Thousands of Companies Are Driving China's AI Boom. How the Cyberspace Administration of China inadvertently made a guide to the country's homegrown AI revolution. When DeepSeek burst onto the global stage in January 2025, it seemed to appear out of nowhere. But the large language model was just one of the thousands of generative AI tools that have been released in China since 2023--and there's a public archive of every single one of them. Here are 23 ways China is rewiring the future .

Control contraction metrics (CCMs) provide a framework to co-synthesize a controller and a corresponding contraction metric -- a positive-definite Riemannian metric under which a closed-loop system is guaranteed to be incrementally exponentially stable. However, the synthesized controller only ensures that all the trajectories of the system converge to one single trajectory and, as such, does not impose any notion of optimality across an entire trajectory. Furthermore, constructing CCMs requires a known dynamics model and non-trivial effort in solving an infinite-dimensional convex feasibility problem, which limits its scalability to complex systems featuring high dimensionality with uncertainty. To address these issues, we propose to integrate CCMs into reinforcement learning (RL), where CCMs provide dynamics-informed feedback for learning control policies that minimize cumulative tracking error under unknown dynamics. We show that our algorithm, called contraction actor-critic (CAC), formally enhances the capability of CCMs to provide a set of contracting policies with the long-term optimality of RL in a fully automated setting. Given a pre-trained dynamics model, CAC simultaneously learns a contraction metric generator (CMG) -- which generates a contraction metric -- and uses an actor-critic algorithm to learn an optimal tracking policy guided by that metric. We demonstrate the effectiveness of our algorithm relative to established baselines through extensive empirical studies, including simulated and real-world robot experiments, and provide a theoretical rationale for incorporating contraction theory into RL.

LLM-based optimization has shown remarkable potential in enhancing agentic systems. However, the conventional approach of prompting LLM optimizer with the whole training trajectories on training dataset in a single pass becomes untenable as datasets grow, leading to context window overflow and degraded pattern recognition. To address these challenges, we propose Fine-Grained Optimization (FGO), a scalable framework that divides large optimization tasks into manageable subsets, performs targeted optimizations, and systematically combines optimized components through progressive merging. Evaluation across ALFWorld, LogisticsQA, and GAIA benchmarks demonstrate that FGO outperforms existing approaches by 1.6-8.6% while reducing average prompt token consumption by 56.3%. Our framework provides a practical solution for scaling up LLM-based optimization of increasingly sophisticated agent systems. Further analysis demonstrates that FGO achieves the most consistent performance gain in all training dataset sizes, showcasing its scalability and efficiency.

Coronary artery disease (CAD), one of the leading causes of mortality worldwide, necessitates effective risk assessment strategies, with coronary artery calcium (CAC) scoring via computed tomography (CT) being a key method for prevention. Traditional methods, primarily based on UNET architectures implemented on pre-built models, face challenges like the scarcity of annotated CT scans containing CAC and imbalanced datasets, leading to reduced performance in segmentation and scoring tasks. In this study, we address these limitations by incorporating the self-supervised learning (SSL) technique of DINO (self-distillation with no labels), which trains without requiring CAC-specific annotations, enhancing its robustness in generating distinct features. The DINO-LG model, which leverages label guidance to focus on calcified areas, achieves significant improvements, with a sensitivity of 89% and specificity of 90% for detecting CAC-containing CT slices, compared to the standard DINO model's sensitivity of 79% and specificity of 77%. Additionally, false-negative and false-positive rates are reduced by 49% and 59%, respectively, instilling greater confidence in clinicians when ruling out calcification in low-risk patients and minimizing unnecessary imaging reviews by radiologists. Further, CAC scoring and segmentation tasks are conducted using a basic UNET architecture, applied specifically to CT slices identified by the DINO-LG model as containing calcified areas. This targeted approach enhances CAC scoring accuracy by feeding the UNET model with relevant slices, significantly improving diagnostic precision, reducing both false positives and false negatives, and ultimately lowering overall healthcare costs by minimizing unnecessary tests and treatments, presenting a valuable advancement in CAD risk assessment.

Multi-view action clustering leverages the complementary information from different camera views to enhance the clustering performance. Although existing approaches have achieved significant progress, they assume all camera views are available in advance, which is impractical when the camera view is incremental over time. Besides, learning the invariant information among multiple camera views is still a challenging issue, especially in continual learning scenario. Aiming at these problems, we propose a novel continual action clustering (CAC) method, which is capable of learning action categories in a continual learning manner. To be specific, we first devise a category memory library, which captures and stores the learned categories from historical views. Then, as a new camera view arrives, we only need to maintain a consensus partition matrix, which can be updated by leveraging the incoming new camera view rather than keeping all of them. Finally, a three-step alternate optimization is proposed, in which the category memory library and consensus partition matrix are optimized. The empirical experimental results on 6 realistic multi-view action collections demonstrate the excellent clustering performance and time/space efficiency of the CAC compared with 15 state-of-the-art baselines.

Soon after China's artificial intelligence rules came into effect last month, a series of new AI chatbots began trickling onto the market, with government approval. The rules have already been watered down from what was initially proposed, and so far, China hasn't enforced them as strictly as it could, experts say. China's regulatory approach will likely have huge implications for the technological competition between the country and its AI superpower rival the U.S. The Cyberspace Administration of China's (CAC) Generative AI Measures, which came into effect on Aug. 15, are some of the strictest in the world. They state that the generative AI services should not generate content "inciting subversion of national sovereignty or the overturn of the socialist system," or "advocating terrorism or extremism, promoting ethnic hatred and ethnic discrimination, violence and obscenity, as well as fake and harmful information." Preventing AI chatbots from spewing out unwanted or even toxic content has been a challenge for AI developers around the world.

China's cyberspace regulator said it has issued draft rules to oversee the security management of facial recognition technology in the country, following concerns raised in public about the overuse of the technology. The Cyberspace Administration of China (CAC) said on Tuesday that facial recognition technology can only be used to process facial information when there is a specific purpose and sufficient necessity as well as with strict protective measures. The use of the technology will also require an individual's consent, the CAC said in a statement. It added that non-biometric identification solutions should be favoured over facial recognition in cases where such methods are equally effective. Biometric identification, especially facial recognition, has become widespread in China.

China on Tuesday revealed its proposed assessment measures for prospective generative artificial intelligence (AI) tools, telling companies they must submit their products before launching to the public. The Cyberspace Administration of China (CAC) proposed the measures in order to prevent discriminatory content, false information and content with the potential to harm personal privacy or intellectual property, the South China Morning Press reported. Such measures would ensure that the products do not end up suggesting regime subversion or disrupting economic or social order, according to the CAC. A number of Chinese companies, including Baidu, SenseTime and Alibaba, have recently shown of new AI models to power a number of applications from chatbots to image generators, prompting concern from officials over the impending boom in use. The CAC also stressed that the products must align with the country's core socialist values, Reuters reported.

Deep learning algorithms enable computers to perform tasks that seem beyond what we can program them to do using traditional techniques. In recent years, we have seen the emergence of unbeatable computer go players, practical speech recognition systems, and self-driving cars. These algorithms also have applications to image recognition, bioinformatics, and many other domains. Yet, on the theoretical side, we are only starting to understand why deep learning works so well. Recently, Cohen et al. [16] used tensor theory to explain the superiority of deep learning over shallow learning for one specific learning architecture called convolutional arithmetic circuits (CACs). Machine learning algorithms attempt to model abstractions of their input data.

China will begin enforcing its strict new rules around the creation of deepfakes from today. Deepfakes are increasingly being used for manipulation and humiliation. We've seen deepfakes of figures like disgraced FTX founder Sam Bankman-Fried to commit fraud, Ukrainian President Volodymyr Zelenskyy to spread disinformation, and US House Speaker Nancy Pelosi to make her appear drunk. Last month, the Cyberspace Administration of China (CAC) announced rules to clampdown on deepfakes. "In recent years, in-depth synthetic technology has developed rapidly. While serving user needs and improving user experiences, it has also been used by some criminals to produce, copy, publish, and disseminate illegal and bad information, defame, detract from the reputation and honour of others, and counterfeit others," explains the CAC.