Announcing the Test of Time Awards from ICLR 2016

We are honored to announce the Test of Time awards for ICLR 2026. This award recognizes papers published ten years ago at ICLR 2016 that have had a lasting impact on the field. The 2026 program chairs reviewed the papers published at ICLR 2016, and selected the two papers below for their profound influence and impact on machine learning today.

Congratulations to the authors of the Test of Time winners!

If you are at ICLR, please come to hear the winners talk on Friday at 5:45 pm (there will be two back-to-back talks).

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Alec Radford, Luke Metz, Soumith Chintala

This paper, colloquially called DCGAN, was one of the first successful demonstrations that learning-based generative models could synthesize diverse, realistic, and complex images. The results kickstarted the subfield of image generation, which is one of the most active areas of machine learning research today and has led to many highly successful applications in industry. While the techniques have changed (from GANs to diffusion models), DCGAN stands the test of time as a key step in the establishment of this important field.

Continuous control with deep reinforcement learning

Timothy P. Lillicrap, Jonathan J. Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, Daan Wierstra

The Test of Time award recognizes works that fundamentally alter the trajectory of their field, and Continuous Control with Deep Reinforcement Learning did exactly that. Before this paper introduced the Deep Deterministic Policy Gradient (DDPG) algorithm, applying reinforcement learning to physical systems was highly bottlenecked. Engineers were trapped manually hand-crafting state features or battling the curse of dimensionality caused by discretizing complex motor controls. DDPG was the first algorithm to successfully address both limitations. By ingeniously combining a deterministic Actor-Critic architecture with the stabilizing techniques of DQN, the algorithm allowed neural networks to translate raw sensor data directly into precise, continuous physical actions. Ultimately, DDPG showed it was possible for deep RL to venture into continuous control, changing the course of the field and sparking an RL revolution.