Pith. sign in

REVIEW 4 major objections 5 minor 80 references

THOR: A Theta-Gamma Hierarchical Oscillatory Reasoning Framework for Multi-hop QA

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read THOR — a theta–gamma control loop with slot-schema memory and verifier-gated repair — more than doubles exact-match accuracy on MuSiQue and suppresses drift and compounding errors in multi-hop QA

desk verdict A well-engineered agentic wrapper with solid internal ablation, but the headline comparison against prior work rests on unverified numbers; the core mechanism deserves closer study. read the letter →

arxiv 2607.20459 v1 pith:QMQF4DYJ submitted 2026-05-15 cs.CL cs.AI

Ziyang Ling , Ronald X. Xu , Mingzhai Sun This is my paper · ORCID
classification cs.CLcs.AI
keywords multi-hopquestionansweringtheta-gammaoscillationattentiondecayerroraccumulationverificationandrepairslot-schemamemoryretrieval-augmentedgenerationLLMreasoning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Multi-hop question answering fails in two recurring ways, this paper argues: the model's attention drifts as the reasoning chain grows, and small early mistakes get committed and compounded. THOR is a proposed fix that organizes reasoning as a two-timescale loop: a slow 'theta' planner maintains the global reasoning frame in a typed slot-schema memory, while a fast 'gamma' executor retrieves evidence, answers one hop, and runs three verification checks. The check results drive explicit state transitions — continue, retrieve, repair, replan — so errors are caught and corrected by a controller instead of silently accumulating. On HotpotQA, 2WikiMultiHopQA, and MuSiQue, THOR reports large gains over a GPT-3.5-turbo baseline, including exact match moving from 19.7 to 48.5 on MuSiQue, with lower frame-shift and anchor-shift rates that persist at four hops. If the central claim holds, multi-hop accuracy can be improved primarily through the control loop and structured memory, not just through a stronger underlying model.

What carries the argument

The load-bearing mechanism is a two-timescale finite-state controller inspired by theta–gamma phase-amplitude coupling. The 'theta' outer loop is a planner (iPFC) that owns a slot-schema memory — a typed frame in which each hop binds named slots (sub-question, core entity, expected answer type, sub-answer, completion flag) to explicit values. The 'gamma' inner loop is an executor/verifier pair: iHPC retrieves evidence using topic-top-3 and context-top-3 fallback strategies plus replay-based query refinement, and iACC checks whether the candidate answer anchors on the intended entity, matches the expected type, and is supported by the evidence. The controller converts that three-bit check vec

What would settle it

Re-run THOR and each baseline in the main comparison with one fixed backbone, the same retriever, the same evidence budget, and the same answer-matching judge. If THOR's MuSiQue exact-match advantage over the strongest agentic baseline nearly vanishes, the claim that the framework itself — rather than backbone or retriever differences — drives the improvement is falsified.

Watch

Extended reading notes

Core claim

THOR's central claim: multi-hop failures are control failures — drift plus compounding — correctable by an explicit two-timescale loop. A slow theta planner (iPFC) maintains a slot-schema frame (sub-questions, entities, answer types); a fast gamma executor (iHPC) retrieves and proposes a sub-answer; a verifier (iACC) emits three binary checks (anchoring, type, support). These drive a finite-state controller: continue, retrieve, repair, or replan. With GPT-3.5-turbo on MuSiQue, THOR lifts exact match from 19.7 to 48.5, reduces 4-hop frame-shift to 15.3% and anchor-shift to 10.4%, and degrades less under adversarial injection. Ablations show every component matters; removing the planner costs

Load-bearing premise

The headline comparison assumes every baseline in the main table was evaluated under identical conditions — same backbone LLM, same retriever, same evidence budget, same evaluation split, and same answer-judging protocol — as THOR; if the baseline numbers were taken from their original papers under different conditions, the reported gains are not purely from the framework.

