Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data
cs.LG, cs.AI
Submitted: 2026-01-21
Updated: 2026-11-06
Terminology
Sources
- Generalization on the Unseen, Logic Reasoning and Degree Curriculum
- Can LLMs perform structured graph reasoning?
- What Makes Data Suitable for a Locally Connected Neural Network? A Necessary and Sufficient Condition Based on Quantum Entanglement
- The pitfalls of next-token prediction
- Facts in Stats: Impacts of Pretraining Diversity on Language Model Generalization
- Bypassing the Exponential Dependency: Looped Transformers Efficiently Learn In-context by Multi-step Gradient Descent
- Evaluating Large Language Models Trained on Code
- A Simple Framework for Contrastive Learning of Visual Representations
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback
- Continuous vs. Discrete Optimization of Deep Neural Networks
- Revisiting Fundamentals of Experience Replay
- Does Learning Require Memorization? A Short Tale about a Long Tail
- Looped Transformers as Programmable Computers
- The Llama 3 Herd of Models
- Automated Curriculum Learning for Neural Networks
- How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations
- Transformers Learn to Implement Multi-step Gradient Descent with Chain of Thought
- Transformers Provably Learn Chain-of-Thought Reasoning with Length Generalization
- On the Emergence of Implicit Curriculum in RLVR Learning Dynamics
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis