Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Vilas, Martina G., Yousefi, Safoora, Nushi, Besmira, Horvitz, Eric, Balachandran, Vidhisha
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909839942221824
author Vilas, Martina G.
Yousefi, Safoora
Nushi, Besmira
Horvitz, Eric
Balachandran, Vidhisha
author_facet Vilas, Martina G.
Yousefi, Safoora
Nushi, Besmira
Horvitz, Eric
Balachandran, Vidhisha
contents Reasoning models improve their problem-solving ability through inference-time scaling, allocating more compute via longer token budgets. Identifying which reasoning traces are likely to succeed remains a key opportunity: reliably predicting productive paths can substantially reduce wasted computation and improve overall efficiency. We introduce Latent-Trajectory signals that characterize the temporal evolution of a model's internal representations during the generation of intermediate reasoning tokens. By measuring the overall change in latent representations between the start and end of reasoning, the change accumulated across intermediate steps, and the extent to which these changes advance toward the final state, we show that these signals predict solution accuracy more reliably than both cross-layer metrics and output-based confidence measures. When used to guide answer selection across multiple sampled generations, Latent-Trajectory signals make test-time scaling more effective and efficient than majority voting, reducing token usage by up to 70% while preserving and even improving accuracy by 2.6% on average. Moreover, these predictive signals often emerge early in the reasoning trace, enabling early selection and allocation of compute to the most promising candidates. Our findings contribute not only practical strategies for inference-time efficiency, but also a deeper interpretability perspective on how reasoning processes are represented and differentiated in latent space.
format Preprint
id arxiv_https___arxiv_org_abs_2510_10494
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning
Vilas, Martina G.
Yousefi, Safoora
Nushi, Besmira
Horvitz, Eric
Balachandran, Vidhisha
Artificial Intelligence
Reasoning models improve their problem-solving ability through inference-time scaling, allocating more compute via longer token budgets. Identifying which reasoning traces are likely to succeed remains a key opportunity: reliably predicting productive paths can substantially reduce wasted computation and improve overall efficiency. We introduce Latent-Trajectory signals that characterize the temporal evolution of a model's internal representations during the generation of intermediate reasoning tokens. By measuring the overall change in latent representations between the start and end of reasoning, the change accumulated across intermediate steps, and the extent to which these changes advance toward the final state, we show that these signals predict solution accuracy more reliably than both cross-layer metrics and output-based confidence measures. When used to guide answer selection across multiple sampled generations, Latent-Trajectory signals make test-time scaling more effective and efficient than majority voting, reducing token usage by up to 70% while preserving and even improving accuracy by 2.6% on average. Moreover, these predictive signals often emerge early in the reasoning trace, enabling early selection and allocation of compute to the most promising candidates. Our findings contribute not only practical strategies for inference-time efficiency, but also a deeper interpretability perspective on how reasoning processes are represented and differentiated in latent space.
title Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning
topic Artificial Intelligence
url https://arxiv.org/abs/2510.10494