International Conference on Learning Representations (ICLR) 2026

ICLR 2026 will be held in person from April 23–27 in Rio de Janeiro, Brazil, and Apple is participating again this year as a sponsor. During the exhibition period, Apple will operate Booth #204, open Thursday through Saturday from 9:30 AM to 5:30 PM (BRT).

The presentation schedule is densely packed, centered on posters. On the first day, Pretraining with Hierarchical Memories, The Potential of CoT for Reasoning, Revisiting the Scaling Properties of Downstream Metrics in Large Language Model Training, and Synthetic Bootstrapped Pretraining will be introduced, followed in the afternoon session by research on SelfReflect, Flow Matching with Semidiscrete Couplings, GenCtrl, Scaling Synthetic Task Generation for Agents via Exploration, and Mobile multimodal/vision/robotics/LLM.

On the second day, April 24, topics related to efficient training, reasoning, vision, and safety stand out, including ParaRNN: Unlocking Parallel Training of Nonlinear RNNs for Large Language Models, Compute-Optimal Quantization-Aware Training, Sharp Monocular View Synthesis in Less Than a Second, Semantic Regexes, VLSU, AbstRaL, RL for Reasoning by Adaptively Revealing Rationales, and LaDiR. In the afternoon, presentations continue with Reusing Pre-Training Data at Test Time is a Compute Multiplier, EgoDex, Closing the Gap Between Text and Speech Understanding in LLMs, Adaptive Thinking, and BED-LLM.

April 25 includes Adapting Self-Supervised Representations as a Latent Space for Efficient Generation, MobileCLIP2, Trained on Tokens, Calibrated on Concepts, Learning to Reason as Action Abstractions with Scalable Mid-Training RL, and DiffuCoder. On the same day, To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models will be presented at Oral Session 5C, and ParaRNN will be introduced once again as an Apple Expo Talk.

At the weekend workshops, topics covering data, safety, and adaptive updates continue, including Time Series in the Age of Large Models, Navigating and Addressing Data Problems for Foundation Models, Test-Time Updates, and Interpretability, Robustness, and Safety across Modalities. Overall, this participation brings together Apple's latest research portfolio centered on LLM, reasoning, multimodal, efficiency, and safety.