Form Follows Function: Recursive Stem Model

• URL: http://arxiv.org/abs/2603.15641v1
• Date: Tue, 03 Mar 2026 00:55:00 GMT
• Title: Form Follows Function: Recursive Stem Model
• Abstract summary: We introduce Recursive Stem Model (RSM) to solve compute-heavy and NP puzzles.<n>RSM treats early iterations as detached "warm-up" steps, and applies loss only at the final step.<n>On Sudoku-Extreme, RSM reaches 97.5% exact accuracy with test-time compute.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recursive reasoning models such as Hierarchical Reasoning Model (HRM) and Tiny Recursive Model (TRM) show that small, weight-shared networks can solve compute-heavy and NP puzzles by iteratively refining latent states, but their training typically relies on deep supervision and/or long unrolls that increase wall-clock cost and can bias the model toward greedy intermediate behavior. We introduce Recursive Stem Model (RSM), a recursive reasoning approach that keeps the TRM-style backbone while changing the training contract so the network learns a stable, depth-agnostic transition operator. RSM fully detaches the hidden-state history during training, treats early iterations as detached "warm-up" steps, and applies loss only at the final step. We further grow the outer recursion depth $H$ and inner compute depth $L$ independently and use a stochastic outer-transition scheme (stochastic depth over $H$) to mitigate instability when increasing depth. This yields two key capabilities: (i) $>20\times$ faster training than TRM while improving accuracy ($\approx 5\times$ reduction in error rate), and (ii) test-time scaling where inference can run for arbitrarily many refinement steps ($\sim 20,000 H_{\text{test}} \gg 20 H_{\text{train}}$), enabling additional "thinking" without retraining. On Sudoku-Extreme, RSM reaches 97.5% exact accuracy with test-time compute (within ~1 hour of training on a single A100), and on Maze-Hard ($30 \times 30$) it reaches ~80% exact accuracy in ~40 minutes using attention-based instantiation. Finally, because RSM implements an iterative settling process, convergence behavior provides a simple, architecture-native reliability signal: non-settling trajectories warn that the model has not reached a viable solution and can be a guard against hallucination, while stable fixed points can be paired with domain verifiers for practical correctness checks.

