Part of the AI People directory: a structured reference to the individuals shaping artificial intelligence across research, industry, governance, ethics and public discourse.
Geoffrey Hinton, deep learning and AI safety, Canada. The godfather of deep learning, whose persistence through two AI winters kept neural network research alive. His work on backpropagation, deep belief networks and representation learning enabled modern AI. He left Google in 2023 to speak openly about existential risks, and is a co-recipient of the 2024 Nobel Prize in Physics. *Key works: backpropagation papers; deep belief networks; Boltzmann machines.*
Yann LeCun, convolutional networks and computer vision, United States. Invented convolutional neural networks in the 1980s, the architecture that now powers image recognition, medical imaging and autonomous vehicles. Chief AI Scientist at Meta and a vocal advocate for self-supervised learning. Co-recipient of the 2018 Turing Award. *Key works: LeNet; convolutional neural networks for vision.*
Yoshua Bengio, deep learning and AI safety, Canada. Pioneered neural language models and deep learning techniques and co-authored the Deep Learning textbook that trained a generation. He has since pivoted to AI safety research, becoming one of the most prominent technical researchers warning about advanced AI risks. Co-recipient of the 2018 Turing Award. *Key works: Deep Learning (2016); neural language models; safety research.*
Jürgen Schmidhuber, recurrent networks and sequence learning, Switzerland. Co-invented LSTM networks with Sepp Hochreiter, solving the vanishing-gradient problem that had stymied recurrent neural networks. LSTM became the foundation for speech recognition, machine translation and language models until the transformer era. *Key works: the LSTM architecture; recurrent learning.*
Sepp Hochreiter, LSTM and recurrent networks, Austria. Co-invented Long Short-Term Memory networks, enabling neural networks to learn from sequences and remember over long time spans, powering a decade of advances in speech recognition and natural language processing. *Key works: Long Short-Term Memory (1997).*
Ian Goodfellow, generative models and adversarial ML, United States. Invented Generative Adversarial Networks in 2014, creating a new approach for generating realistic images, video and other content, and pioneered adversarial-examples research revealing how easily neural networks can be fooled. *Key works: Generative Adversarial Nets (2014); adversarial ML research.*
Alex Krizhevsky, computer vision and GPU computing, Canada. Created AlexNet with Hinton and Sutskever, winning the 2012 ImageNet competition by a dramatic margin and triggering the deep learning revolution, and demonstrated that GPUs could train neural networks far faster than CPUs. *Key works: ImageNet Classification with Deep Convolutional Neural Networks (2012).*
Ilya Sutskever, deep learning and foundation models, United States. Co-created AlexNet and sequence-to-sequence learning, and as Chief Scientist at OpenAI led the research behind the GPT models. Co-founded Safe Superintelligence Inc. in 2024. One of the most influential figures of the foundation-model era. *Key works: AlexNet; seq2seq; GPT research leadership.*
Kaiming He, computer vision and architecture design, United States. Invented ResNet (residual networks), which solved the degradation problem in very deep networks and enabled training of networks with hundreds of layers, and created Mask R-CNN for instance segmentation. *Key works: ResNet; Mask R-CNN.*
Fei-Fei Li, computer vision and AI ethics, United States. Created ImageNet, the dataset and competition that catalysed the deep learning revolution, and founded Stanford's Human-Centered AI Institute, advocating for AI development that considers human values. *Key works: the ImageNet dataset and challenge; Human-Centered AI leadership.*
Diederik P. Kingma, generative models and optimisation, United States. Co-invented Variational Autoencoders and created the Adam optimiser, now the default for training neural networks. His work on generative models and optimisation is embedded in virtually every modern deep learning system. *Key works: VAE papers; the Adam optimiser.*
Kunihiko Fukushima, neural networks and vision, Japan. Created the neocognitron in 1980, a precursor to convolutional neural networks that introduced hierarchical feature extraction and anticipated modern computer vision architectures by decades. *Key works: the Neocognitron (1980).*
Oriol Vinyals, sequence learning and game AI, United Kingdom. Led the AlphaStar project that reached grandmaster level in StarCraft II, and pioneered sequence-to-sequence learning and attention mechanisms across neural machine translation and game-playing AI. *Key works: seq2seq contributions; AlphaStar; pointer networks.*
Chelsea Finn, meta-learning and robotics, United States. Created Model-Agnostic Meta-Learning (MAML), enabling AI systems to learn new tasks from small amounts of data, addressing one of AI's key limitations: the need for massive training data. *Key works: MAML; robot learning research.*