acceptodds
Under review as a conference paper at ICLR 2027

Energy-guided Recursive Model

Abstract

Recursive models show promise on reasoning and language tasks, yet their test-time scaling lacks a principled criterion for selecting trajectories or determining recurrent depth. We introduce Energy-guided Recursive Model (ERM), which uses Hopfield-type memories of valid local and global structures to assign intrinsic energies to candidate trajectories. These energies guide candidate selection and suggest an effective range of recurrent depths, implying that deeper recurrence does not necessarily improve reasoning accuracy. They also enable sampling methods such as parallel tempering to improve exploration. For reasoning tasks, ERM achieves optimal solutions on Sudoku (), Pencil Puzzle Bench (PPBench, ) and Maze (), reaching the best accuracy in recursive modeling. On language modeling, ERM reduces RedPajama-V2 perplexity by with marginal inference overhead. The results support energy guidance as a practical framework for improving test-time scaling in recursive models.

Primary area:
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Keywords:
Recursive reasoning, Energy-based models, Test-time computation, Hopfield networks

Reviews

Among:
Shape:
1234this paper, Top 49%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.33 over 3 reviews, above 51% of the others.
Reviewer_RmcxOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

The paper proposes Energy-guided Recursive Model (ERM), an inference framework for fixed iterative models. On puzzles, it scores candidate outputs with soft retrieval over valid local patterns and explicit global constraints. It uses these scores for selection, parallel-tempering exchanges, and a batch-level stopping rule. On language modeling, it combines candidate probabilities with retrieved continuation statistics through a normalized geometric mean. ERM matches the candidate-pool oracle on the three main puzzle experiments. Parallel tempering adds three Sudoku solves. On a RedPajama-V2 proxy, the language correction reduces perplexity from 44.189 to 43.419. The paper also reports rule-checker controls, memory ablations, and reduced-depth Maze inference.

Critical strengths

The shared-pool evaluation separates candidate generation from candidate selection. Table 1 measures how much each selector recovers from the same available answers. ERM improves PPBench from 251 to 265 solves relative to q-head selection. It also recovers every available solution in the main Sudoku and Maze pools. Reporting the oracle prevents these selection gains from being confused with generating new solutions. Tables 9 and 10 provide valuable additional controls using simpler rule scores.

Critical weaknesses

The central reasoning results do not establish a benefit specific to Hopfield retrieval. In Table 9, global-only scoring, unweighted violation count, first-valid selection, and full ERM all solve exactly 993/1000 Maze instances. In Table 10, a discrete score using the complete local and global rules solves the same 265/301 PPBench instances as full ERM. Thus, these controls already explain the headline selection benefit through ordinary task verification. The local-versus-global ablations show that all relevant rules matter. They do not demonstrate that the soft memory formulation adds value over evaluating those rules directly. The claimed broader role in sampling and stopping could distinguish the method, but this is not established against checker-based PT or stopping. PT adds only three Sudoku solves, without repeated seeds. Table 4 compares different candidate pools, and stable batch energy is not a correctness certificate. The evidence is useful, but it supports a narrower contribution than the main Hopfield-guidance claim.

The language result and the claim of marginal overhead need a cleaner evaluation boundary. Equations 9-10 reduce exactly to p_ERM(v) proportional to sqrt(p0(v) * ((1-g_t)p0(v) + g_t q_t(v))). This is a memory-based probability correction. The supplied mixture and memory-form ablations are useful, but they do not establish a special connection to recursive reasoning. Training-free memory augmentation for language prediction already has a direct precedent in Grave et al., "Improving Neural Language Models with a Continuous Cache," ICLR 2017: https://openreview.net/pdf?id=B184E5qee . More importantly, Appendix B.2 states that the same 33k-document proxy supports development and evaluation. The reported bootstrap interval measures document-sampling uncertainty within that proxy. It does not account for model or memory choices made using it. Finally, the 2.83-hour language runtime covers the full comparison, and the development timing covers all branches together. Table 13 excludes retrieval and constraint scoring. These records do not isolate the added cost required to substantiate "marginal inference overhead."

Questions for authors
  • On matched candidate pools and repeated seeds, where does the soft Hopfield score improve over the complete discrete verifier? Please include checker-based PT or stopping if those are the intended advantages beyond final selection.

  • Which documents and results informed memory, gate, checkpoint, and variant selection? Please report the frozen final configuration on a separate evaluation split that was not used for development. Does the gain persist with one candidate, or is there evidence that recurrence or multiple trajectories are important to the correction?

  • Please separate candidate-generation time, memory construction, lookup, constraint scoring, normalization, and peak memory. What is the measured incremental cost over candidate averaging and over a single candidate? For Maze stopping, what is the wall-clock comparison with matched PT fixed-depth and first-valid stopping?

  • What exactly are the PTRM and DeepSeek-V4 bars in Figure 1? Please provide their generator identities, prompts, candidate budgets, and evaluation protocol, and explain how these bars relate to the shared EqR selectors in Table 1.

Reviewer_KQg9OpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The authors propose ERM, an energy-based approach for reasoning that leverages energy minimization to do reasoning on tasks such as Sudoku, Pencil Puzzle Bench, and Maze. The authors also demonstrate ERM for language modeling as well as new inference time sampling techniques.

Critical strengths
  • The authors demonstrate that energy-based reasoning can achieve strong performance compared to existing models.
  • The ERM enables different sampling strategies, which the authors demonstrate enable useful capabilities.
  • The ERM is decently calibrated with sample quality.
  • The proposed approach has little to no training cost (this can be seen as an advantage, but is also a massive disadvantage for scalability, as discussed in weaknesses).
Critical weaknesses
  • The biggest weakness, which is quite large, is that the proposed ERM is not a learned energy function (at least, in the modern sense of what the term learning is). While this achieves strong performance on several of the proposed tasks, I strongly believe this is a massive limiting factor. Progress over the last 15 years has demonstrated the core importance of learning for scalability, generalization, etc. Not learning the energy functions means the ERM approach is not very scalable to real-world tasks and thus vastly limits applicability.
  • The comparison to EBTs is also not directly comparable, with EBT selecting per document (and just measuring scalar energy comparison) whereas ERM builds a new distribution.
Questions for authors

Why did the authors prioritize not learning the energy function? How does the energy-based stopping compare to EqR's q-head? How would the authors use the proposed approach for large scale problems?

Reviewer_iKxYOpenReview ↗
Rating
3
Confidence
3
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary
  • The authors introduce Energy-guided Recursive Models (ERMs) which use local and global structures to assign intrinsic Hopfield-style energies to candidate trajectories.
  • The authors find that ERMs achieve optimal solutions on a range of reasoning tasks, thus supporting the idea of energy guidance as a practical framework for improving test-time scaling in recursive models.
Critical strengths
  • The idea is principally sound and generally works.
Critical weaknesses
  • The gains oftentimes are fairly marginal.
Questions for authors
  • It would be useful to see how much of the fixed point problem is avoided by the ERM. Likewise, analyses to check if these models also quickly commit to an answer vs actually use extra steps to refine answers would be insightful.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.