GTS: Inference-Time Scaling of Latent Reasoning with a Learnable Gaussian Thought Sampler
This work addresses a domain-specific bottleneck in latent reasoning models, offering an incremental improvement over existing methods.
The paper tackled the problem of inefficient exploration in inference-time scaling for latent reasoning models by proposing a Gaussian Thought Sampler (GTS) that learns context-dependent perturbation distributions, achieving more reliable scaling than heuristic baselines on GSM8K.
Inference-time scaling (ITS) in latent reasoning models typically introduces stochasticity through heuristic perturbations, such as dropout or fixed Gaussian noise. While these methods increase trajectory diversity, their exploration behavior is not explicitly modeled and can be inefficient under finite sampling budgets. We observe that stronger perturbations do not necessarily translate into more effective candidate trajectories, as unguided noise may disrupt internal decision structure rather than steer it. To provide a more structured alternative, we model latent thought exploration as conditional sampling from learnable densities and instantiate this idea as a Gaussian Thought Sampler (GTS). GTS predicts context-dependent perturbation distributions over continuous reasoning states and is trained with GRPO-style policy optimization while keeping the backbone frozen. Experiments on GSM8K with two latent reasoning architectures show that GTS achieves more reliable inference-time scaling than heuristic baselines. These findings indicate that improving latent ITS requires structured and optimizable exploration mechanisms rather than simply amplifying stochasticity.