Editorial extensions

If this is right

  • A given LLM can become a much stronger multi-hop solver without changing the model: THOR reports exact-match gains of roughly 2× on MuSiQue for GPT-3.5-turbo and consistent gains across eight regular and five reasoning backbones.
  • Ablation evidence implies the four framework components are complementary: removing the planner costs 20.1 exact-match points on MuSiQue, while removing the retriever, verifier, or slot-schema memory each costs several points on all three benchmarks.
  • The anti-drift benefit grows with chain depth: THOR's frame-shift and anchor-shift rates are lowest at 4 hops, where chain-of-thought, chain-of-retrieval, and reversible-agent baselines drift most.
  • Iterative retrieval refinement is a concrete source of the gains: with up to three per-hop retries, recall@15 on MuSiQue rises from 36.4 to 44.5, surpassing a chain-of-retrieval baseline.
  • The retry budget Imax provides an explicit accuracy–cost dial: Imax=3 gives the best accuracy, and Imax=1 already achieves a better accuracy–token trade-off than a reversible multi-agent baseline.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension: replace the LLM planner and verifier with a rule-based planner and a cheaper verifier; if the anti-drift effect survives, per-question token cost could drop well below the reported figures.
  • The controller is essentially a deterministic protocol over typed signals, so the same design could transfer to other sequential tasks with explicit constraints — multi-step tool use, database query composition, or theorem proving — where the failure mode is also silent drift rather than missing information.
  • A stress-test for any future multi-hop system falls out of the adversarial protocol: inject a document that preserves the surface answer pattern but breaks the entity bridge, and measure the relative drop in exact match; this directly probes error accumulation independent of retrieval quality.
  • Because the paper's frame-shift and anchor-shift metrics rely on LLM and lightweight judges, an extension would be to publish the judged annotations and measure how sensitive the reported drift reductions are to judge choice — the framework's robustness claim depends on those metrics being reproducible.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. THOR is a closed-loop, brain-inspired reasoning wrapper for multi-hop QA. It alternates between a global 'theta' planner (iPFC), a local 'gamma' executor/retriever (iHPC), and a verifier (iACC), with a slot-schema memory and a finite-state controller that decides between Continue, Retrieve, Repair, and Replan. The authors claim that this design mitigates attention decay and error accumulation, and they report large EM/F1 gains over the plain GPT-3.5-turbo backbone on HotpotQA, 2WikiMultiHopQA, and MuSiQue, plus a backbone-adaptation study across eight regular and five reasoning LLMs. They also report FSR/ASR 'attention decay' metrics, an adversarial document-injection experiment, retrieval recall comparisons, and an accuracy-cost frontier. The main results are accompanied by ablations and a controlled add-on experiment, but the headline comparison to prior systems rests on baseline numbers whose experimental conditions are not documented.

Significance. If the central empirical claims hold, THOR would be a practically useful, largely model-agnostic reasoning wrapper: its controlled add-on experiment (Section 6.1, Table 7) isolates the contribution of finite-state control plus typed slot-schema memory, and the external EM/F1 and adversarial-drop benchmarks are appropriate targets. The paper also ships code, reports a human-reliability check for FSR (Table 10), and includes a broad backbone sweep — these are genuine strengths. The main reservation is that the strongest headline claim ('outperforms the vast majority of methods') depends on a baseline table whose provenance and comparability are not established. The mechanism-specific FSR/ASR evidence is also, in part, circular because the metrics are defined in terms of the same anchor/type/support checks the verifier is designed to optimize. These are addressable but load-bearing issues.

major comments (4)
  1. [§5.2, Table 1; Appendix C.2] The head-to-head comparison against 27 prior methods is not auditable. The paper never states whether any baseline was re-run under THOR's exact conditions: backbone (GPT-3.5-turbo), retriever (BGE-M3/BGE-Reranker), evaluation split and subset size (especially MuSiQue), answer normalization, and the same EM/F1 judging script. Section 5.1 describes only THOR's implementation, and Appendix C.2 gives conceptual descriptions of each method without experimental conditions. An internal red flag is the BELLE row for MuSiQue: F1=42.1 with EM=50.5, which is impossible for a single prediction set and suggests the table mixes numbers from heterogeneous sources. Please either rerun all baselines under matched conditions, or clearly source each number and provide a comparability table. Without this, the superiority claim over prior methods is not supported.
  2. [§5.1, §5.2.2, Algorithm 2] The FSR/ASR evidence for 'attention decay mitigation' is partially circular with THOR's own verification checks. ASR is essentially the negation of C_ei from Algorithm 2 ('anchor mention in evidence'), and FSR is judged by a GPT-4o rubric for sub-question alignment with gold decomposition, which is closely related to the type/constraint checks C_tau and C_r. Since iACC explicitly rejects anchor-missing evidence and drives repair, low ASR is to some extent guaranteed by the construction. The authors report human agreement only for FSR (Table 10), not for ASR. To make the mechanistic claim non-circular, please (i) report FSR/ASR with human-validated rubrics on both metrics, (ii) show that FSR/ASR improvements predict EM/F1 gains beyond the effect of retrieval quality, or (iii) rely on the external adversarial and EM/F1 results as the primary evidence and present FSR/ASR as descriptive diag
  3. [§5.2, Tables 2, 3, 6, 7] No variance or statistical testing is reported anywhere. In Tables 2, 3, 6, and 7, every number is a single run. Several comparisons are within 1–2 EM/F1 points (e.g., Table 6: Qwen-3-Thinking 2Wiki F1 72.9→73.0; Table 7: FSR 13.1 vs 12.1), which is well within typical LLM stochasticity unless explicitly controlled. Please report means and standard deviations over at least 3–5 seeds (or fixed seeds plus sensitivity analysis), and run paired significance tests for the main backbone and the add-on comparison. At minimum, the ablation and backbone claims should not rely on differences that are smaller than expected run-to-run variance.
  4. [§5.2.5, Table 6] The claim that THOR 'consistently yields large gains' and 'generalizes' is contradicted by several cells in the authors' own table. Concretely: DeepSeek-R1 2Wiki F1 drops from 70.7 to 69.2; Gemini-2.5-Pro 2Wiki F1 drops from 82.9 to 82.3; O1 MuSiQue drops from 41.7 to 39.1 EM and from 55.1 to 49.4 F1. There are also unexplained numeric differences with Table 2 for the same GPT-3.5-turbo baseline (EM 31.9 vs 31.2). These patterns weaken the 'plug-and-play wrapper' story. Please report per-backbone wins/losses, discuss cases where THOR hurts, and consider whether the generalization claim should be softened to 'most backbones'.
minor comments (5)
  1. [Table 4] The THOR row for 'Drop' is printed as '12.79.3', which should be '12.7 9.3'. Please format the two columns clearly.
  2. [Algorithm 1] The symbol 'I' is used for an indicator function but is not defined. Also 'δ_i ← I(Insuff(E_topic))' is unclear: what exactly does iACC return, and how is 'insufficient' operationalized? Please define these variables and the threshold.
  3. [§5.1] 'fixedLLM judge' has a missing space; please proofread formulas and prose for spacing errors (e.g., '0.7 9.8' in Table 3, 'THOR-Iter@273.1' in Table 5 caption).
  4. [§5.2.6 / Figure 4] The accuracy-cost discussion would be clearer with the exact token/call/latency numbers from Table 11 referenced in the main text, and with a statement on whether the cost includes the verification calls or only generation tokens.
  5. [Appendix C.3] The FSR judge uses the gold decomposition and evaluates only the first h predicted hops. This is reasonable, but please state explicitly in the main text that FSR/ASR are computed on MuSiQue only, since the main text's 'all three benchmarks' phrasing in §5.2.2 is ambiguous.

Circularity Check

1 steps flagged · score 6.0 of 10

ASR validation metric is identical to THOR's own evidence-anchoring gate, making the anchor-shift reduction claim partly by construction; EM/F1 and gold-based FSR results remain independent.

  1. self definitional [Section 5.1 (ASR definition) with Algorithm 2 (Evidence Anchoring Check) and Algorithm 3 (controller gating)]
    "Define hop-level anchor alignment aa_t(x)∈{0,1} , where aa_t(x)=1 if the anchor entity in the predicted sub-question is mentioned in the retrieved evidence at hop t. ... (1) Evidence Anchoring Check: C_ei ← I[Mention(ce_t, ei)] ... if Acc_i = ⟨1,1,1⟩ then c_t ← CONTINUE(θ→γ)"

