Xinshao Wang

AdaTriplet-RA: Domain Matching via Adaptive Triplet and Reinforced Attention for Unsupervised Domain Adaptation

Nov 16, 2022
Xinyao Shu, Shiyang Yan, Zhenyu Lu, Xinshao Wang, Yuan Xie

Figure 2 for AdaTriplet-RA: Domain Matching via Adaptive Triplet and Reinforced Attention for Unsupervised Domain Adaptation

Figure 3 for AdaTriplet-RA: Domain Matching via Adaptive Triplet and Reinforced Attention for Unsupervised Domain Adaptation

Figure 4 for AdaTriplet-RA: Domain Matching via Adaptive Triplet and Reinforced Attention for Unsupervised Domain Adaptation

Access Paper or Ask Questions

Misspecified Phase Retrieval with Generative Priors

Oct 11, 2022
Zhaoqiang Liu, Xinshao Wang, Jiulong Liu

Figure 2 for Misspecified Phase Retrieval with Generative Priors

Figure 3 for Misspecified Phase Retrieval with Generative Priors

Figure 4 for Misspecified Phase Retrieval with Generative Priors

* NeurIPS 2022

Access Paper or Ask Questions

ProSelfLC: Progressive Self Label Correction Towards A Low-Temperature Entropy State

Jun 30, 2022
Xinshao Wang, Yang Hua, Elyor Kodirov, Sankha Subhra Mukherjee, David A. Clifton, Neil M. Robertson

Figure 2 for ProSelfLC: Progressive Self Label Correction Towards A Low-Temperature Entropy State

Figure 3 for ProSelfLC: Progressive Self Label Correction Towards A Low-Temperature Entropy State

Figure 4 for ProSelfLC: Progressive Self Label Correction Towards A Low-Temperature Entropy State

* This work presents a new insightful finding to complement a previous one "deep neural networks easily fit random labels [99]": Deep models fit and generalise significantly less confident when more random labels exist. ProSelfLC redirects and promotes entropy minimisation, which is in marked contrast to recent practices of confidence penalty [16, 59, 72]

Access Paper or Ask Questions

Not All Knowledge Is Created Equal

Jun 02, 2021
Ziyun Li, Xinshao Wang, Haojin Yang, Di Hu, Neil M. Robertson, David A. Clifton, Christoph Meinel

Figure 2 for Not All Knowledge Is Created Equal

Figure 3 for Not All Knowledge Is Created Equal

Figure 4 for Not All Knowledge Is Created Equal

* Selective mutual knowledge distillation

Access Paper or Ask Questions

ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks

Jun 08, 2020
Xinshao Wang, Yang Hua, Elyor Kodirov, Neil M. Robertson

Figure 2 for ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks

Figure 3 for ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks

Figure 4 for ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks

* Open discussion: we show it's fine for a learner (student) to be confident towards a correct low entropy status. Then more research attention should be paid to the definition of correct knowledge, as in genera we accept, human annotations used for learning supervision may be biased, subjective, and wrong

Access Paper or Ask Questions

ProSelfLC: Progressive Self Label Correction for Target Revising in Label Noise

May 17, 2020
Xinshao Wang, Yang Hua, Elyor Kodirov, Neil M. Robertson

Figure 2 for ProSelfLC: Progressive Self Label Correction for Target Revising in Label Noise

Figure 3 for ProSelfLC: Progressive Self Label Correction for Target Revising in Label Noise

Figure 4 for ProSelfLC: Progressive Self Label Correction for Target Revising in Label Noise

* Learning target revising, softer targets, entropy regularisation, EM algorithm. A target label distribution should define both the semantic class and similarity structure!

Access Paper or Ask Questions

Instance Cross Entropy for Deep Metric Learning

Nov 22, 2019
Xinshao Wang, Elyor Kodirov, Yang Hua, Neil Robertson

Figure 2 for Instance Cross Entropy for Deep Metric Learning

Figure 3 for Instance Cross Entropy for Deep Metric Learning

Figure 4 for Instance Cross Entropy for Deep Metric Learning

Access Paper or Ask Questions

ID-aware Quality for Set-based Person Re-identification

Nov 20, 2019
Xinshao Wang, Elyor Kodirov, Yang Hua, Neil M. Robertson

Figure 2 for ID-aware Quality for Set-based Person Re-identification

Figure 3 for ID-aware Quality for Set-based Person Re-identification

Figure 4 for ID-aware Quality for Set-based Person Re-identification

* A Set-based Person Re-identification Baseline: Simple Average Fusion of Global Spatial Representations, without temporal information, without parts/poses/attributes information

Access Paper or Ask Questions

Emphasis Regularisation by Gradient Rescaling for Training Deep Neural Networks with Noisy Labels

May 27, 2019
Xinshao Wang, Yang Hua, Elyor Kodirov, Neil Robertson

Figure 2 for Emphasis Regularisation by Gradient Rescaling for Training Deep Neural Networks with Noisy Labels

Figure 3 for Emphasis Regularisation by Gradient Rescaling for Training Deep Neural Networks with Noisy Labels

Figure 4 for Emphasis Regularisation by Gradient Rescaling for Training Deep Neural Networks with Noisy Labels

* Question: What training examples should be focused and how much more should they be emphasised when training DNNs under label noise? Answer: When noise rate is higher, we can improve a model's robustness by focusing on relatively less difficult examples

Access Paper or Ask Questions

1

2

>>