Lin Li
Lin Li has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Jianing Qiu
The Chinese University of Hong Kong
6 shared publications - Michael William Spratling
King’s College London
4 shared publications - Benny Lo
Hamlyn Centre, Department of Surgery and Cancer, Imperial College London, London SW7 2AZ, UK
3 shared publications - Kyle Lam
Department of Surgery and Cancer, Imperial College, London, UK
3 shared publications - Wu Yuan
Department of Biomedical Engineering, The Chinese University of Hong Kong, Hong Kong SAR, China
3 shared publications - Bo Xiao
Hamlyn Centre for Robotic Surgery, Imperial College London, London, U.K
2 shared publications
Latest Publications
- Open Access
- 27 February 2025
- Published by Elsevier in Pattern Recognition
- Vol. 163
Robust shortcut and disordered robustness: Improving adversarial training through adaptive smoothing
- 27 February 2025
- Published by Elsevier in Pattern Recognition
- Vol. 163
Open AccessRobust shortcut and disordered robustness: Improving adversarial training through adaptive smoothing
- Top Cited Papers
- 27 November 2024
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 1 (12)
Development and Validation of a Multimodal Multitask Vision Foundation Model for Generalist Ophthalmic Artificial Intelligence
- 27 November 2024
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 1 (12)
Top Cited PapersDevelopment and Validation of a Multimodal Multitask Vision Foundation Model for Generalist Ophthalmic Artificial Intelligence
- 23 September 2024
- Published by SAGE Publications in The International Journal of Robotics Research
- Vol. 44 (4) , 513-531
Advancing robots with greater dynamic dexterity: A large-scale multi-view and multi-modal dataset of human-human throw&catch of arbitrary objects
Learning and imitating behavioral intelligence from human demonstrations is a promising approach towards the intuitive programming of robots for enhanced dynamic dexterity. However, there has been no publicly available dataset in this domain. To address this gap, we introduce the first large-scale dataset and recording framework specifically designed for studying human collaborative dynamic dexterity in throw&catch tasks. The dataset, named H 2 TC, contains 15,000 multi-view and multi-modal synchronized recordings of diverse Human-Human Throw-and-Catch activities. It involves 34 human subjects with typical motor abilities and a variety of 52 objects frequently manipulated through throw&catch in domestic and/or industrial scenarios. The dataset is supplemented with a hierarchy of manually annotated semantic and dense labels, such as the ground truth human body, hand and object motions captured with specialized high-precision motion tracking systems. These rich annotations make the dataset well-suited for a wide range of robot studies, including both low-level motor skill learning and high-level cognitive planning and recognition. We envision that the proposed dataset and recording framework will facilitate learning pipelines to extract insights on how humans coordinate both intra- and interpersonally to throw and catch objects, ultimately leading to the development of more capable and collaborative robots. The dataset, along with a suite of utility tools, such as those for visualization and annotation, can be accessed from our project page at https://h2tc-roboticsx.github.io/ .
- 23 September 2024
- Published by SAGE Publications in The International Journal of Robotics Research
- Vol. 44 (4) , 513-531
Advancing robots with greater dynamic dexterity: A large-scale multi-view and multi-modal dataset of human-human throw&catch of arbitrary objects
Learning and imitating behavioral intelligence from human demonstrations is a promising approach towards the intuitive programming of robots for enhanced dynamic dexterity. However, there has been no publicly available dataset in this domain. To address this gap, we introduce the first large-scale dataset and recording framework specifically designed for studying human collaborative dynamic dexterity in throw&catch tasks. The dataset, named H 2 TC, contains 15,000 multi-view and multi-modal synchronized recordings of diverse Human-Human Throw-and-Catch activities. It involves 34 human subjects with typical motor abilities and a variety of 52 objects frequently manipulated through throw&catch in domestic and/or industrial scenarios. The dataset is supplemented with a hierarchy of manually annotated semantic and dense labels, such as the ground truth human body, hand and object motions captured with specialized high-precision motion tracking systems. These rich annotations make the dataset well-suited for a wide range of robot studies, including both low-level motor skill learning and high-level cognitive planning and recognition. We envision that the proposed dataset and recording framework will facilitate learning pipelines to extract insights on how humans coordinate both intra- and interpersonally to throw and catch objects, ultimately leading to the development of more capable and collaborative robots. The dataset, along with a suite of utility tools, such as those for visualization and annotation, can be accessed from our project page at https://h2tc-roboticsx.github.io/ .