    ASR's condition—anchor entity mentioned in retrieved evidence—is exactly the condition iACC verifies as C_ei, and Algorithm 3 only commits a hop to CONTINUE when Acc_i = ⟨1,1,1⟩, i.e., when C_ei = 1. Therefore every hop that THOR commits is anchor-present by construction, so THOR's low ASR is a restatement of its own verification gate rather than independent evidence that attention decay / anchor shift is mitigated. The 'validation' metric and the mechanism are the same check.

full rationale

The central accuracy and robustness claims rest on external EM/F1 benchmarks (HotpotQA, 2WikiMultiHopQA, MuSiQue) and on gold-decomposition-based FSR, so the paper is not globally circular. The specific attention-decay evidence, however, includes ASR, whose definition is the same condition as THOR's Evidence Anchoring Check that gates whether a hop may be committed. Reporting low ASR as proof of anchor-shift mitigation is therefore partly a construction: the controller enforces the measured property. The controlled add-on experiment and external EM/F1 comparisons give independent support for the framework's value, and FSR is not reduced by construction. The Table 1 baseline comparability issue (including the impossible BELLE F1<EM cell) is a separate reproducibility/data-quality concern, not a circularity finding.

Assumptions & free parameters 2 free parameters · 3 assumptions · 3 invented entities

The framework does not introduce new physical entities; it assumes benchmark validity, verifier reliability, and gold-decomposition ground truth. The main hand-set design choices are Imax=3 and top-k=3 retrieval widths.

free parameters (2)
  • Imax (max per-hop retry budget) = 3 (default; swept over 1,3,5 in §5.2.6)
    Chosen by hand; controls when repair escalates to replan; results in Table 11 depend on it.
  • topic_top_k and context_top_k = 3 and 3
    Retrieval width in Algorithm 1; fixed without sensitivity analysis.
assumptions (3)
  • domain assumption Gold decomposition of MuSiQue provides ground truth for hop-level frame alignment
    FSR computation compares predicted sub-questions to gold sub-questions from the dataset (C.3.1); assumes gold hops are unambiguous and that GPT-4o can judge semantic equivalence reliably.
  • domain assumption LLM verification signals are reliable enough to drive control decisions
    iACC's three binary checks gate continue/retrieve/repair/replan; if the verifier is systematically wrong, the framework's control policy degrades. Paper does not measure verifier accuracy in isolation.
  • domain assumption The three benchmark datasets are valid measures of multi-hop reasoning capability
    HotpotQA/2Wiki/MuSiQue are used as ground truth; standard but still an assumption about what these benchmarks measure.
invented entities (3)
  • iPFC (Prefrontal Cortex-inspired planner)
    purpose: Global frame planning, repair, replan; maintains theta memory
    Software module; its name and brain mapping are analogical, not empirically established as the reason for gains.
  • iHPC (Hippocampus-inspired executor)
    purpose: Per-hop retrieval and sub-answer generation
    Software module; no independent evidence that hippocampal analogy contributes beyond the retrieval algorithm.
  • iACC (Anterior Cingulate-inspired verifier)
    purpose: Three binary checks that drive state transitions
    Verifier design is central; only in-paper experiments support its usefulness; no external validation of the check reliability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of THOR: A Theta-Gamma Hierarchical Oscillatory Reasoning Framework for Multi-hop QA." pith.science (2026). https://pith.science/paper/QMQF4DYJ

@misc{pith2026260720459,
  author       = {Pith},
  title        = {Pith review of: THOR: A Theta-Gamma Hierarchical Oscillatory Reasoning Framework for Multi-hop QA},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QMQF4DYJ}},
  note         = {Machine review of arXiv:2607.20459}
}
read the original abstract

Multi-hop question answering requires retrieving and integrating evidence from multiple contexts. Despite the rapid progress of current research, multi-hop reasoning remains constrained by two persistent limitations: attention decay, where the model's focus on main question degrades as the reasoning chain grows, and error accumulation, where mistakes propagate across hops and compounds into final failure. Inspired by Theta-Gamma hierarchical oscillation which decouples global planning from local retrieval, enabling efficient attention transfer between hops and a verification and repair mechanism that interrupts the accumulation of errors in the wrong paths, we present THOR, a brain-inspired Theta-Gamma hierarchical oscillatory reasoning framework. Extensive comparative experiments and specific validation experiments on multi-hop QA benchmarks demonstrate that THOR improves answer accuracy and robustness while mitigating limitations, showcasing its generalization across different backbones.

