33,628 Hits in 4.9 sec

Learning Compositional Representations for Effective Low-Shot Generalization [article]

Samarth Mishra, Pengkai Zhu, Venkatesh Saligrama
2022 arXiv   pre-print
We find that this type of learning inspired by human cognition can overcome hurdles faced by deep convolutional networks in low-shot generalization tasks, like zero-shot learning, few-shot learning and  ...  Finally, we propose an application of these interpretable encodings in the form of generating synthetic attribute annotations for evaluating zero-shot learning methods on new datasets.  ...  The authors would like to thank Ruizhao Zhu for helpful discussions.  ... 
arXiv:2204.08090v1 fatcat:3dealbb62zdwtnymninemwrfci

Visual Compositional Learning for Human-Object Interaction Detection [article]

Zhi Hou, Xiaojiang Peng, Yu Qiao, Dacheng Tao
2020 arXiv   pre-print
We devise a deep Visual Compositional Learning (VCL) framework, which is a simple yet efficient framework to effectively address this problem.  ...  thus largely alleviates the long-tail distribution problem and benefits low-shot or zero-shot HOI detection.  ...  Low-shot and Zero-shot Learning Our work also ties with low-shot learning [33] within long-tailed distribution [22] and zero-shot learning recognition [35] . Shen et al.  ... 
arXiv:2007.12407v2 fatcat:65aefbhdejfe5lbc433f2kow2m

Intelligent problem-solving as integrated hierarchical reinforcement learning

Manfred Eppe, Christian Gumbsch, Matthias Kerzel, Phuong D. H. Nguyen, Martin V. Butz, Stefan Wermter
2022 Nature Machine Intelligence  
Therefore, we first review the literature in cognitive psychology to highlight the importance of compositional abstraction and predictive processing.  ...  Then we relate the gained insights with contemporary hierarchical reinforcement learning methods.  ...  For example, it can generate "Goldilocks" subgoals that are neither too hard nor too easy to achieve for the low-level layer.  ... 
doi:10.1038/s42256-021-00433-9 fatcat:5rwjbkbqhve53l2dsgr3fuj3aa

Detecting Human-Object Interaction via Fabricated Compositional Learning [article]

Zhi Hou, Baosheng Yu, Yu Qiao, Xiaojiang Peng, Dacheng Tao
2021 arXiv   pre-print
Specifically, we introduce an object fabricator to generate effective object representations, and then combine verbs and fabricated objects to compose new HOI samples.  ...  Inspired by this, we devise a novel HOI compositional learning framework, termed as Fabricated Compositional Learning (FCL), to address the problem of open long-tailed HOI detection.  ...  [57] present a new method for low-shot learning that directly learns to hallucinate examples that are useful for classification.  ... 
arXiv:2103.08214v2 fatcat:x72bszy2jjetpikcxqgxxyvl3q

BatchFormer: Learning to Explore Sample Relationships for Robust Representation Learning [article]

Zhi Hou, Baosheng Yu, Dacheng Tao
2022 arXiv   pre-print
zero-shot learning, domain generalization, and contrastive learning.  ...  Despite the success of deep neural networks, there are still many challenges in deep representation learning due to the data scarcity issues such as data imbalance, unseen distribution, and domain shift  ...  Generalized Zero-Shot Learning We also evaluate BatchFormer on generalized zero-shot learning task.  ... 
arXiv:2203.01522v2 fatcat:2nra2zxp55ch7kizekdrtvtije

Improving Compositional Generalization with Self-Training for Data-to-Text Generation [article]

Sanket Vaibhav Mehta, Jinfeng Rao, Yi Tay, Mihir Kale, Ankur P. Parikh, Emma Strubell
2022 arXiv   pre-print
Such representations are compositional and it is costly to collect responses for all possible combinations of atomic meaning schemata, thereby necessitating few-shot generalization to novel MRs.  ...  In this work, we systematically study the compositional generalization of the state-of-the-art T5 models in few-shot data-to-text tasks.  ...  However, our work focuses on compositional generalization, rather than the pure few-shot learning setup.  ... 
arXiv:2110.08467v2 fatcat:vpczy74ctjcihkj3nsayffym4m

Learning Compositional Representations for Few-Shot Recognition [article]

Pavel Tokmakov, Yu-Xiong Wang, Martial Hebert
2019 arXiv   pre-print
We demonstrate the value of compositional representations on three datasets: CUB-200-2011, SUN397, and ImageNet, and show that they require fewer examples to learn classifiers for novel categories.  ...  Instrumental to this rapid learning ability is the compositional structure of concept representations in the human brain --- something that deep learning models are lacking.  ...  The effect of compositional representations for Matching Networks is more pronounced, allowing them to outperform the linear classifier in the 1-shot evaluation setting.  ... 
arXiv:1812.09213v3 fatcat:dtoimu7qy5bgjp6ppfndsder4i

Using Shape to Categorize: Low-Shot Learning with an Explicit Shape Bias [article]

Stefan Stojanov, Anh Thai, James M. Rehg
2021 arXiv   pre-print
about 3D shape can be used to improve low-shot learning methods' generalization performance.  ...  Our new approach improves the performance of image-only low-shot learning approaches on multiple datasets.  ...  41, 50, 16, 35, 56, 46, 9] , the question of how shape cues could be used to learn effective representations for image-based low-shot categorization has not been investigated previously.  ... 
arXiv:2101.07296v2 fatcat:qbam6yf7uvdtxaqkl6ibihnbvi

