Pith. sign in

REVIEW 4 major objections 5 minor 30 references

This paper establishes that feedback-augmented self-distillation fails to improve retrieval-interleaved search agents because models settle into input-agnostic decoding templates, and that stabilizing the self-teacher only partially fixes t

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 17:38 UTC pith:NHH3DAQX

load-bearing objection A genuinely useful negative result about feedback-augmented self-distillation for search agents, but the named mechanism—'decoding collapse'—is inferred rather than demonstrated. the 4 major comments →

arxiv 2607.17558 v1 pith:NHH3DAQX submitted 2026-07-20 cs.AI

Why Does Feedback-Augmented Self-Distillation Fail to Improve Retrieval-Interleaved Search Agents?

Fan Yang , Rui Meng , Yuxin Wen This is my paper
classification cs.AI
keywords self-distillationretrieval-interleaved search agentsdecoding collapseon-policy distillationteacher regularizationEMA teacheropen-domain QAnegative results
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish why feedback-augmented self-distillation (FA-SD), a training method that distills token-level behavior from a model prompted with a successful search rollout, does not improve retrieval-interleaved search agents. Across seven open-domain QA benchmarks, FA-SD variants fail to sustain improvement even though the feedback-augmented teacher branch is often stronger than the student. The paper's causal diagnosis is decoding collapse: trajectories look diverse but follow recurring reasoning-and-search templates that ignore the question, so the KL signal carries no useful information. Stabilizing the teacher with an exponential moving average partially rescues training—raising final average score from 0.114 for the base model to 0.206—but shows that self-teacher instability, especially prompt inconsistency, is a central obstacle. The work matters because it isolates a failure mode that aggregate metrics miss and locates the problem in supervision consistency rather than teacher capability.

Core claim

The central claim is that feedback-augmented self-distillation (FA-SD) does not reliably transfer the advantage of seeing a successful rollout into the unconditioned search policy. In FA-SD, the same policy is prompted with a correct earlier attempt and used with a stop-gradient as a per-token teacher for the unconditioned student; the paper finds this supervision collapses: the model adopts recurring reasoning-and-search templates that produce diverse-looking but input-agnostic trajectories, making the reverse-KL distillation signal uninformative. The paper supports this with training dynamics (near-zero pass rates, declining effective sample ratio) and by showing the self-teacher branch ou

What carries the argument

The central object is the FA-SD loss, which distills token-level distributions from a feedback-augmented self-teacher—the same policy conditioned on a successful rollout from the rollout group—into the unconditioned student via reverse KL, with a stop-gradient on the teacher branch. The paper's named failure mode is decoding collapse, defined as recurring reasoning-and-search output templates that appear diverse but are agnostic to the input question, rendering the KL signal uninformative. The stabilizing machinery is teacher regularization: a fixed reference teacher (the student at the start of training) or an exponential moving average (EMA) teacher that slows parameter drift and makes sup

Load-bearing premise

The load-bearing premise is that the observed training collapse is caused by decoding collapse—recurring, input-agnostic templates—rather than by another property of the optimization (e.g., noisy single-sample reverse-KL gradients over long trajectories, noisy success labels, or the short 200-step training budget).

What would settle it

At a late FA-SD training step, take rollouts for many different questions and measure (1) token/query overlap across questions (e.g., self-BLEU or entity overlap) and (2) the KL loss between the unconditioned student and the feedback-augmented self-teacher. If trajectories remain clearly question-specific while pass rates hover near zero, the decoding-collapse explanation is falsified; if trajectories are near-identical across questions while the KL loss stays low, it is confirmed.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the central claim is correct, FA-SD without teacher regularization will not improve retrieval-interleaved search agents even when successful rollouts are available as privileged demonstrations.
  • EMA-regularized FA-SD can improve over the base model (0.206 vs 0.114 average score) but still falls far short of an RL-trained search agent (0.348), suggesting dense self-distillation alone is not a substitute for outcome-level reinforcement.
  • Prompt-time feedback can elicit stronger behavior from the same model, but that advantage is not reliably internalized by the unconditioned policy through token-level KL distillation.
  • Feedback augmentation does not straightforwardly generalize to external-teacher on-policy distillation: adding a successful rollout to the teacher's prompt (FA-MOPD) underperforms standard MOPD, indicating privileged feedback can introduce prompt inconsistency even with a strong teacher.
  • Training diagnostics should track the effective sample ratio—the fraction of rollouts providing a valid distillation signal—not just pass rate or final EM, because aggregate metrics can miss decoding collapse.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: decoding collapse could be tested directly by measuring trajectory overlap across questions at a late training step—e.g., self-BLEU of search queries or entity overlap—rather than inferring collapse from pass rates alone.
  • Beyond the paper: the same inconsistency decomposition (model drift vs. prompt conditioning) likely applies to other self-improvement settings, not just search agents; EMA stabilization may be broadly useful whenever a model distills from its own conditioned outputs.
  • A plausible alternative explanation the paper leaves open is that single-sample reverse-KL gradients over long interleaved rollouts are too high-variance to train reliably; comparing multi-sample teacher estimates or lower-variance estimators could separate optimization noise from template collapse.
  • If decoding collapse is real, evaluation protocols for agentic RL should include input-specificity checks—such as requiring that search queries contain question-specific entities—so that diverse-looking but degenerate policies are not scored as successful.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper studies Feedback-Augmented Self-Distillation (FA-SD) for retrieval-interleaved search agents. FA-SD uses successful rollouts from the current policy as privileged demonstrations to construct a feedback-augmented self-teacher and performs token-level KL-style distillation into the unconditioned student. Across seven open-domain QA benchmarks, the authors report that FA-SD does not sustain improvement: unclipped and PPO-style clipped variants show pass rates close to zero for most of training, with a transient rebound around step 180. They attribute this failure to 'decoding collapse,' a tendency to produce recurring reasoning-and-search templates that are input-question-agnostic, making the KL signal uninformative. They further decompose the instability into model inconsistency (drifting self-teacher) and prompt inconsistency (feedback conditioning), and show that a fixed reference or EMA teacher stabilizes training, with the EMA variant reaching 0.206 average EM versus 0.114 for the base model. They extend feedback augmentation to external-teacher MOPD and find it underperforms standard MOPD, which they interpret as additional evidence of prompt inconsistency.

Significance. If the empirical measurements are taken at face value, the paper provides a useful, honestly reported negative result on a timely question: whether self-distillation from privileged successful rollouts can provide dense supervision for long-horizon search agents. The multi-benchmark evaluation, the comparison against GRPO (0.348) and Qwen7B-Instruct (0.304) as strong upper reference points, and the systematic ablation of teacher regularization are concrete strengths. The EMA/fixed-reference comparison is a sensible diagnostic that supports the practical recommendation that self-teacher drift must be controlled. However, the paper's central causal claim—that decoding collapse is the mechanism behind FA-SD's failure—is not directly evidenced by the metrics presented. The effective sample ratio is definitionally coupled to the pass rate, and no trajectory-level diagnostic is provided. The scientific contribution would be stronger if the mechanism were treated as a hypothesis and the headline finding framed as a measured negative result with proposed explanations, rather than as an established failure mode.