Figures

Figures reproduced from arXiv: 2607.20459 by the authors.

Figure 1
Figure 1. Comparison of multi-hop pipelines: Chain-of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Current methods still have the problem of [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of the framework of THOR. THOR alternates between a global [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Accuracy-cost trade-off experiments on [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The proportion of incorrect but with all evidence present samples remains relatively high on MuSiQue, revealing a faithfulness gap between evi￾dence availability and conclusion integration. inconsistent. As shown in table 8 and 9, THOR differs by detecting error and co…
Figure 6
Figure 6. Figure 6: ASR across hop settings on frame-correct but answer-error cases and all error cases. tem implicitly commits to the wrong referent. (3) Context-induced hijacking: a retrieved passage con￾tains a prominent related entity that attracts subse￾quent retrieval and reasoning,…
Figure 8
Figure 8. Figure 8: Illustration of error accumulation across hops. A locally plausible anchor error at the first hop is committed [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Theta–Gamma hierarchical oscillation in hu [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Ablation experiments on MuSiQue. From left to right: [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Accuracy-cost results with 10-binned density distribution on THOR. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Example of fake-answer substitution with bridge-entity breaking. The shortcut-bearing answer pattern is [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Prompt template for iPFC. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Prompt template for iHPC. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: Prompt template for iACC. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Example of frame shift caused by ambiguous surface cues. THOR rejects the incorrect sense of “Apple” [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: Example of error accumulation from an under-specified intermediate binding. THOR explicitly binds the [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: Example where local repair is insufficient and THOR must trigger REPLAN. By refining a coarse [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 3 canonical work pages

  1. [1]

    2025 , eprint=

    Multi-Hop Question Answering: When Can Humans Help, and Where do They Struggle? , author=. 2025 , eprint=

    2025

  2. [2]

    2018 , eprint=

    HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering , author=. 2018 , eprint=

    2018

  3. [3]

    2022 , eprint=

    MuSiQue: Multihop Questions via Single-hop Question Composition , author=. 2022 , eprint=

    2022

  4. [4]

    Foundations and Trends in Information Retrieval , volume =

    Multi-hop Question Answering , author =. Foundations and Trends in Information Retrieval , volume =. 2024 , publisher =

    2024

  5. [5]

    ACM Computing Surveys , volume =

    Multi-step Reasoning with Large Language Models, a Survey , author =. ACM Computing Surveys , volume =. 2025 , doi =

    2025

  6. [6]

    arXiv preprint arXiv:2510.04493 , year=

    Multi-Hop Question Answering: When Can Humans Help, and Where do They Struggle? , author=. arXiv preprint arXiv:2510.04493 , year=

    arXiv

  7. [7]

    Avoiding Reasoning Shortcuts: Adversarial Evaluation, Training, and Model Development for Multi-Hop QA

    Jiang, Yichen and Bansal, Mohit. Avoiding Reasoning Shortcuts: Adversarial Evaluation, Training, and Model Development for Multi-Hop QA. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. doi:10.18653/v1/P19-1262

    doi:10.18653/v1/p19-1262 2019

  8. [8]

    Science , volume=

    Entrainment of neuronal oscillations as a mechanism of attentional selection , author=. Science , volume=. 2008 , publisher=

    2008

Show all 80 references
  1. [9]

    Neuron , volume=

    Gamma and beta bursts underlie working memory , author=. Neuron , volume=. 2016 , publisher=

    2016

  2. [10]

    Transactions of the Association for Computational Linguistics , volume =

    Lost in the Middle: How Language Models Use Long Contexts , author =. Transactions of the Association for Computational Linguistics , volume =. 2024 , publisher =

    2024

  3. [11]

    Advances in Neural Information Processing Systems , volume =

    Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting , author =. Advances in Neural Information Processing Systems , volume =. 2023 , url =

    2023

  4. [12]

    and Jensen, Ole , title =

    Lisman, John E. and Jensen, Ole , title =. Neuron , year =

  5. [13]

    2025 , eprint=

    ComposeRAG: A Modular and Composable RAG for Corpus-Grounded Multi-Hop Question Answering , author=. 2025 , eprint=

    2025

  6. [14]

    Current Opinion in Behavioral Sciences , year =

    Ursino, Mauro and Pirazzini, Gabriele , title =. Current Opinion in Behavioral Sciences , year =

  7. [15]

    and Rosen, Andrew M

    Tamura, Makoto and Spellman, Timothy J. and Rosen, Andrew M. and Gogos, Joseph A. and Gordon, Joshua A. , title =. Nature Communications , year =

  8. [16]

    and Foster, David J

    Pfeiffer, Brad E. and Foster, David J. , title =. Nature , year =

  9. [17]

    and Cohen, Jonathan D

    Miller, Earl K. and Cohen, Jonathan D. , title =. Annual Review of Neuroscience , year =

  10. [18]

    and Cohen, Jonathan D

    Shenhav, Amitai and Botvinick, Matthew M. and Cohen, Jonathan D. , title =. Neuron , year =

  11. [19]

    Annual Review of Psychology , year =

    Baddeley, Alan , title =. Annual Review of Psychology , year =

  12. [20]

    Nature Neuroscience , volume =

    Episodic sequence memory is supported by a theta--gamma phase code , author =. Nature Neuroscience , volume =. 2016 , doi =

    2016

  13. [21]

    Neuron , volume =

    What is a cognitive map? Organizing knowledge for flexible behavior , author =. Neuron , volume =. 2018 , doi =

    2018

  14. [22]

    Neuron , volume =

    Prefrontal phase locking to hippocampal theta oscillations , author =. Neuron , volume =. 2005 , doi =

    2005

  15. [23]

    Nature Reviews Neuroscience , volume =

    Prefrontal--hippocampal interactions in episodic memory , author =. Nature Reviews Neuroscience , volume =. 2017 , doi =

    2017

  16. [24]

    Nature Neuroscience , volume =

    Prioritized memory access explains planning and hippocampal replay , author =. Nature Neuroscience , volume =. 2018 , doi =

    2018

  17. [25]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , pages =

    Self-prompted Chain-of-Thought on Large Language Models for Open-domain Multi-hop Reasoning , author =. Findings of the Association for Computational Linguistics: EMNLP 2023 , pages =. 2023 , address =. 2310.13552 , archivePrefix =

    2023 arXiv

  18. [26]

    2024 , eprint=

    Tree of Reviews: A Tree-based Dynamic Iterative Retrieval Framework for Multi-hop Question Answering , author=. 2024 , eprint=

    2024

  19. [27]

    2023 , eprint=

    Tree of Thoughts: Deliberate Problem Solving with Large Language Models , author=. 2023 , eprint=

    2023

  20. [28]

    2024 , eprint=

    FSM: A Finite State Machine Based Zero-Shot Prompting Paradigm for Multi-Hop Question Answering , author=. 2024 , eprint=

    2024

  21. [29]

    2023 , eprint=

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author=. 2023 , eprint=

    2023

  22. [30]

    2023 , eprint=

    Least-to-Most Prompting Enables Complex Reasoning in Large Language Models , author=. 2023 , eprint=

    2023

  23. [31]

    2022 , eprint=

    Atlas: Few-shot Learning with Retrieval Augmented Language Models , author=. 2022 , eprint=

    2022

  24. [32]

    Measuring and Narrowing the Compositionality Gap in Language Models

    Press, Ofir and Zhang, Muru and Min, Sewon and Schmidt, Ludwig and Smith, Noah and Lewis, Mike. Measuring and Narrowing the Compositionality Gap in Language Models. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023. doi:10.18653/v1/2023.findings-emnlp.378

    2023 doi

  25. [33]

    Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions

    Trivedi, Harsh and Balasubramanian, Niranjan and Khot, Tushar and Sabharwal, Ashish. Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Vo...

    2023 doi

  26. [34]

    Active Retrieval Augmented Generation

    Jiang, Zhengbao and Xu, Frank and Gao, Luyu and Sun, Zhiqing and Liu, Qian and Dwivedi-Yu, Jane and Yang, Yiming and Callan, Jamie and Neubig, Graham. Active Retrieval Augmented Generation. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing....

    2023 doi

  27. [35]

    Probabilistic Tree-of-thought Reasoning for Answering Knowledge-intensive Complex Questions

    Cao, Shulin and Zhang, Jiajie and Shi, Jiaxin and Lv, Xin and Yao, Zijun and Tian, Qi and Hou, Lei and Li, Juanzi. Probabilistic Tree-of-thought Reasoning for Answering Knowledge-intensive Complex Questions. Findings of the Association for Computational Linguistics: EMNLP 2023...

    2023 doi

  28. [36]

    2024 , eprint=

    BeamAggR: Beam Aggregation Reasoning over Multi-source Knowledge for Multi-hop Question Answering , author=. 2024 , eprint=

    2024

  29. [37]

    2023 , eprint=

    Enhancing Retrieval-Augmented Large Language Models with Iterative Retrieval-Generation Synergy , author=. 2023 , eprint=

    2023

  30. [38]

    2025 , eprint=

    Chain-of-Retrieval Augmented Generation , author=. 2025 , eprint=

    2025

  31. [39]

    2024 , eprint=

    EfficientRAG: Efficient Retriever for Multi-Hop Question Answering , author=. 2024 , eprint=

    2024

  32. [40]

    2024 , eprint=

    Chain of Agents: Large Language Models Collaborating on Long-Context Tasks , author=. 2024 , eprint=

    2024

  33. [41]

    2025 , eprint=

    HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models , author=. 2025 , eprint=

    2025

  34. [42]

    2024 , eprint=

    KAG: Boosting LLMs in Professional Domains via Knowledge Augmented Generation , author=. 2024 , eprint=

    2024

  35. [43]

    2025 , eprint=

    ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA , author=. 2025 , eprint=

    2025

  36. [44]

    2025 , eprint=

    GeAR: Graph-enhanced Agent for Retrieval-augmented Generation , author=. 2025 , eprint=

    2025

  37. [45]

    2025 , note =

    Nahid, Md and Rafiei, Davood , journal =. 2025 , note =

    2025

  38. [46]

    2025 , eprint=

    Search-o1: Agentic Search-Enhanced Large Reasoning Models , author=. 2025 , eprint=

    2025

  39. [47]

    2025 , eprint=

    Talk to Right Specialists: Routing and Planning in Multi-agent System for Question Answering , author=. 2025 , eprint=

    2025

  40. [48]

    2025 , eprint=

    BELLE: A Bi-Level Multi-Agent Reasoning Framework for Multi-Hop Question Answering , author=. 2025 , eprint=

    2025

  41. [49]

    Constructing a Multi-hop

    Ho, Xanh and Nguyen, Anh-Khoa Duong and Sugawara, Saku and Aizawa, Akiko , booktitle =. Constructing a Multi-hop. 2020 , address =

    2020

  42. [50]

    and Jensen, Ole , title =

    Lisman, John E. and Jensen, Ole , title =. Neuron , year =. doi:10.1016/j.neuron.2013.03.007 , url =

    2013 doi

  43. [51]

    and Kastner, Sabine , title =

    Fiebelkorn, Ian C. and Kastner, Sabine , title =. Trends in Cognitive Sciences , year =. doi:10.1016/j.tics.2018.11.009 , url =

    2018 doi

  44. [52]

    and Buhl, Eberhard H

    Fisahn, Andreas and Pike, Francis G. and Buhl, Eberhard H. and Paulsen, Ole , title =. Nature , year =. doi:10.1038/28179 , url =

    doi

  45. [53]

    Theta oscillations in the hippocampus , journal =

    Buzs. Theta oscillations in the hippocampus , journal =. 2002 , volume =. doi:10.1016/S0896-6273(02)00586-X , url =

    2002 doi

  46. [54]

    and Wilson, Matthew A

    Jones, Matthew W. and Wilson, Matthew A. , title =. PLOS Biology , year =. doi:10.1371/journal.pbio.0030402 , url =

    doi

  47. [55]

    Read , title =

    Schultz, Wolfram and Dayan, Peter and Montague, P. Read , title =. Science , year =. doi:10.1126/science.275.5306.1593 , url =

  48. [56]

    and Wilson, Matthew A

    Foster, David J. and Wilson, Matthew A. , title =. Hippocampus , year =. doi:10.1002/hipo.20345 , url =

    doi

  49. [57]

