399 Hits in 4.3 sec

Select, Attend, and Transfer: Light, Learnable Skip Connections [article]

Saeid Asgari Taghanaki, Aicha Bentaieb, Anmol Sharma, S. Kevin Zhou, Yefeng Zheng, Bogdan Georgescu, Puneet Sharma, Sasa Grbic, Zhoubing Xu, Dorin Comaniciu, Ghassan Hamarneh
2018 arXiv   pre-print
We propose light, learnable skip connections which learn to first select the most discriminative channels and then attend to the most discriminative regions of the selected feature maps.  ...  We evaluate the proposed method on three different 2D and volumetric datasets and demonstrate that the proposed light, learnable skip connections can outperform the traditional heavy skip connections in  ...  , that is, only those channels with non-zero weights are selected; 2) Attend: discovering the most salient spatial locations within Figure 1 : We propose light, learnable skip connections that improve  ... 
arXiv:1804.05181v3 fatcat:mkah3fhafbfabnoayqt66cq64i

Conditionally Adaptive Multi-Task Learning: Improving Transfer Learning in NLP Using Fewer Parameters Less Data [article]

Jonathan Pilault, Amine Elhattami, Christopher Pal
2022 arXiv   pre-print
However, MTL must deal with challenges such as: overfitting to low resource tasks, catastrophic forgetting, and negative task transfer, or learning interference.  ...  Multi-Task Learning (MTL) networks have emerged as a promising method for transferring learned knowledge across different tasks.  ...  We would like to thank Colin Raffel, Sandeep Subramanian, and Nicolas Gontier for their useful feedback and the anonymous reviewers for helpful comments, discussions and suggestions.  ... 
arXiv:2009.09139v3 fatcat:3xlrlsexybaipczwkvn4vgjatu

Fine-grained Image-to-Image Transformation towards Visual Recognition [article]

Wei Xiong, Yutong He, Yixuan Zhang, Wenhan Luo, Lin Ma, Jiebo Luo
2020 arXiv   pre-print
In order to preserve the fine-grained contextual details of the input image during the deformable transformation, a constrained nonalignment connection method is proposed to construct learnable highways  ...  Moreover, an adaptive identity modulation mechanism is proposed to transfer the identity information into the output image effectively.  ...  Acknowledgement This work is supported in part by NSF awards #1704337, #1722847, #1813709, and our corporate sponsors.  ... 
arXiv:2001.03856v2 fatcat:3uaudw4jerc67jnsn6h65tux5m

A Theory of Relation Learning and Cross-domain Generalization [article]

Leonidas A. A. Doumas, Guillermo Puebla, Andrea E. Martin, John E. Hummel
2021 arXiv   pre-print
and Pong) and between several psychological tasks.  ...  The model is an extension of the LISA and DORA models of relational inference and learning.  ...  One network was trained using random frame skipping and the other with fixed frame skipping. The inputs to the network were the output of the visual preprocessor described above.  ... 
arXiv:1910.05065v5 fatcat:dak3dffaifbanm2kt5f4nx454u

Focus U-Net: A novel dual attention-gated CNN for polyp segmentation during colonoscopy [article]

Michael Yeung, Evis Sala, Carola-Bibiane Schönlieb, Leonardo Rundo
2021 arXiv   pre-print
The Focus U-Net further incorporates short-range skip connections and deep supervision.  ...  selective learning of polyp features.  ...  The upsampling operation is replaced with a learnable kernel weight using a transposed convolution, but otherwise the skip connection and gating signal are resampled to matching dimensions.  ... 
arXiv:2105.07467v2 fatcat:y2b5j6fgdrhq5iahv7wk4ctjve

A Survey of Transformers [article]

Tianyang Lin, Yuxin Wang, Xiangyang Liu, Xipeng Qiu
2021 arXiv   pre-print
We first briefly introduce the vanilla Transformer and then propose a new taxonomy of X-formers.  ...  X-formers) have been proposed, however, a systematic and comprehensive literature review on these Transformer variants is still missing.  ...  transfer.  ... 
arXiv:2106.04554v2 fatcat:pjctgoqeffhq7ntyw52jqwfzsy

Lightweight End-to-End Speech Enhancement Generative Adversarial Network Using Sinc Convolutions

Lujun Li, Wudamu, Ludwig Kürzinger, Tobias Watzel, Gerhard Rigoll
2021 Applied Sciences  
Second, the system derives a customized filter bank, tuned for the desired application compactly and efficiently.  ...  Moreover, we employ a set of data augmentation techniques in the time domain, which further improve the system performance and its generalization abilities.  ...  Acknowledgments: The authors wish to thank the editors and reviewers for their valuable comments.  ... 
doi:10.3390/app11167564 fatcat:kd2jajp3zzec5ndjl7a6ouxn4q

Tensor-to-Image: Image-to-Image Translation with Vision Transformers [article]