major comments (4)
  1. [§4.4, Figures 1 and 3] The named mechanism 'decoding collapse' is asserted, not demonstrated. The evidence is (i) near-zero pass rates and (ii) a declining 'effective sample ratio' defined as the fraction of trajectories that provide a valid feedback signal—i.e., that come from a rollout group containing at least one successful rollout. This ratio is definitionally coupled to the pass rate, so it cannot independently diagnose template collapse. No metric directly tests input-question-agnostic template use: no question-conditioned trajectory similarity, no template entropy, no mutual information between question and generated tokens, and no example of a collapsed trajectory. The claim that such behavior 'can be missed by existing evaluation metrics' requires a metric that detects it; none is provided. This unmeasured interpretive layer underpins the causal chain in Section 4.7, so the main 'why' contribution re
  2. [Table 2, Eq. (4), Figure 1] Alternative explanations for the observed dynamics are not ruled out. FA-SD is trained for only 200 steps while external-teacher OPD gets 1000 steps; the single-sample reverse-KL estimator in Eq. (4) has high variance over long interleaved rollouts; and success-rollout selection via exact match is noisy in multi-hop QA. The transient rebound around step 180 and the failure of the student to track the self-teacher (Figure 2) are equally consistent with optimization instability, variance, or insufficient budget as with template collapse. To make decoding collapse load-bearing, the paper should include a control (larger budget, multi-sample KL estimator, or a direct trajectory-similarity metric) or explicitly reframe the mechanism as a hypothesis.
  3. [§4.5, Table 5, Title/Abstract] The EMA-regularized FA-SD improves the average from 0.114 to 0.206, which conflicts with the title's blanket claim that FA-SD 'fails to improve.' The paper qualifies this as 'does not sustain improvement across variants,' but the title and abstract overstate the negative result. If the intended claim is that FA-SD is dominated by GRPO (0.348) and Qwen7B-Instruct (0.304), that is a different and defensible claim, but it should be stated as such. This framing issue affects how readers interpret the paper's central contribution.
  4. [§4.6–4.7, Figure 5] The interpretation that FA-MOPD underperforms MOPD 'indicating that privileged feedback can introduce additional prompt inconsistency' is not supported by direct measurement. The degradation could be due to distribution shift in the teacher prompt, to conditioning on noisy successful rollouts, or to interaction between the feedback and the student's on-policy prefixes. Without ablating the feedback template (e.g., providing the same successful rollout to the student, or using oracle feedback), the decomposition into model versus prompt inconsistency remains speculative. Because this decomposition is part of the paper's explanation for FA-SD's failure, it needs additional empirical support.
minor comments (5)
  1. [Section 1] The text 'identify and characterizedecoding collapse' contains a missing space and should read 'identify and characterize decoding collapse.'
  2. [Tables 5–6 and Appendix C.4] The model is referred to inconsistently as 'Qwen2.5-3B-Base' and 'Qwen3B-Base' in the tables and text; unify the terminology.
  3. [Equation (3)] The divergence D is not specified when the loss is introduced. Since the gradient derivation in Eq. (4) assumes reverse KL, state this when defining Eq. (3) to avoid ambiguity for readers.
  4. [Section 4.4] The 'effective sample ratio' is described qualitatively but never given a formula. A precise definition (e.g., fraction of rollout groups with at least one successful trajectory) would clarify why it is correlated with pass rate and help the reader evaluate whether it can serve as an independent diagnostic.
  5. [Appendix C.4] The sentence 'We include this result to complement the training-dynamics analysis' appears in both C.3 and C.4 with identical wording; rephrase to avoid duplication, and provide a brief interpretation of the OPD-after-GRPO degradation rather than only reporting the numbers.

Circularity Check

0 steps flagged

No significant circularity: the paper's core claims are empirical measurements and its diagnostics, while under-supported, do not reduce to their inputs.

full rationale

The central result—FA-SD fails to sustain improvement across variants—is an empirical measurement on standard benchmarks (EM and pass rate) with no constants fitted to produce it. The FA-SD objective (Eq. 3) is defined from successful rollouts, but the final negative result and the improvement from EMA-regularized training (Table 5) are not predetermined by that definition; they are external, falsifiable observations. The paper's causal explanation (decoding collapse) is indeed under-demonstrated: Figures 1 and 3 mainly show pass-rate dynamics, and the effective sample ratio in Figure 3 is definitionally coupled to pass rate, as the paper itself acknowledges by stating that it 'closely tracks the training pass rate.' However, that metric is used as a diagnostic illustration of signal sparsity, not as an independent derivation of the collapse mechanism; it does not make the central claim circular. No load-bearing self-citation appears: the related works cited for self-distillation, reasoning collapse, and OPD are external to the present authors. The absence of a direct trajectory-similarity or question-dependence metric is a correctness/evidence gap, not a circularity under the specified criteria. Thus the derivation chain does not reduce to its inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 1 invented entities