    Science , volume =

    Neuronal oscillations in cortical networks , author =. Science , volume =. 2004 , doi =

    2004

  50. [58]

    Science , volume =

    High gamma power is phase-locked to theta oscillations in human neocortex , author =. Science , volume =. 2006 , doi =

    2006

  51. [59]

    Neuron , volume =

    The - neural code , author =. Neuron , volume =. 2013 , doi =

    2013

  52. [60]

    Proceedings of the National Academy of Sciences , volume =

    Theta-gamma coupling increases during the learning of item-context associations , author =. Proceedings of the National Academy of Sciences , volume =. 2009 , doi =

    2009

  53. [61]

    Proceedings of the National Academy of Sciences , volume =

    Cross-frequency coupling supports multi-item working memory in the human hippocampus , author =. Proceedings of the National Academy of Sciences , volume =. 2010 , doi =

    2010

  54. [62]

    PLoS Biology , volume =

    Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task , author =. PLoS Biology , volume =. 2005 , doi =

    2005

  55. [63]

    Neuron , volume =

    Coherent theta oscillations and reorganization of spike timing in the hippocampal-prefrontal network upon learning , author =. Neuron , volume =. 2010 , doi =

    2010

  56. [64]

    Trends in Cognitive Sciences , volume =

    Frontal theta as a mechanism for cognitive control , author =. Trends in Cognitive Sciences , volume =. 2014 , doi =

    2014

  57. [65]

    Nature Neuroscience , volume =

    Common medial frontal mechanisms of adaptive control in humans and rodents , author =. Nature Neuroscience , volume =. 2013 , doi =

    2013

  58. [66]

    2006 , doi =

    High Gamma Power Is Phase-Locked to Theta Oscillations in Human Neocortex , journal =. 2006 , doi =

    2006

  59. [67]

    Neuron , volume =

    The - Neural Code , author =. Neuron , volume =. 2013 , doi =

    2013

  60. [68]

    Nature , volume =

    Hippocampal place-cell sequences depict future paths to remembered goals , author =. Nature , volume =. 2013 , doi =

    2013

  61. [69]

    Neuron , volume =

    Rhythms for Cognition: Communication through Coherence , author =. Neuron , volume =. 2015 , doi =

    2015

  62. [70]

    Transactions of the Association for Computational Linguistics , volume =

    Lost in the Middle: How Language Models Use Long Contexts , author =. Transactions of the Association for Computational Linguistics , volume =. 2024 , doi =

    2024

  63. [71]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , address =

    Do Large Language Models Latently Perform Multi-Hop Reasoning? , author =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , address =. 2024 , doi =

    2024

  64. [72]

    2022 , eprint =

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author =. 2022 , eprint =

    2022

  65. [73]

    doi:10.48550/arXiv.2410.21276 , url =

    arXiv preprint , year =. doi:10.48550/arXiv.2410.21276 , url =. 2410.21276 , archivePrefix=

    doi arXiv

  66. [74]

    2025 , month = apr, url =

    2025

  67. [75]

    doi:10.48550/arXiv.2412.19437 , url =

    arXiv preprint , year =. doi:10.48550/arXiv.2412.19437 , url =. 2412.19437 , archivePrefix=

    doi arXiv

  68. [76]

    doi:10.48550/arXiv.2501.12948 , url =

    arXiv preprint , year =. doi:10.48550/arXiv.2501.12948 , url =. 2501.12948 , archivePrefix=

    doi arXiv

  69. [77]

    doi:10.48550/arXiv.2412.15115 , url =

    arXiv preprint , year =. doi:10.48550/arXiv.2412.15115 , url =. 2412.15115 , archivePrefix=

    doi arXiv

  70. [78]

    doi:10.48550/arXiv.2505.09388 , url =

    arXiv preprint , year =. doi:10.48550/arXiv.2505.09388 , url =. 2505.09388 , archivePrefix=

    doi arXiv

  71. [79]

    2024 , eprint =

    M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation , author =. 2024 , eprint =

    2024

  72. [80]

    Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '94) , pages=

    Some simple effective approximations to the 2-Poisson model for probabilistic weighted retrieval , author=. Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '94) , pages=. 1994 , organization=

    1994

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.