🛰️ Daily AI Frontier
‹ back to 2026-09-08

Trace2Tower: Transition-Aware EigenTrace Induction of Multi-Level Skills for LLM Agents

Research LLM Agents

Ranking

Overall 78
Content 95
Popularity 37

Observed public metrics from 1 member.

Merged summary

TL;DR - Trace2Tower distills agent execution traces into a multi-level hierarchy of action templates, routines, and task strategies using transition-aware graph modeling and contrastive spectral decomposition. It improves task success and experience reuse while reducing inefficient or invalid actions.

  • Converts step-level interactions into canonical events linked by semantic compatibility, transition dynamics, and outcome evidence.
  • Extracts stable, success-aligned behavioral modes while suppressing failure-prone shortcuts.
  • Dynamically refines learned skills through verifier-guided feedback.
  • Achieves 87.31% success on ALFWorld with 10.35 steps and 0.26 invalid actions, plus 50.67% exact success on WebShop.

Sources (1)

Trace2Tower: Transition-Aware EigenTrace Induction of Multi-Level Skills for LLM Agents

arXiv cs.AI Jiazheng Sun, Boyu Yang, Binhao Yuan, Mingxuan Li, Xin Peng 2026-09-04 arXiv:2609.05261
Public signals Semantic Scholar citations 0 · Semantic Scholar influential citations 0
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · Citations 0 · Influential citations 0 X · N/A Fetched 2026-09-12 14:15:12.742998 UTC

TL;DR - Trace2Tower distills agent execution traces into a multi-level hierarchy of action templates, routines, and task strategies using transition-aware graph modeling and contrastive spectral decomposition. It improves task success and experience reuse while reducing inefficient or invalid actions.

  • Converts step-level interactions into canonical events linked by semantic compatibility, transition dynamics, and outcome evidence.
  • Extracts stable, success-aligned behavioral modes while suppressing failure-prone shortcuts.
  • Dynamically refines learned skills through verifier-guided feedback.
  • Achieves 87.31% success on ALFWorld with 10.35 steps and 0.26 invalid actions, plus 50.67% exact success on WebShop.
item →