Model improves math problem solving with activation-based learning
Activation-Conditioned Self-Distillation
Machine Learning
Summary
Solving math problems by machines can be difficult when they need to check long answers step by step. The authors present a new method where the model learns from itself by comparing how well it does on successful attempts versus unsuccessful ones. Using these comparisons, it creates guidance to improve future answers without needing extra help or fixed solutions. This approach improved math problem-solving accuracy across several tested models.
What this means in practice
- •For ai model trainers: Train language models to solve math problems more accurately using activation-conditioned guidance without needing external references.
- •For software developers: Improve the reliability of AI-based coding assistants or calculators by integrating models distilled with activation-conditioned self-distillation.
Authors
Zhexi Lu, Subhajit Chaudhury, Tejaswini Pedapati, Keerthiram Murugesan, Lei Yu
Abstract
On-policy self-distillation uses a model as its own teacher to provide dense supervision for reasoning, often through reference-solution conditioning. Providing privileged information does not by itself ensure effective token-level supervision throughout long responses. We introduce Activation-Conditioned Self-Distillation (ACSD), which extracts a steering vector by contrasting activations of self-generated trajectories that reach verified correct answers within a generation budget with those of all remaining trajectories. A frozen copy of the base model applies this vector at each prediction position, and the student learns from its next-token distributions on student-generated prefixes. Outcome verification is used for direction construction and calibration; distillation requires neither problem-specific reference text nor teacher parameter updates. The distilled student is used alone at inference. On each of five models, ACSD achieves the highest mean accuracy over four mathematical benchmarks among the evaluated methods. On DeepSeek-R1-0528-Qwen3-8B, mean mathematical accuracy reaches 71.9\% and LiveCodeBench v6 pass@12 reaches 70.9\%, compared with 69.0\% and 66.3\% for the reference-conditioned OPSD baseline. Contrasts among correct trajectories also support distillation, and extracted directions can be reused across mathematical training datasets. On fixed student trajectories, ACSD maintains more stable late-position logit-update magnitudes than OPSD.