Yiğit Gündüç
2021 arXiv   pre-print
Transformers gain huge attention since they are first introduced and have a wide range of applications.  ...  With the help of self-attention, our model was able to generalize and apply to different problems without a single modification.  ...  U-Net adds skip connections between layers. To formulate these skip connections between layers is i th layer and the (n−i) th layer is directly connected, where n is the layer number.  ... 
arXiv:2110.08037v1 fatcat:5z7g2m5kyfg67g62nlvyrkl6cy

A Survey of Visual Transformers [article]

Yang Liu, Yao Zhang, Yixin Wang, Feng Hou, Jin Yuan, Jiang Tian, Yang Zhang, Zhongchao Shi, Jianping Fan, Zhiqiang He
2022 arXiv   pre-print
Because of their differences on training settings and dedicated vision tasks, we have also evaluated and compared all these existing visual Transformers under different configurations.  ...  segmentation) as well as multiple sensory data stream (images, point clouds, and vision-language data).  ...  A skip-Transformer for adjacent layers further refines the spatial-context features between parents and children to enhance their connection regions.  ... 
arXiv:2111.06091v3 fatcat:a3fq6lvvzzgglb3qtus5qwrwpe

Dynamic Neural Networks: A Survey [article]

Yizeng Han, Gao Huang, Shiji Song, Le Yang, Honghui Wang, Yulin Wang
2021 arXiv   pre-print
3) temporal-wise dynamic models that perform adaptive inference along the temporal dimension for sequential data such as videos and texts.  ...  The important research problems of dynamic networks, e.g., architecture design, decision making scheme, optimization technique and applications, are reviewed systematically.  ...  This scheme is typically implemented on structures with skip connections (e.g.  ... 
arXiv:2102.04906v4 fatcat:zelspxwv6nel7kv2yu6ynakyuu

Vit-GAN: Image-to-image Translation with Vision Transformes and Conditional GANS [article]

Yiğit Gündüç
2021 arXiv   pre-print
We used a unique vision transformers-based generator architecture and Conditional GANs(cGANs) with a Markovian Discriminator (PatchGAN) (  ...  U-Net adds skip connections between layers. if one wants to formulate skip connections between the i th and the (n − i) th layers, where n is the layer number, one need to create a link in between which  ...  The generated images mostly contain two colors, one for the sky and the other for the land. U-Net is an autoencoder with skip connections between encoder and decoder layers.  ... 
arXiv:2110.09305v1 fatcat:shda3rwzizhphki4luyhltwtka

Trading with the Momentum Transformer: An Intelligent and Interpretable Architecture [article]

Kieran Wood, Sven Giegerich, Stephen Roberts, Stefan Zohren
2021 arXiv   pre-print
Through the addition of an interpretable variable selection network, we observe how CPD helps our model to move away from trading predominantly on daily returns data.  ...  The time series is thus segmented into regimes and the model tends to focus on previous time-steps in alike regimes.  ...  ACKNOWLEDGEMENTS We would like to thank the Oxford-Man Institute of Quantitative Finance for financial and computing support.  ... 
arXiv:2112.08534v1 fatcat:2sz62hwvercyfhztbywedp6ppi

ICE-GAN: Identity-aware and Capsule-Enhanced GAN with Graph-based Reasoning for Micro-Expression Recognition and Synthesis [article]

Jianhui Yu, Chaoyi Zhang, Yang Song, Weidong Cai
2021 arXiv   pre-print
detects the image authenticity and expression classes.  ...  The generator produces synthetic faces with controllable micro-expressions and identity-aware features, whose long-ranged dependencies are captured through the graph reasoning module (GRM), and the discriminator  ...  Instead of leveraging skip connections to directly transfer the multi-scale spatial information, we design to firstly flatten the spatial dimension and directly apply self-attention on feature channels  ... 
arXiv:2005.04370v2 fatcat:sx5oxoutbjfdfkgc3wblwqefoe

Class-Aware Generative Adversarial Transformers for Medical Image Segmentation [article]

Chenyu You, Ruihan Zhao, Fenglin Liu, Siyuan Dong, Sandeep Chinchali, Ufuk Topcu, Lawrence Staib, James S. Duncan
2022 arXiv   pre-print
Further qualitative experiments provide a more detailed picture of the model's inner workings, shed light on the challenges in improved transparency, and demonstrate that transfer learning can greatly  ...  and low-level anatomical features.  ...  (Ronneberger et al., 2015) proposed a deep 2D U-Net architecture, combining skip connections between opposing convolution and deconvolution layers to achieve promising performance on a diverse set of  ... 
arXiv:2201.10737v3 fatcat:opygzwskbngtxkxza7ppend4he

Cross-Modal Representation [chapter]

Zhiyuan Liu, Yankai Lin, Maosong Sun
2020 Representation Learning for Natural Language Processing  
After that, we review several real-world applications related to cross-modal representation learning including image captioning, visual relation detection, and visual question answering.  ...  The image and text features pass through two fully connected layers, where single-modal and cross-modal knowledge transfer are performed.  ...  co-attention image Answer:green stop the stop light color stop light lit stop light light ... ... ... ...  ... 
doi:10.1007/978-981-15-5573-2_9 fatcat:duazhghcevejzd27ncvzdpshqq
« Previous Showing results 1 — 15 out of 399 results