# Towards Compositional Generalization of LLMs via Skill Taxonomy Guided Data Synthesis

arXiv: https://arxiv.org/abs/2601.03676 · HTML: https://mlsanity.com/paper/2601.03676/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Self-evolving LLM agents, one of 13 directions in Synthetic data and self-improvement for LLMs.
- Builds on 14 of its 34 references that are in the map.
- First posted on arXiv 7 Jan 2026.

## Placed in

- Synthetic data and self-improvement for LLMs / [Self-evolving LLM agents](https://mlsanity.com/synth/self-evolving-llm-agents/)

## Builds on (most cited first, 14 of 14)

- [Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning](https://mlsanity.com/paper/2503.09516/) (2025)
- [Faith and Fate: Limits of Transformers on Compositionality](https://mlsanity.com/paper/2305.18654/) (2023)
- [WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild](https://mlsanity.com/paper/2406.04770/) (2024)
- [Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models](https://mlsanity.com/paper/2307.14430/) (2023)
- [A Theory for Emergence of Complex Skills in Language Models](https://mlsanity.com/paper/2307.15936/) (2023)
- [Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task](https://mlsanity.com/paper/2310.09336/) (2023)
- [Skill-Mix: a Flexible and Expandable Family of Evaluations for AI models](https://mlsanity.com/paper/2310.17567/) (2023)
- [Infinity Instruct: Scaling Instruction Selection and Synthesis to Enhance Language Models](https://mlsanity.com/paper/2506.11116/) (2025)
- [BiMix: A Bivariate Data Mixing Law for Language Model Pretraining](https://mlsanity.com/paper/2405.14908/) (2024)
- [Can Models Learn Skill Composition from Examples?](https://mlsanity.com/paper/2409.19808/) (2024)
- [Do Deep Neural Networks Capture Compositionality in Arithmetic Reasoning?](https://mlsanity.com/paper/2302.07866/) (2023)
- [The Information of Large Language Model Geometry](https://mlsanity.com/paper/2402.03471/) (2024)
- [Structural Entropy Guided Agent for Detecting and Repairing Knowledge Deficiencies in LLMs](https://mlsanity.com/paper/2505.07184/) (2025)
- [Scaling Towards the Information Boundary of Instruction Sets: The Infinity Instruct Subject Technical Report](https://mlsanity.com/paper/2507.06968/) (2025)

## Built on by (most cited first, 4 of 4)

- [SKT: Skill-Use Training at Scale via Verified Synthetic Data Generation](https://mlsanity.com/paper/2608.02287/) (2026)
- [Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning](https://mlsanity.com/paper/2608.05139/) (2026)
- [Toward Latent Language Model Skills Steering and Optimization: An Empirical Study](https://mlsanity.com/paper/2608.29459/) (2026)
- [SkillFusion: Towards Reliable Multi-Skill Composition for LLM Agents](https://mlsanity.com/paper/2610.15481/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) and agent-based systems often struggle with compositional generalization due to a data bottleneck in which complex skill combinations follow a long-tailed, power-law distribution, limiting both instruction-following performance and generalization in agent-centric tasks. To address this challenge, we propose STEPS, a Skill Taxonomy guided Entropy-based Post-training data Synthesis framework for generating compositionally challenging data. STEPS explicitly targets compositional generalization by uncovering latent relationships among skills and organizing them into an interpretable, hierarchical skill taxonomy using structural information theory. Building on this taxonomy, we formulate data synthesis as a constrained information maximization problem, selecting skill combinations that maximize marginal structural information within the hierarchy while preserving semantic coherence. Experiments on challenging instruction-following benchmarks show that STEPS outperforms existing data synthesis baselines, while also yielding improved compositional generalization in downstream agent-based evaluations.