diff --git a/content/pages/home.md b/content/pages/home.md index 7680412a..b644e3b7 100644 --- a/content/pages/home.md +++ b/content/pages/home.md @@ -1,38 +1,53 @@ --- title: 对抗评测平台 blocks: - - titleen: Latest updates + - titleen: Latest news viewMoreen: View more updates: - - titleen: Frontier thoughts on AI and scientific insights - subtitleen: AI - contenten: > - Laborious was can, nor some from all teachings because ever example - extremely exercise rationally know, it pain explorer avoids no. - timeen: 2024 Feb 25 | by OpenTAI + - titleen: Introducing VisionSafety Platform + subtitleen: '#Vision' + contenten: >- + As the safety of vision models remains a cornerstone of AI research, + we are proud to launch the VisionSafety Platform. This innovative + platform is designed to evaluate the safety of vision models through + the creation of more powerful, transferable adversarial attacks, + alongside the release of million-scale, real-world adversarial + datasets. This initiative marks a significant step forward in + enhancing the security and resilience of vision-based AI systems. + timeen: 2024 Dec 4 | by Vision Team href: 'https://www.baidu.com' img: /uploads/latestUpdates1.png - - titleen: Frontier thoughts on AI and scientific insights - subtitleen: GPT - contenten: > - Laborious was can, nor some from all teachings because ever example - extremely exercise rationally know, it pain explorer avoids no. - timeen: 2024 Feb 25 | by OpenTAI + - titleen: >- + Launching the Multimodal Safety Research Project: Tackling New Risks + in AI + subtitleen: '#Multimodal' + contenten: >- + The rise of multimodal AI presents significant new risks. In response, + we are launching the Multimodal Safety Research Project, which aims to + drive community-led research on securing multimodal AI systems. This + initiative seeks not only to build safe and secure multimodal models + but also to develop techniques that prevent these systems from being + misused or turning harmful. + timeen: 2024 Dec 17 | by Multimodal Team img: /uploads/latestUpdates2.png - - titleen: Frontier thoughts on AI and scientific insights - subtitleen: ALGORITHMS - contenten: > - Laborious was can, nor some from all teachings because ever example - extremely exercise rationally know, it pain explorer avoids no. - timeen: 2024 Feb 25 | by OpenTAI + - titleen: Do We Truly Understand Large Language Models? + subtitleen: '#Language' + contenten: >- + As LLMs revolutionize technology, a crucial question emerges: Do we + really understand how they work? Often described as sophisticated + next-token predictors, LLMs excel in compressing vast amounts of + information to generate human-like text. But is this mere pattern + matching, or is there a deeper intelligence at play? This intriguing + debate challenges us to explore the true nature of these models. Join + the conversation and share your insights! + timeen: 2024 Dec 17 | by Language Team img: /uploads/latestUpdates3.png _template: updates - - title: Our mission + - title: Our Mission body: > - OpenTAI is an open source platform to support the ever-growing research in - Trustworthy AI, a place where emerging topics can be quickly implemented, - new ideas can be easily tested, and attacks/defenses can be symmetrically - evaluated. + OpenTAI is an open-source platform that drives cutting-edge Trustworthy AI + research and fosters open collaboration to build a secure and equitable AI + future. _template: content - titleen: Research titlezh: Research @@ -97,6 +112,19 @@ blocks: - titleen: Benchmarks titlezh: Benchmarks items: + - benchMarkName: VisionSafety + description: >- + This platform provides datasets, algorithms, and tools needed for + large-scale and transferable adversarial robustness evaluation of + computer vision models. Every vision model deserves thorough and + scalable adversarial evaluations before real-world deployment. + subTitle: 'An Adversarial Evaluation Platform for Computer Vision Models ' + learnMore: Learn More > + benchMarksImg: /uploads/eye-acc.png + tags: + - tagName: vision + - tagName: adversarial + - tagName: million-scale - benchMarkName: Vision Safety description: >- Adversarial attacks and defenses.Adversarial attacks and @@ -104,41 +132,46 @@ blocks: defenses. subTitle: 'world''s #1 benchmark' learnMore: Learn More > - benchMarksImg: /uploads/BenchMarks2.png - tags: - - tagName: LLM1 - - tagName: DEEPFAKE + benchMarksImg: /BenchMarks2.png + tags: [] + - benchMarkName: Vision Safety + description: >- + Adversarial attacks and defenses.Adversarial attacks and + defenses.Adversarial attacks and defenses.Adversarial attacks and + defenses. + subTitle: 'world''s #1 benchmark' + learnMore: Learn More > + benchMarksImg: /BenchMarks2.png + tags: [] _template: benchMarks - titleen: Datasets titlezh: Datasets items: - - datasetsName: 系外行星检测 - desc: >- - But actual has painful explain born and pain no dislikes of - praising.But actual has painful explain born and pain no dislikes of - praising. - subTitle: DEEPFAKE + - datasetsName: CC1M-Adv-C/F + desc: Two million-scale adversarial image datasets. + subTitle: Transfer Attack datasetsBackground: /uploads/datasets2.png - - datasetsName: 系外行星检测 - desc: >- - But actual has painful explain born and pain no dislikes of - praising.But actual has painful explain born and pain no dislikes of - praising. - subTitle: DEEPFAKE + - datasetsName: AdvPatch-1K + desc: 'A adversarial T-shirt dataset of 1,131 images from 20 participants.' + subTitle: Physical Attack datasetsBackground: /uploads/datasets2.png - - datasetsName: 系外行星检测 + - datasetsName: WildDeepfake desc: >- - But actual has painful explain born and pain no dislikes of - praising.But actual has painful explain born and pain no dislikes of - praising. - subTitle: DEEPFAKE + WildDeepfake is a dataset of 7,314 face sequences from 707 deepfake + videos. + subTitle: Deepfake datasetsBackground: /uploads/datasets2.png - - datasetsName: 系外行星检测 - desc: >- - But actual has painful explain born and pain no dislikes of - praising.But actual has painful explain born and pain no dislikes of - praising. - subTitle: DEEPFAKE + - datasetsName: DeepSafe + desc: A safety dataset of 100K questions used by the DeepSafe benchmark. + subTitle: LLM + datasetsBackground: /uploads/datasets2.png + - datasetsName: VLJailbreak + desc: A multimodal jailbreak dataset for VLMs used by the VLJailbreakBench. + subTitle: Multimodal + datasetsBackground: /uploads/datasets2.png + - datasetsName: X-Transfer + desc: A universal adversarial perturbation dataset for vision and VLMs. + subTitle: Multimodal datasetsBackground: /uploads/datasets2.png _template: datasets - titleen: Tools @@ -155,6 +188,11 @@ blocks: - img: /uploads/tag2.png - img: /uploads/tag3.png - img: /uploads/tag4.png + - name: taiadv.vision + description: A Comprehensive Benchmark for Adversarial Attacks on Vision Models + learnMore: Learn More > + img: /uploads/BenchMarks2.png + tagsImage: [] _template: tools - titleen: Partners titlezh: Partners @@ -171,26 +209,13 @@ blocks: - titleen: Contributors titlezh: Contributors items: - - name: Joseph Moore - - name: 张伟 - - name: David Davis - - name: Mark Thompson - - name: 周阳 - - name: Joseph Moore - - name: 张伟 - - name: David Davis - - name: Mark Thompson - - name: 周阳 - - name: Joseph Moore - - name: 张伟 - - name: David Davis - - name: Mark Thompson - - name: 周阳 - - name: Joseph Moore - - name: 张伟 - - name: David Davis - - name: Mark Thompson - - name: 周阳 + - name: Xingjun Ma + - name: Weijie Zheng + - name: Yong Xie + - name: Zhixiang Wang + - name: Hanxun Huang + - name: Bojia Zi + - name: Yugang Jiang _template: contributors --- diff --git a/public/uploads/eye-acc.png b/public/uploads/eye-acc.png new file mode 100644 index 00000000..2c1d4145 Binary files /dev/null and b/public/uploads/eye-acc.png differ