Unverified50% confidenceBenchmarkExact time
According to statistics in recent years, PyTorch accounts for more than 60% of paper implementations at conferences such as CVPR, NeurIPS and ICLR, significantly ahead of TensorFlow and JAX.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Related Entities
Related Claims
Unverified顶级 AI 会议论文中使用 PyTorch 的比例已超过70%，PyTorch 在学术研究领域的使用率已大幅超过 TensorFlow67% similarUnverified在从50%到92%的准确率提升中，REPL贡献最大（直接从50%跳至74%），随后领域知识注入、模糊搜索、公式追踪、系统prompt优化和bug修复层层累积63% similarUnverifiedTop main conferences in the machine learning field include NeurIPS, ICML, ICLR, and CVPR, with acceptance rates typically between 15% and 30%.62% similarUnverifiedIn the illustrative sample constructed in the article, a certain tool's first-call success rate was 76% on the claude-desktop client and 21% on cursor, with an average of 62% across all clients.61% similarUnverifiedMedIQ多智能体系统最终达到57%准确率，摘要智能体带来约5%提升，而拥有全部前置信息的Oracle上限为66%61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1086498API
curl https://kongchang.com/api/v1/knowledge/claims/1086498MCP
get_claim(id=1086498)