Looped Transformers as Optimizers

arXiv:2609.37379 · cs.LG · Submitted 2026-09-29 · Read on arXiv

cs.LG

Submitted: 2026-09-29

Updated: 2026-09-29

Terminology

• Transformer — 1× in this paper · explained in 2 episodes

Sources

• PIQA: Reasoning about Physical Commonsense in Natural Language
• Training-Free Looped Transformers
• Evaluating Large Language Models Trained on Code
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• Training Verifiers to Solve Math Word Problems
• MoEUT: Mixture-of-Experts Universal Transformers
• Solve the Loop: Attractor Models for Language and Reasoning
• Loop the Loopies!
• Can Looped Transformers Learn to Implement Multi-step Gradient Descent for In-context Learning?
• Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
• Adaptive Computation Time for Recurrent Neural Networks
• Efficiently Modeling Long Sequences with Structured State Spaces
• Gated Recurrent Transformers: Expressive Depth through Recurrent Modulation
• Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters
• Scaling Laws for Neural Language Models
• Kimi Linear: An Expressive, Efficient Attention Architecture
• Large Language Models are Zero-Shot Reasoners
• Muon is Scalable for LLM Training
• Pointer Sentinel Mixture Models
• Recirculation

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks