accounted for. Next, we demonstrate that reinforcement learning (RL) techniques can be leveraged to solve the problem computationally. To this end, we provide some technical and practical considerations to facilitate the application of the RL algorithms and improve the efficiency of the solutions. Finally, through experiments in grid world environments and Gazebo simulator, we show that reinforcement learning-based algorithms efficiently cover realistic unknown indoor environments, and outperform the current state of the art.

show abstract

Stacked What-Where Auto-encoders

Cited by 83 publications

References 18 publications

RankingMatch: Delving into Semi-Supervised Learning with Consistency Regularization and Ranking Loss

RankingMatch: Delving into Semi-Supervised Learning with Consistency Regularization and Ranking Loss

Unsupervised Hyperbolic Representation Learning via Message Passing Auto-Encoders

Reinforcement Learning-Based Coverage Path Planning with Implicit Cellular Decomposition

Contact Info

Product

Resources

About