From Pareto to Preference: Personalized Test-Time Scaling via Amortized Agentic Policy Discovery

Key points

• Test-time scaling (TTS) improves the reasoning capabilities of large language models by allocating additional inference computation.1
• Existing approaches to improving TTS efficiency largely optimize accuracy against one resource dimension at a time, advancing either the accuracy--cost or accuracy--latency Pareto frontier.1
• We formulate Personalized Test-Time Scaling as discovering executable controllers that maximize the joint satisfaction rate of user-specific requirements.1
• Experiments on AIME and HMMT show that PersonTTS substantially outperforms strong TTS baselines in joint requirement satisfaction on unseen user profiles and held-out problems.1

Sources (1)

• [1]From Pareto to Preference: Personalized Test-Time Scaling via Amortized Agentic Policy Discovery
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 08:46 AM

  To reduce the overhead of repeated policy discovery for new user profiles, we propose PersonTTS, an amortized agentic policy-discovery framework that reuses prior search experience through requirement-matched controller initialization and source-distilled procedural guidance, while retaining target-profile evaluation for every candidate.

  Test-time scaling (TTS) improves the reasoning capabilities of large language models by allocating additional inference computation.

Extractive summary: sentences quoted from the sources.