The central claim rests on measured training and evaluation curves, not fitted constants; no free parameter is tuned to produce the headline numbers. The genuinely load-bearing unverified inputs are interpretive: the decoding-collapse mechanism, the two-way inconsistency decomposition, and the validity of the RDS-style token-level signal in a retrieval-interleaved environment. The EMA momentum and 200-step budget are hand-chosen and unswept. One invented construct (decoding collapse) is introduced without an independent detector, so its falsifiability is currently internal to the paper's narrative.

free parameters (2)
  • EMA momentum alpha = 0.95
    Hand-set in Section 3.3 (unless otherwise specified, alpha = 0.95); the central stabilization claim is demonstrated only at this value with no sensitivity sweep.
  • Self-teacher training budget = 200 max steps
    Table 2: max steps 200 for self-teacher versus 1000 for external teacher; the claim that FA-SD does not sustain improvement is measured inside this budget, and the asymmetry confounds cross-method stability comparisons.
axioms (4)
  • domain assumption Token-level reverse-KL self-teacher gradients (Equations 3-4, carried from RDS/Hübotter et al. 2026) provide a usable dense supervision signal when retrieved observations are interleaved and masked
    Section 3.2.2. The paper itself questions this premise in Section 1, noting that matching token-level distributions may be insufficient to transfer the long-horizon search decisions required for effective retrieval.
  • ad hoc to paper Near-zero pass rates and a declining effective sample ratio imply template-collapse (decoding collapse) rather than optimization instability, high-variance gradients, or reward/filter noise
    Section 4.4. The mechanism is inferred from Figures 1-3; no direct measurement of input-question-agnostic template recurrence is provided, so the diagnosis depends on this interpretation.
  • domain assumption A correct trajectory from another attempt in the same rollout group is valid privileged feedback for the self-teacher
    Section 3.2.2 FA-SD construction. The FA-MOPD result in Section 4.6 partially undermines this premise when the teacher is already strong.
  • domain assumption Qwen2.5-3B-Base with 2018 Wikipedia and E5 top-3 retrieval is a representative testbed for retrieval-interleaved search agents
    Section 4.1, borrowed from prior work (Jin et al. 2025; Li et al. 2026); generality across student scales, retrievers, and corpora is untested.
invented entities (1)
  • Decoding collapse (named failure mode) no independent evidence
    purpose: Explains why FA-SD fails: recurring reasoning-and-search templates produce diverse-looking but question-agnostic trajectories that make the KL signal uninformative; motivates the EMA and fixed-reference fixes
    The paper admits such behavior can be missed by existing evaluation metrics (Section 4.4) yet ships no metric, detector, or quantitative definition that catches it, so the construct has no falsifiable handle outside the paper's own interpretation.

pith-pipeline@v1.3.0-alltime-deepseek · 13219 in / 23492 out tokens · 190207 ms · 2026-08-01T17:38:52.300301+00:00 · methodology

