Test-time Diverse Reasoning by Riemannian Activation Steering

• URL: http://arxiv.org/abs/2511.08305v1
• Date: Wed, 12 Nov 2025 01:51:59 GMT
• Title: Test-time Diverse Reasoning by Riemannian Activation Steering
• Authors: Ly Tran Ho Khanh, Dongxuan Zhu, Man-Chung Yue, Viet Anh Nguyen,
• Abstract summary: Best-of-$N$ reasoning improves the accuracy of language models in solving complex tasks by sampling multiple candidate solutions and then selecting the best one based on some criteria.<n>A critical bottleneck for this strategy is the output limit diversity, which occurs when the model generates similar outputs despite sampling, and hence recites the same error.<n>We propose a novel strategy that simultaneously optimize the steering vectors for multiple reasoning trajectories at test time.
• Score: 16.26456436031057
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Best-of-$N$ reasoning improves the accuracy of language models in solving complex tasks by sampling multiple candidate solutions and then selecting the best one based on some criteria. A critical bottleneck for this strategy is the output diversity limit, which occurs when the model generates similar outputs despite stochastic sampling, and hence recites the same error. To address this lack of variance in reasoning paths, we propose a novel unsupervised activation steering strategy that simultaneously optimizes the steering vectors for multiple reasoning trajectories at test time. At any synchronization anchor along the batch generation process, we find the steering vectors that maximize the total volume spanned by all possible intervened activation subsets. We demonstrate that these steering vectors can be determined by solving a Riemannian optimization problem over the product of spheres with a log-determinant objective function. We then use a Riemannian block-coordinate descent algorithm with a well-tuned learning rate to obtain a stationary point of the problem, and we apply these steering vectors until the generation process reaches the subsequent synchronization anchor. Empirical evaluations on popular mathematical benchmarks demonstrate that our test-time Riemannian activation steering strategy outperforms vanilla sampling techniques in terms of generative diversity and solution accuracy.

Related papers

• What If We Allocate Test-Time Compute Adaptively? [2.1713977971908944]
  Test-time scaling allocates inference computation uniformly, uses fixed sampling strategies, and applies verification only for reranking.<n>We propose a verifier-guided adaptive framework treating reasoning as iterative trajectory generation and selection.<n>Across datasets, our dynamic, PRM-guided approach consistently outperforms direct test-time scaling.
  arXiv  Detail & Related papers  (2026-02-01T07:30:22Z)
• Exploring Diverse Generation Paths via Inference-time Stiefel Activation Steering [15.160204778653302]
  Language models often default to a narrow set of high-probability outputs, leaving their generation paths homogeneous and prone to mode collapse.<n>We introduce STARS, a training-free, inference-time intervention method that transforms activation steering into an exploration engine.<n>For test case generation and scientific discovery benchmarks, STARS consistently outperforms standard sampling methods.
  arXiv  Detail & Related papers  (2026-01-29T17:17:04Z)
• Differentiable Knapsack and Top-k Operators via Dynamic Programming [10.762219154434709]
  Knapsack and Top-k operators are useful for selecting discrete subsets of variables.<n>We propose a unified framework casting these operators as dynamic programs, and derive differentiable relaxations.<n>On the algorithmic side, we develop efficient algorithms supporting both deterministic and forward passes.<n>On the theoretical side, we prove that Shannon entropy is the unique regularization choice yielding permutation-equivariant operators.
  arXiv  Detail & Related papers  (2026-01-29T14:25:35Z)
• Fractional Reasoning via Latent Steering Vectors Improves Inference Time Compute [60.151643048803145]
  We propose Fractional Reasoning, a framework that enables continuous control over reasoning intensity at inference time.<n>Our method operates by extracting the latent steering vector associated with deeper reasoning and reapplying it with a tunable scaling factor.<n> Experiments on GSM8K, MATH500, and GPQA demonstrate that Fractional Reasoning consistently improves performance across diverse reasoning tasks and models.
  arXiv  Detail & Related papers  (2025-06-18T21:15:59Z)
• Vector Optimization with Gaussian Process Bandits [7.049738935364297]
  Learning problems in which multiple objectives must be considered simultaneously often arise in various fields, including engineering, drug design, and environmental management.<n>Traditional methods for dealing with multiple black-box objective functions have limitations in incorporating objective preferences and exploring the solution space accordingly.<n>We propose Vector Optimization with Gaussian Process (VOGP), a probably approximately correct adaptive elimination algorithm that performs black-box vector optimization using Gaussian process bandits.
  arXiv  Detail & Related papers  (2024-12-03T14:47:46Z)
• Truncating Trajectories in Monte Carlo Policy Evaluation: an Adaptive Approach [51.76826149868971]
  Policy evaluation via Monte Carlo simulation is at the core of many MC Reinforcement Learning (RL) algorithms. We propose as a quality index a surrogate of the mean squared error of a return estimator that uses trajectories of different lengths. We present an adaptive algorithm called Robust and Iterative Data collection strategy Optimization (RIDO)
  arXiv  Detail & Related papers  (2024-10-17T11:47:56Z)
• Harmonic Path Integral Diffusion [0.4527270266697462]
  We present a novel approach for sampling from a continuous multivariate probability distribution, which may either be explicitly known (up to a normalization factor) or represented via empirical samples. Our method constructs a time-dependent bridge from a delta function centered at the origin of the state space at $t=0$, transforming it into the target distribution at $t=1$. We contrast these algorithms with other sampling methods, particularly simulated and path integral sampling, highlighting their advantages in terms of analytical control, accuracy, and computational efficiency.
  arXiv  Detail & Related papers  (2024-09-23T16:20:21Z)
• Exploring the Algorithm-Dependent Generalization of AUPRC Optimization with List Stability [107.65337427333064]
  optimization of the Area Under the Precision-Recall Curve (AUPRC) is a crucial problem for machine learning. In this work, we present the first trial in the single-dependent generalization of AUPRC optimization. Experiments on three image retrieval datasets on speak to the effectiveness and soundness of our framework.
  arXiv  Detail & Related papers  (2022-09-27T09:06:37Z)
• A Stochastic Newton Algorithm for Distributed Convex Optimization [62.20732134991661]
  We analyze a Newton algorithm for homogeneous distributed convex optimization, where each machine can calculate gradients of the same population objective. We show that our method can reduce the number, and frequency, of required communication rounds compared to existing methods without hurting performance.
  arXiv  Detail & Related papers  (2021-10-07T17:51:10Z)
• Adaptive Sampling Quasi-Newton Methods for Zeroth-Order Stochastic Optimization [1.7513645771137178]
  We consider unconstrained optimization problems with no available gradient information. We propose an adaptive sampling quasi-Newton method where we estimate the gradients of a simulation function using finite differences within a common random number framework. We develop modified versions of a norm test and an inner product quasi-Newton test to control the sample sizes used in the approximations and provide global convergence results to the neighborhood of the optimal solution.
  arXiv  Detail & Related papers  (2021-09-24T21:49:25Z)
• Navigating to the Best Policy in Markov Decision Processes [68.8204255655161]
  We investigate the active pure exploration problem in Markov Decision Processes. Agent sequentially selects actions and, from the resulting system trajectory, aims at the best as fast as possible.
  arXiv  Detail & Related papers  (2021-06-05T09:16:28Z)
• Adaptive Sampling for Best Policy Identification in Markov Decision Processes [79.4957965474334]
  We investigate the problem of best-policy identification in discounted Markov Decision (MDPs) when the learner has access to a generative model. The advantages of state-of-the-art algorithms are discussed and illustrated.
  arXiv  Detail & Related papers  (2020-09-28T15:22:24Z)
• Provably Convergent Working Set Algorithm for Non-Convex Regularized Regression [0.0]
  This paper proposes a working set algorithm for non-regular regularizers with convergence guarantees. Our results demonstrate high gain compared to the full problem solver for both block-coordinates or a gradient solver.
  arXiv  Detail & Related papers  (2020-06-24T07:40:31Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.