Variational Autoencoder with Disentanglement Priors for Low-Resource Task-Specific Natural Language Generation [article]

Zhuang Li, Lizhen Qu, Qiongkai Xu, Tongtong Wu, Tianyang Zhan, Gholamreza Haffari
2022 arXiv   pre-print
In order to improve compositional generalization, our model performs disentangled representation learning by introducing a prior for the latent content space and another prior for the latent label space  ...  We can also sample diverse content representations from the content space without accessing data of the seen tasks, and fuse them with the representations of novel tasks for generating diverse texts in  ...  Methodology To Disentangled representation learning for compositional generalization in low resource NLG tasks poses three major challenges.  ... 
arXiv:2202.13363v2 fatcat:e6tjmp4hcjbpxdl2qrsytvthjq

Learning to Generate Novel Scene Compositions from Single Images and Videos [article]

Vadim Sushko, Juergen Gall, Anna Khoreva
2021 arXiv   pre-print
In this work, we introduce One-Shot GAN that can learn to generate samples from a training set as little as one image or one video.  ...  Training GANs in low-data regimes remains a challenge, as overfitting often leads to memorization or training divergence.  ...  One-Shot GAN Content-Layout Discriminator. We introduce a solution to overcome the memorization effect but still to generate images of high quality in the one-shot setting.  ... 
arXiv:2105.05847v1 fatcat:4aprmi23mzgpzj2yfmj2jtry6y

Hierarchical principles of embodied reinforcement learning: A review [article]

Manfred Eppe, Christian Gumbsch, Matthias Kerzel, Phuong D.H. Nguyen, Martin V. Butz, Stefan Wermter
2022 arXiv   pre-print
Among the most promising computational approaches to provide comparable learning-based problem-solving abilities for artificial agents and robots is hierarchical reinforcement learning.  ...  There exists pressing evidence that the cognitive mechanisms that enable problem-solving skills in these species build on hierarchical mental representations.  ...  Their method successfully executes compositional policies that it has never executed before, effectively performing zero-shot problem-solving.  ... 
arXiv:2012.10147v2 fatcat:u46d7xx65fhfpoixpgta2j443e

Learning Compositional Shape Priors for Few-Shot 3D Reconstruction [article]

Mateusz Michalkiewicz, Stavros Tsogkas, Sarah Parisot, Mahsa Baktashmotlagh, Anders Eriksson, Eugene Belilovsky
2021 arXiv   pre-print
In this work we experimentally demonstrate that naive baselines fail in this few-shot learning setting, in which the network must learn informative shape priors for inference of new categories.  ...  examples, calling for a model that can successfully generalize to novel object classes.  ...  However, in the low-shot regime, even the zero-shot variant shows better generalization, outperforming ONN.  ... 
arXiv:2106.06440v2 fatcat:ncnobbj7mndbpdgmuzz6cefumq

Generating Novel Scene Compositions from Single Images and Videos [article]

Vadim Sushko, Dan Zhang, Juergen Gall, Anna Khoreva
2022 arXiv   pre-print
In this work, we introduce SIV-GAN, an unconditional generative model that can generate new scene compositions from a single training image or a single video clip.  ...  We show that SIV-GAN successfully deals with a new challenging task of learning from a single video, for which prior GAN models fail to achieve synthesis of both high quality and diversity.  ...  For a given image x, the purpose of D low-level is to learn low-level features and to produce an image representation F (x) = D low-level (x) for the branches.  ... 
arXiv:2103.13389v3 fatcat:rkkjps4d6jaybn45hnebuu3k5e

Sports Video Analysis: Semantics Extraction, Editorial Content Creation and Adaptation

Changsheng Xu, Jian Cheng, Yi Zhang, Yifan Zhang, Hanqing Lu
2009 Journal of Multimedia  
We first propose a generic multi-layer and multi-modal framework for sports video analysis.  ...  Then we introduce several mid-level audio/visual features which are able to bridge the semantic gap between low-level features and high-level understanding.  ...  Moreover, the supervised learning procedure was constructed on the basis of effective mid-level representations [4] instead of exhaustive low-level features.  ... 
doi:10.4304/jmm.4.2.69-79 fatcat:xytusontr5cyxlxpyqgljnhkqu

Zero-Shot Compositional Concept Learning [article]

Guangyue Xu, Parisa Kordjamshidi, Joyce Y. Chai
2021 arXiv   pre-print
Experiments on two widely-used zero-shot compositional learning (ZSCL) benchmarks have demonstrated the effectiveness of the model compared with recent approaches on both conventional and generalized ZSCL  ...  In this paper, we study the problem of recognizing compositional attribute-object concepts within the zero-shot learning (ZSL) framework.  ...  Related Work Compositional Concept Learning. As a specific zero-shot learning (ZSL) problem, zero-shot compositional learning (ZSCL) tries to learn complex concepts by composing element concepts.  ... 
arXiv:2107.05176v1 fatcat:23b53vxv35eytikftt6vyyokc4
« Previous Showing results 1 — 15 out of 33,628 results