Quantifying the Stability of Multi-Step Reasoning via Error Amplification
cs.LG, cs.AI, stat.ML
Submitted: 2026-10-05
Updated: 2026-10-05
Code: https://github.com/VirtuosoResearch/Multi-step-reasoning-experiments
Sources
- Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
- From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step
- CodeMonkeys: Scaling Test-Time Compute for Software Engineering
- LOTION: Smoothing the Optimization Landscape for Quantized Training
- Efficiently Learning Branching Networks for Multitask Algorithmic Reasoning
- A Defense of the Quadratic Model
- Data-Driven Switchback Experiments: Theoretical Tradeoffs and Empirical Bayes Designs
- Unveiling Transformers with LEGO: a synthetic reasoning task
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