- Open Access
- 24 August 2024
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 133 (2) , 929-950
AROID: Improving Adversarial Robustness Through Online Instance-Wise Data Augmentation
Deep neural networks are vulnerable to adversarial examples. Adversarial training (AT) is an effective defense against adversarial examples. However, AT is prone to overfitting which degrades robustness substantially. Recently, data augmentation (DA) was shown to be effective in mitigating robust overfitting if appropriately designed and optimized for AT. This work proposes a new method to automatically learn online, instance-wise, DA policies to improve robust generalization for AT. This is the first automated DA method specific for robustness. A novel policy learning objective, consisting of Vulnerability, Affinity and Diversity, is proposed and shown to be sufficiently effective and efficient to be practical for automatic DA generation during AT. Importantly, our method dramatically reduces the cost of policy search from the 5000 h of AutoAugment and the 412 h of IDBH to 9 h, making automated DA more practical to use for adversarial robustness. This allows our method to efficiently explore a large search space for a more effective DA policy and evolve the policy as training progresses. Empirically, our method is shown to outperform all competitive DA methods across various model architectures and datasets. Our DA policy reinforced vanilla AT to surpass several state-of-the-art AT methods regarding both accuracy and robustness. It can also be combined with those advanced AT methods to further boost robustness. Code and pre-trained models are available at: https://github.com/TreeLLi/AROID.
- 24 August 2024
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 133 (2) , 929-950
Open AccessAROID: Improving Adversarial Robustness Through Online Instance-Wise Data Augmentation
Deep neural networks are vulnerable to adversarial examples. Adversarial training (AT) is an effective defense against adversarial examples. However, AT is prone to overfitting which degrades robustness substantially. Recently, data augmentation (DA) was shown to be effective in mitigating robust overfitting if appropriately designed and optimized for AT. This work proposes a new method to automatically learn online, instance-wise, DA policies to improve robust generalization for AT. This is the first automated DA method specific for robustness. A novel policy learning objective, consisting of Vulnerability, Affinity and Diversity, is proposed and shown to be sufficiently effective and efficient to be practical for automatic DA generation during AT. Importantly, our method dramatically reduces the cost of policy search from the 5000 h of AutoAugment and the 412 h of IDBH to 9 h, making automated DA more practical to use for adversarial robustness. This allows our method to efficiently explore a large search space for a more effective DA policy and evolve the policy as training progresses. Empirically, our method is shown to outperform all competitive DA methods across various model architectures and datasets. Our DA policy reinforced vanilla AT to surpass several state-of-the-art AT methods regarding both accuracy and robustness. It can also be combined with those advanced AT methods to further boost robustness. Code and pre-trained models are available at: https://github.com/TreeLLi/AROID.
- Open Access
One Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models
Large pre-trained Vision-Language Models (VLMs) like CLIP, despite having remarkable generalization ability, are highly vulnerable to adversarial examples. This work studies the adversarial robustness of VLMs from the novel perspective of the text prompt instead of the extensively studied model weights (frozen in this work). We first show that the effectiveness of both adversarial attack and defense are sensitive to the used text prompt. Inspired by this, we propose a method to improve resilience to adversarial attacks by learning a robust text prompt for VLMs. The proposed method, named Adversarial Prompt Tuning (APT), is effective while being both computationally and data efficient. Extensive experiments are conducted across 15 datasets and 4 data sparsity schemes (from 1-shot to full training data settings) to show APT's superiority over hand-engineered prompts and other state-of-the-art adaption methods. APT demonstrated excellent abilities in terms of the in-distribution performance and the generalization under input distribution shift and across datasets. Surprisingly, by simply adding one learned word to the prompts, APT can significantly boost the accuracy and robustness (epsilon=4/255) over the hand-engineered prompts by +13% and +8.5% on average respectively. The improvement further increases, in our most effective setting, to +26.4% for accuracy and +16.7% for robustness. Code is available at https://github.com/TreeLLi/APT.
Open AccessOne Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models
Large pre-trained Vision-Language Models (VLMs) like CLIP, despite having remarkable generalization ability, are highly vulnerable to adversarial examples. This work studies the adversarial robustness of VLMs from the novel perspective of the text prompt instead of the extensively studied model weights (frozen in this work). We first show that the effectiveness of both adversarial attack and defense are sensitive to the used text prompt. Inspired by this, we propose a method to improve resilience to adversarial attacks by learning a robust text prompt for VLMs. The proposed method, named Adversarial Prompt Tuning (APT), is effective while being both computationally and data efficient. Extensive experiments are conducted across 15 datasets and 4 data sparsity schemes (from 1-shot to full training data settings) to show APT's superiority over hand-engineered prompts and other state-of-the-art adaption methods. APT demonstrated excellent abilities in terms of the in-distribution performance and the generalization under input distribution shift and across datasets. Surprisingly, by simply adding one learned word to the prompts, APT can significantly boost the accuracy and robustness (epsilon=4/255) over the hand-engineered prompts by +13% and +8.5% on average respectively. The improvement further increases, in our most effective setting, to +26.4% for accuracy and +16.7% for robustness. Code is available at https://github.com/TreeLLi/APT.
- Open Access
Large AI Models in Health Informatics: Applications, Challenges, and the Future
Large AI models, or foundation models, are models recently emerging with massive scales both parameter-wise and data-wise, the magnitudes of which can reach beyond billions. Once pretrained, large AI models demonstrate impressive performance in various downstream tasks. A prime example is ChatGPT, whose capability has compelled people's imagination about the far-reaching influence that large AI models can have and their potential to transform different domains of our lives. In health informatics, the advent of large AI models has brought new paradigms for the design of methodologies. The scale of multi-modal data in the biomedical and health domain has been ever-expanding especially since the community embraced the era of deep learning, which provides the ground to develop, validate, and advance large AI models for breakthroughs in health-related areas. This article presents a comprehensive review of large AI models, from background to their applications. We identify seven key sectors in which large AI models are applicable and might have substantial influence, including 1) bioinformatics; 2) medical diagnosis; 3) medical imaging; 4) medical informatics; 5) medical education; 6) public health; and 7) medical robotics. We examine their challenges, followed by a critical discussion about potential future directions and pitfalls of large AI models in transforming the field of health informatics.
Open AccessLarge AI Models in Health Informatics: Applications, Challenges, and the Future
Large AI models, or foundation models, are models recently emerging with massive scales both parameter-wise and data-wise, the magnitudes of which can reach beyond billions. Once pretrained, large AI models demonstrate impressive performance in various downstream tasks. A prime example is ChatGPT, whose capability has compelled people's imagination about the far-reaching influence that large AI models can have and their potential to transform different domains of our lives. In health informatics, the advent of large AI models has brought new paradigms for the design of methodologies. The scale of multi-modal data in the biomedical and health domain has been ever-expanding especially since the community embraced the era of deep learning, which provides the ground to develop, validate, and advance large AI models for breakthroughs in health-related areas. This article presents a comprehensive review of large AI models, from background to their applications. We identify seven key sectors in which large AI models are applicable and might have substantial influence, including 1) bioinformatics; 2) medical diagnosis; 3) medical imaging; 4) medical informatics; 5) medical education; 6) public health; and 7) medical robotics. We examine their challenges, followed by a critical discussion about potential future directions and pitfalls of large AI models in transforming the field of health informatics.
- Top Cited PapersOpen Access
- 22 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (12) , 6074-6087
Large AI Models in Health Informatics: Applications, Challenges, and the Future
Large AI models, or foundation models, are models recently emerging with massive scales both parameter-wise and data-wise, the magnitudes of which can reach beyond billions. Once pretrained, large AI models demonstrate impressive performance in various downstream tasks. A prime example is ChatGPT, whose capability has compelled people's imagination about the far-reaching influence that large AI models can have and their potential to transform different domains of our lives. In health informatics, the advent of large AI models has brought new paradigms for the design of methodologies. The scale of multi-modal data in the biomedical and health domain has been ever-expanding especially since the community embraced the era of deep learning, which provides the ground to develop, validate, and advance large AI models for breakthroughs in health-related areas. This article presents a comprehensive review of large AI models, from background to their applications. We identify seven key sectors in which large AI models are applicable and might have substantial influence, including 1) bioinformatics; 2) medical diagnosis; 3) medical imaging; 4) medical informatics; 5) medical education; 6) public health; and 7) medical robotics. We examine their challenges, followed by a critical discussion about potential future directions and pitfalls of large AI models in transforming the field of health informatics.
- 22 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (12) , 6074-6087
Top Cited PapersOpen AccessLarge AI Models in Health Informatics: Applications, Challenges, and the Future
Large AI models, or foundation models, are models recently emerging with massive scales both parameter-wise and data-wise, the magnitudes of which can reach beyond billions. Once pretrained, large AI models demonstrate impressive performance in various downstream tasks. A prime example is ChatGPT, whose capability has compelled people's imagination about the far-reaching influence that large AI models can have and their potential to transform different domains of our lives. In health informatics, the advent of large AI models has brought new paradigms for the design of methodologies. The scale of multi-modal data in the biomedical and health domain has been ever-expanding especially since the community embraced the era of deep learning, which provides the ground to develop, validate, and advance large AI models for breakthroughs in health-related areas. This article presents a comprehensive review of large AI models, from background to their applications. We identify seven key sectors in which large AI models are applicable and might have substantial influence, including 1) bioinformatics; 2) medical diagnosis; 3) medical imaging; 4) medical informatics; 5) medical education; 6) public health; and 7) medical robotics. We examine their challenges, followed by a critical discussion about potential future directions and pitfalls of large AI models in transforming the field of health informatics.
- Open Access
Data Augmentation Alone Can Improve Adversarial Training
Adversarial training suffers from the issue of robust overfitting, which seriously impairs its generalization performance. Data augmentation, which is effective at preventing overfitting in standard training, has been observed by many previous works to be ineffective in mitigating overfitting in adversarial training. This work proves that, contrary to previous findings, data augmentation alone can significantly boost accuracy and robustness in adversarial training. We find that the hardness and the diversity of data augmentation are important factors in combating robust overfitting. In general, diversity can improve both accuracy and robustness, while hardness can boost robustness at the cost of accuracy within a certain limit and degrade them both over that limit. To mitigate robust overfitting, we first propose a new crop transformation, Cropshift, which has improved diversity compared to the conventional one (Padcrop). We then propose a new data augmentation scheme, based on Cropshift, with much improved diversity and well-balanced hardness. Empirically, our augmentation method achieves the state-of-the-art accuracy and robustness for data augmentations in adversarial training. Furthermore, when combined with weight averaging it matches, or even exceeds, the performance of the best contemporary regularization methods for alleviating robust overfitting. Code is available at: https://github.com/TreeLLi/DA-Alone-Improves-AT.
Open AccessData Augmentation Alone Can Improve Adversarial Training
Adversarial training suffers from the issue of robust overfitting, which seriously impairs its generalization performance. Data augmentation, which is effective at preventing overfitting in standard training, has been observed by many previous works to be ineffective in mitigating overfitting in adversarial training. This work proves that, contrary to previous findings, data augmentation alone can significantly boost accuracy and robustness in adversarial training. We find that the hardness and the diversity of data augmentation are important factors in combating robust overfitting. In general, diversity can improve both accuracy and robustness, while hardness can boost robustness at the cost of accuracy within a certain limit and degrade them both over that limit. To mitigate robust overfitting, we first propose a new crop transformation, Cropshift, which has improved diversity compared to the conventional one (Padcrop). We then propose a new data augmentation scheme, based on Cropshift, with much improved diversity and well-balanced hardness. Empirically, our augmentation method achieves the state-of-the-art accuracy and robustness for data augmentations in adversarial training. Furthermore, when combined with weight averaging it matches, or even exceeds, the performance of the best contemporary regularization methods for alleviating robust overfitting. Code is available at: https://github.com/TreeLLi/DA-Alone-Improves-AT.
- Open Access
- 1 December 2022
- Published by Elsevier in Pattern Recognition
- Vol. 136
Understanding and combating robust overfitting via input loss landscape analysis and regularization
- 1 December 2022
- Published by Elsevier in Pattern Recognition
- Vol. 136
Open AccessUnderstanding and combating robust overfitting via input loss landscape analysis and regularization