0 comments
read the original abstract

On-policy self-distillation (OPSD) offers a promising approach for training large language models without relying on a separate teacher model. However, its effectiveness on complex agentic tasks remains largely unexplored. In this work, we instantiate Feedback-Augmented Self-Distillation (FA-SD), a self-distillation algorithm for agentic search that leverages successful demonstrations as privileged information. We identify that models can rely on recurring reasoning-and-search output templates, producing trajectories that appear diverse but are largely agnostic to the input question, making the KL-based self-distillation signal uninformative. We term this phenomenon decoding collapse, a failure mode that can be missed by existing evaluation metrics. To understand its underlying cause, we show that although the self-teacher achieves stronger performance, learning remains inherently unstable due to inconsistent supervision signals. We further decompose this inconsistency into model inconsistency and prompt inconsistency, and show that the latter can significantly degrade the quality of the supervision signal, limiting the effectiveness of self-teacher learning. To mitigate this inconsistency, we introduce an exponential moving average (EMA) teacher to stabilize the self-teacher and provide more consistent supervision signals. Although the EMA teacher requires a warm-up phase during which performance may temporarily regress, it ultimately improves model performance by providing more stable supervision.

Figures

Figures reproduced from arXiv: 2607.17558 by Fan Yang, Rui Meng, Yuxin Wen.

Figure 1
Figure 1. Figure 1: Training pass-rate dynamics for unclipped and PPO-style clipped FA-SD variants over all sampled rollouts [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Evolution of inference-time accuracy of the unconditioned student and feedback-augmented self-teacher [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Effective distillation-signal ratio for different [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training pass-rate dynamics of FA-SD with different self-teacher regularization strategies. The baseline FA [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Training pass-rate dynamics for MOPD and Feedback-Augmented MOPD. Standard MOPD improves [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed training dynamics for the unclipped [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Detailed training dynamics for the PPO-style [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

30 extracted references · 3 canonical work pages

  1. [1]

    2026 , eprint=

    RAGEN-2: Reasoning Collapse in Agentic RL , author=. 2026 , eprint=

    2026

  2. [2]

    arXiv preprint arXiv:2605.12652 , year=

    Multi-Rollout On-Policy Distillation via Peer Successes and Failures , author=. arXiv preprint arXiv:2605.12652 , year=

    Pith/arXiv arXiv

  3. [3]

    Advances in Neural Information Processing Systems , volume =

    Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , author =. Advances in Neural Information Processing Systems , volume =

  4. [4]

    arXiv preprint arXiv:2603.24472 , year =

    Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs? , author =. arXiv preprint arXiv:2603.24472 , year =. doi:10.48550/arXiv.2603.24472 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2603.24472

  5. [5]

    arXiv preprint arXiv:2603.25562 , year =

    Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes , author =. arXiv preprint arXiv:2603.25562 , year =. doi:10.48550/arXiv.2603.25562 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2603.25562

  6. [6]

    arXiv preprint arXiv:2306.13649 , year =

    On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes , author =. arXiv preprint arXiv:2306.13649 , year =. doi:10.48550/arXiv.2306.13649 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.13649

  7. [7]

    arXiv preprint arXiv:2603.09203 , year =

    Evaluate-as-Action: Self-Evaluated Process Rewards for Retrieval-Augmented Agents , author =. arXiv preprint arXiv:2603.09203 , year =

    arXiv

  8. [8]

    Li, Jian and Jin, Yizhang and Liu, Dongqi and Ding, Hang and Wu, Jiafu and Chen, Dongsheng and Shen, Yunhang and Qin, Yulei and Tai, Ying and Wang, Chengjie and Yuan, Xiaotong and Wang, Yabiao , journal =

  9. [9]

    arXiv preprint arXiv:2212.03533 , year =

    Text Embeddings by Weakly-Supervised Contrastive Pre-training , author =. arXiv preprint arXiv:2212.03533 , year =

    Pith/arXiv arXiv

  10. [10]

    From Knowledge to Noise:

    Lindenbauer, Tobias and Groh, Georg and Schuetze, Hinrich , booktitle =. From Knowledge to Noise:. 2025 , address =. doi:10.18653/v1/2025.realm-1.30 , pages =

    doi:10.18653/v1/2025.realm-1.30 2025

  11. [11]

    Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025) , month = jul, year =

    A Study on Leveraging Search and Self-Feedback for Agent Reasoning , author =. Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025) , month = jul, year =. doi:10.18653/v1/2025.realm-1.18 , pages =

    work page doi:10.18653/v1/2025.realm-1.18 2025

  12. [12]

    arXiv preprint arXiv:2601.18734 , year =

    Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models , author =. arXiv preprint arXiv:2601.18734 , year =. doi:10.48550/arXiv.2601.18734 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2601.18734

  13. [13]

    arXiv preprint arXiv:2601.20802 , year =

    Reinforcement Learning via Self-Distillation , author =. arXiv preprint arXiv:2601.20802 , year =. doi:10.48550/arXiv.2601.20802 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2601.20802

  14. [14]

    Aligning Sizes of Intermediate Layers by

    Suzuki, Takeshi and Yamada, Hiroaki and Tokunaga, Takenobu , booktitle =. Aligning Sizes of Intermediate Layers by. 2025 , address =. doi:10.18653/v1/2025.insights-1.10 , pages =

    work page doi:10.18653/v1/2025.insights-1.10 2025

  15. [15]

    Proceedings of the Third Workshop on Insights from Negative Results in NLP , month = may, year =

    Can Question Rewriting Help Conversational Question Answering? , author =. Proceedings of the Third Workshop on Insights from Negative Results in NLP , month = may, year =. doi:10.18653/v1/2022.insights-1.13 , pages =

    work page doi:10.18653/v1/2022.insights-1.13 2022

  16. [16]

    2025 , doi =

    Jin, Bowen and Zeng, Hansi and Yue, Zhenrui and Yoon, Jinsung and Arik, Sercan and Wang, Dong and Zamani, Hamed and Han, Jiawei , journal =. 2025 , doi =

    2025

  17. [17]

    2025 , doi =

    Jiang, Pengcheng and Lin, Jiacheng and Cao, Lang and Tian, Runchu and Kang, SeongKu and Wang, Zifeng and Sun, Jimeng and Han, Jiawei , journal =. 2025 , doi =

    2025

  18. [18]

    2025 , address =

    Jiang, Pengcheng and Xu, Xueqiang and Lin, Jiacheng and Xiao, Jinfeng and Wang, Zifeng and Sun, Jimeng and Han, Jiawei , booktitle =. 2025 , address =. doi:10.18653/v1/2025.emnlp-main.1095 , pages =

    doi:10.18653/v1/2025.emnlp-main.1095 2025

  19. [19]

    arXiv preprint arXiv:2602.21320 , year =

    Acikgoz, Emre Can and Qian, Cheng and H. arXiv preprint arXiv:2602.21320 , year =. doi:10.48550/arXiv.2602.21320 , url =

    doi:10.48550/arxiv.2602.21320

  20. [20]

    2026 , doi =

    Wang, Prince Zizhuang and Jiang, Shuli , journal =. 2026 , doi =

    2026

  21. [21]

    arXiv preprint arXiv:2507.21046 , year =

    A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence , author =. arXiv preprint arXiv:2507.21046 , year =. doi:10.48550/arXiv.2507.21046 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2507.21046

  22. [22]

    2025 , doi =

    Xia, Peng and Zeng, Kaide and Liu, Jiaqi and Qin, Can and Wu, Fang and Zhou, Yiyang and Xiong, Caiming and Yao, Huaxiu , journal =. 2025 , doi =

    2025

  23. [23]

    2025 , doi =

    Lin, Jiaye and Guo, Yifu and Han, Yuzhen and Hu, Sen and Ni, Ziyi and Wang, Licheng and Chen, Mingguang and Liu, Hongzhang and Chen, Ronghao and He, Yangfan and Jiang, Daxin and Jiao, Binxing and Hu, Chen and Wang, Huacan , journal =. 2025 , doi =

    2025

  24. [24]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  25. [25]

    Publications Manual , year = "1983", publisher =

    1983

  26. [26]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    arXiv 1981

  27. [27]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  28. [28]

    Dan Gusfield , title =. 1997

    1997

  29. [29]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  30. [30]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =