Diederik P. Kingma

Diederik P. "Durk" Kingma 是一位荷兰计算机科学家和研究员，对现代人工智能产生了根本性的塑造作用。他以变分自编码器（VAE）的首席发明者和 Adam 优化器的共同创造者而闻名，这两项基础技术如今驱动着无数机器学习系统。在谷歌学术上拥有超过 36.2 万次引用，Kingma 是深度学习和生成式 AI 领域最具影响力的研究者之一。

Kingma 的工作弥合了概率建模与神经网络之间的鸿沟，使机器能够生成逼真的图像、文本和视频。他的创新支撑着 Stable Diffusion 和 DALL-E 等工具，其算法已成为行业大规模模型训练的标准。他曾任职于 OpenAI、Google Brain、DeepMind，目前就职于 Anthropic，专注于大规模机器学习。[编辑]

关于 Kingma 早期生活的公开细节有限，因为他保持低调的个人形象。他于 1983 年出生在荷兰，早年就展现出对计算机科学的才能。2009 年和 2012 年，他在纽约大学担任初级研究科学家，师从 Yann LeCun，在深度学习爆发前便接触了该领域。

2010 年至 2012 年，Kingma 共同创立并担任技术主管的初创公司 Advanza 于 2016 年被成功收购。这段创业经历发生在他转向学术之前。2013 年，他在阿姆斯特丹大学师从 Max Welling 攻读博士学位，专注于生成模型。他于 2017 年以优异成绩获得博士学位，其论文提出了变分自编码器。[编辑]

Kingma 的第一个重大突破是 2013 年变分自编码器（VAE）的预印本，发表于 ICLR 2014。VAE 将变分推理与深度神经网络相结合，实现了潜在表示的可扩展学习。该论文已被引用超过 10 万次，成为生成式 AI 的基石。

2014 年，他共同撰写了 Adam 优化器论文，提出了一种自适应学习率方法，成为深度学习领域的默认优化算法。凭借超过 20 万次引用，Adam 几乎被用于所有主要 AI 系统。Kingma 后来还贡献了 Glow（2018），一种生成流模型，以及推动图像和视频生成的变分扩散模型。

Kingma 是 2015 年 OpenAI 创始团队的一员，在 2018 年之前领导算法团队。随后他在 Google Brain 和 DeepMind 工作了六年，研究文本、图像和视频的生成模型。2024 年，他加入 Anthropic 继续大规模机器学习的研究。[编辑]

媒体将 Kingma 描绘为一位低调、才华横溢的学者，其技术贡献悄然推动着现代人工智能革命。他的公众形象是学术性的，而非名人驱动的；他避免社交媒体上的炒作，很少接受主流采访。没有官方粉丝名称，也没有既定的粉丝文化仪式。在机器学习社区中，他被尊为“基础研究员”，其论文是研究生阅读清单的常客。他的网络声音（主要在 Twitter 上）简洁、技术性强，专注于研究清晰度。[编辑]

Diederik P. Kingma holds the distinction of being the 72nd most famous person from Netherlands according to the latest global metrics, a position he maintains with steady consistency. This standing within the iFANN Global Fame Rankings reflects his profound impact on the technical landscape rather than traditional celebrity culture. While his work powers the generative models that dominate modern media, the community surrounding him lacks a formal name or organized rituals. There is no "Durk Army" or fan club; instead, his influence is carried by researchers, engineers, and developers who rely on his algorithms in their daily work. His presence is felt through code repositories and academic citations rather than social media campaigns or public adoration. The individuals who revere his contributions do so through professional adoption, treating his papers as essential reading and his open-source tools as industry standards. Unlike pop stars who have dedicated fanbases creating content for them, Kingma's audience engages with his legacy by building upon his mathematical frameworks. He interacts exclusively through professional channels, sharing research updates and technical insights without seeking personal fame. The absence of a named fandom underscores his role as a foundational architect whose work speaks for itself, leaving no room for the spectacle that defines other types of icons.[编辑]

Diederik P. Kingma uses the Frisian nickname "Durk," pronounced like "Dirk," a nod to his heritage in the Netherlands. Before his rise to prominence in deep learning, he co-founded the startup Advanza, which was acquired in 2016, predating his major breakthroughs. His seminal paper on Variational Autoencoders was first posted on arXiv in December 2013, arriving months before Generative Adversarial Networks were introduced to the world. During his early career, he worked under Yann LeCun at New York University, laying the groundwork for his future success. He also played a key role as a co-founder of OpenAI alongside Elon Musk, Sam Altman, Greg Brockman, and Ilya Sutskever. In the realm of technical lore, the Adam optimizer stands as his most cited achievement, revolutionizing how neural networks are trained by enabling faster convergence. Another legendary moment in his career is the creation of the VAE architecture, which introduced the reparameterization trick and remains central to understanding diffusion models. These technical milestones serve as the stories fans retell when discussing his impact, replacing viral incidents with the quiet revolution of algorithmic innovation.[编辑]

关于 Kingma 的家庭、父母或兄弟姐妹，没有公开信息。他严格保护个人生活隐私。[编辑]

Kingma 的博士导师是 Max Welling，一位著名的机器学习研究员。他在纽约大学师从 Yann LeCun，并与 OpenAI 联合创始人 Ilya Sutskever 和 Greg Brockman 合作。他最常合作的作者包括 Xi Chen、Pieter Abbeel 和 Stefano Ermon。[编辑]

Diederik Kingma没有公开的恋爱关系报道。他对个人生活严格保密，没有可信的媒体报道过他的伴侣、订婚、婚姻或子女。截至2026年，他被报道为单身。[编辑]

Heading into 2026, Diederik P. Kingma's net worth is estimated at $5 Million per iFANN's editorial analysis. The majority of this wealth stems from his salary and equity compensation as a senior researcher at Anthropic, where base salaries for machine learning researchers typically exceed $600,000 plus stock options. Prior to joining Anthropic, he earned competitive compensation during his tenures at Google Brain and DeepMind. Kingma also benefited significantly from the acquisition of his startup Advanza in 2016, though the exact financial details of that deal were not disclosed to the public. He does not engage in endorsements or public business ventures outside his core research roles. His net worth has grown steadily over recent years as he transitioned from OpenAI to Google and then to Anthropic, with equity packages likely increasing at each move.[编辑]

Kingma因其VAE论文获得了ICLR 2014杰出论文奖。他还以最高荣誉（cum laude）获得了博士学位，这是荷兰最高的学术荣誉。[编辑]