Related papers

• HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation [56.14416807766864]
  Homomorphic encryption allows a server to run a language model directly on encrypted user prompts.<n>Existing approaches fix the iteration counts uniformly across the model rather than tailoring them to each site's error tolerance.<n>We introduce Homomorphic Encryption-Aware Training (HEAT), a fine-tuning method that makes the per-nonline iteration counts learnable.
  arXiv  Detail & Related papers  (2026-09-01T18:02:02Z)
• ThinkRetrieve: Retrieval-Augmented Reasoning Traces for Test-Time Scaling [60.48279351544583]
  ThinkRetrieve is a test-time scaling framework for large reasoning models.<n>It retrieves relevant exemplars at each intermediate step and injects them directly into the thinking trace.<n>It consistently improves accuracy over standard test-time scaling.
  arXiv  Detail & Related papers  (2026-08-11T13:58:07Z)
• Depth-Entropy Guided Sampling for Training-Free LLM Reasoning [7.057431471793534]
  Reinforcement learning (RL) has become the dominant paradigm for improving the reasoning capabilities of large language models.<n>We introduce Depth-Entropy Guided Sampling (DEGS), a training-free, test-time method that exploits layer-wise entropy collapse as an intrinsic quality signal.
  arXiv  Detail & Related papers  (2026-06-19T12:36:23Z)
• $\textbf{Re}^{2}$: Unlocking LLM Reasoning via Reinforcement Learning with Re-solving [75.29519604607111]
  Reinforcement learning with verifiable rewards (RLVR) has shown promise in enhancing the reasoning performance of large language models (LLMs)<n>We show that when the initial direction or quality of the chain-of-thought (CoT) is suboptimal, the model often fails to reach the correct answer.<n>We introduce Reinforcement Learning with Re-solving (Re$2$), in which LLMs learn to flexibly abandon unproductive reasoning paths and restart the solution process when necessary.
  arXiv  Detail & Related papers  (2026-03-07T13:17:46Z)
• $\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space [71.23672814629448]
  $nabla$-Reasoner is an iterative generation framework that integrates differentiable optimization over token logits into the decoding loop.<n>$nabla$-Reasoner achieves over 20% accuracy improvement on a challenging mathematical reasoning benchmark.
  arXiv  Detail & Related papers  (2026-03-05T08:42:54Z)
• Optimal Learning-Rate Schedules under Functional Scaling Laws: Power Decay and Warmup-Stable-Decay [9.371921537573346]
  We study optimal learning-rate schedules (LRSs) under the functional scaling law.<n>LRSs accurately models the loss dynamics of both linear regression and large language model (LLM) pre-training.<n>We analyze optimal shape-fixed schedules, where only the peak learning rate is tuned.
  arXiv  Detail & Related papers  (2026-02-06T15:52:30Z)
• R-Stitch: Dynamic Trajectory Stitching for Efficient Reasoning [80.104336426172]
  Chain-of-thought (CoT) enhances problem-solving ability of large language models.<n>CoT incurs substantial inference cost due to long autoregressive trajectories.<n>We introduce R-Stitch, a training-free hybrid decoding framework.
  arXiv  Detail & Related papers  (2025-07-23T08:14:36Z)
• Statistical-Computational Trade-offs for Recursive Adaptive Partitioning Estimators [11.77169131123735]
  We show that greedy algorithms for high-dimensional regression get stuck at local optima.<n>We show that greedy training requires $exp(Omega(d))$ to achieve low estimation error.<n>This dichotomy mirrors that of two-layer neural networks trained with gradient descent (SGD) in the mean-field regime.
  arXiv  Detail & Related papers  (2024-11-07T03:11:53Z)
• Truncated Consistency Models [57.50243901368328]
  Training consistency models requires learning to map all intermediate points along PF ODE trajectories to their corresponding endpoints.<n>We empirically find that this training paradigm limits the one-step generation performance of consistency models.<n>We propose a new parameterization of the consistency function and a two-stage training procedure that prevents the truncated-time training from collapsing to a trivial solution.
  arXiv  Detail & Related papers  (2024-10-18T22:38:08Z)
• Reinforcement Learning from Human Feedback without Reward Inference: Model-Free Algorithm and Instance-Dependent Analysis [16.288866201806382]
  We develop a model-free RLHF best policy identification algorithm, called $mathsfBSAD$, without explicit reward model inference.<n>The algorithm identifies the optimal policy directly from human preference information in a backward manner.
  arXiv  Detail & Related papers  (2024-06-11T17:01:41Z)
• Training Overparametrized Neural Networks in Sublinear Time [14.918404733024332]
  Deep learning comes at a tremendous computational and energy cost. We present a new and a subset of binary neural networks, as a small subset of search trees, where each corresponds to a subset of search trees (Ds) We believe this view would have further applications in analysis analysis of deep networks (Ds)
  arXiv  Detail & Related papers  (2022-08-09T02:29:42Z)
• Posterior Coreset Construction with Kernelized Stein Discrepancy for Model-Based Reinforcement Learning [78.30395044401321]
  We develop a novel model-based approach to reinforcement learning (MBRL) It relaxes the assumptions on the target transition model to belong to a generic family of mixture models. It can achieve up-to 50 percent reduction in wall clock time in some continuous control environments.
  arXiv  Detail & Related papers  (2022-06-02T17:27:49Z)
• Predicting Training Time Without Training [120.92623395389255]
  We tackle the problem of predicting the number of optimization steps that a pre-trained deep network needs to converge to a given value of the loss function. We leverage the fact that the training dynamics of a deep network during fine-tuning are well approximated by those of a linearized model. We are able to predict the time it takes to fine-tune a model to a given loss without having to perform any training.
  arXiv  Detail & Related papers  (2020-08-28T04:29:54Z)

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.