Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Interlocking-free Selective Rationalization Through Genetic-based Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read GenSPP eliminates interlocking in select-then-predict rationalization by evolving the generator with a genetic algorithm and training a fresh predictor for every candidate, so no module can dominate the other during training.

desk verdict Novel genetic-search rationalization idea, but Eq. 8 inverts the fitness reward direction, undermining the central claim; the paper still deserves review because the idea and artifacts are concrete. read the letter →

arxiv 2412.10312 v2 pith:UW3ZBJDM submitted 2024-12-13 cs.LG cs.AIcs.CLcs.NE

Federico Ruggeri , Gaetano Signorelli This is my paper · ORCID
classification cs.LGcs.AIcs.CLcs.NE
keywords selectiverationalizationinterlockinggeneticalgorithmneuroevolutionrationaleextractionexplainableNLPhatespeechdetection
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

Select-then-predict rationalization couples a generator that picks highlight tokens with a predictor that classifies the masked input; when trained jointly, the two modules can lock into a suboptimal equilibrium called interlocking. This paper proposes GenSPP, a framework that claims to be the first interlocking-free solution that needs no heuristics, weight sharing, or auxiliary regularizers: the generator is evolved by a genetic algorithm, and every candidate generator is evaluated by training a fresh predictor from scratch on its masked inputs. The evaluation is driven by a fitness function that requires masked-input classification to stay within a tolerance of an unmasked-input predictor's loss while minimizing sparsity and contiguity penalties. On a controlled synthetic dataset and on HateXplain, GenSPP reports significantly higher highlight F1 than four SPP baselines (76.0 versus 65.7 on Toy, 42.6 versus 36.2 on HateXplain), comparable classification F1, lower selection ratios, and lower variance across seeds. The paper also shows GenSPP recovers when the initial population is deliberately skewed toward a degenerate generator.

What carries the argument

The carrying mechanism is the Eq. (6) reformulation of select-then-predict as a constrained generator-only optimization, solved by genetic global search. Each individual is a full set of generator parameters; roulette-wheel selection, one-point crossover, Gaussian mutation, and half-elitism survival produce new candidate generators, and the fitness function of Eqs. (7)-(8) gives nonzero score only when masked-input classification loss falls below the hand-set tolerance $l+\epsilon$ and then rewards low sparsity and contiguity regularization. Disjointness is enforced by training a fresh predictor from scratch for every individual with the generator frozen.

What would settle it

On HateXplain, set the tolerance $l+\epsilon=0.8$ so that an empty mask predicting the majority class (cross-entropy near 0.64) satisfies the constraint, then run the full GenSPP search; if high-fitness individuals select zero tokens on most examples while test highlight F1 collapses toward chance, the fitness criterion is insufficient to enforce meaningful highlights.

Watch

Extended reading notes

Core claim

GenSPP's central discovery is that interlocking disappears once the generator and predictor are never co-adapted: the joint minimization $\min_\theta \min_\omega L(f_\omega(g_\theta(x)\odot x),y)$ is replaced by the constrained problem $\min_\theta \Omega(m)$ subject to $\min_\omega L(f_\omega(g_\theta(x)\odot x),y)\le l+\epsilon$, where $l$ is the loss of a predictor trained on the unmasked input. A genetic population of generator parameter vectors is searched globally, and each individual is scored by a fitness $\tilde h$ that combines normalized classification loss and mask regularization; predictors are re-initialized and trained from scratch for each evaluation. The author's claim is that this makes interlocking structurally impossible rather than merely unlikely, because the predictor cannot overfit to a stale mask and the generator is not updated through predictor gradients. The reported results support the claim on two benchmarks by showing better highlight quality and recovery from a deliberately skewed generator initialization.

Load-bearing premise

The framework assumes that matching the loss of an unmasked-input predictor within a hand-set tolerance is the right test of a good highlight; if the tolerance is loose enough, a degenerate mask such as selecting almost nothing can satisfy the constraint and receive high fitness, decoupling the search from genuine rationalization.

Editorial extensions

If this is right

  • If GenSPP is correct, interlocking is not a pathology that must be smoothed, guided, or regularized away; it is an artifact of co-adaptation, and disjoint training eliminates it at the architectural level.
  • Rationalization models no longer need a sparsity threshold $\alpha$ or surrogate differentiable losses; the fitness function can directly encode sparsity and contiguity preferences, removing dataset-specific hyperparameter tuning.
  • The method inherits a robustness property: a deliberately bad generator placed in the initial population does not trap the search, so recovery from interlocking is a population-level property rather than a gradient-escape event.
  • The cost of these properties is computational: a single GenSPP seed run takes roughly 36 minutes on Toy and 78 minutes on HateXplain versus roughly 8 and 4 minutes for baselines, so gains in highlight quality are bought with search budget.

Reading between the lines

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

  • GenSPP's disjointness argument is not tied to genetic algorithms specifically; any non-differentiable global search, such as evolutionary strategies or random search with enough budget, should exhibit the same interlocking-free property, so the core claim is testable by swapping the search operator.
  • The hand-set tolerance $l+\epsilon$ is the main hidden dial: the paper fixes it at 0.1 (Toy) and 0.6 (HateXplain) with no sensitivity analysis, so an immediate extension is to measure how highlight quality degrades as the tolerance widens toward the loss of a trivial predictor.
  • On longer inputs with multiple rationales, selective rationalization typically moves to sentence-level selection; applying GenSPP there would test whether the genetic search overhead and contiguity penalty scale, and whether the disjointness advantage survives when the mask space is much larger.
  • The reported ability to select no tokens on negative examples in HateXplain suggests GenSPP can learn per-example sparsity rather than a global ratio; a testable extension is whether this emerges only because the fitness threshold is loose enough to let empty masks pass on non-hate posts.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript proposes GenSPP, a selective rationalization framework that replaces the end-to-end select-then-predict objective with a disjoint two-stage optimization solved by a genetic algorithm. Each individual in the population encodes a generator, is evaluated by training a fresh predictor on the masked inputs, and receives a fitness combining classification loss and highlight sparsity/contiguity. The authors claim GenSPP is the first interlocking-free architecture that requires no heuristics or additional regularizers, and report superior rationale quality with comparable classification accuracy on a new synthetic toy dataset and HateXplain. The paper also introduces a synthetic benchmark and provides code/data.

Significance. The idea of breaking interlocking by genetic search over generator parameters is novel, and the paper is commendable for releasing code and data, constructing a controlled synthetic dataset, and reporting five-seed runs with Wilcoxon significance tests. If the proposed fitness function were correct, the method would be a nontrivial advance over existing mitigation-based approaches. However, the central optimization objective as written is inverted, so the empirical claims are not currently supported; the significance of the contribution therefore depends on whether the objective can be corrected and the experiments redone.

major comments (3)
  1. [Section 5.2, Eq. (8)] The fitness function h in Eq. (8) is defined as h = 1/(tilde_h + hat_epsilon). Because Eq. (7) sets tilde_h = 0 whenever Lt >= l + epsilon, an individual whose predictor violates the classification constraint receives the maximum possible fitness h = 1/hat_epsilon, while an individual satisfying the constraint receives the smaller value h = 1/(1 - L + hat_epsilon). Since the genetic algorithm maximizes h (roulette-wheel selection and half-elitism, Section 5.3), the search is actively driven toward generators whose predictors fail the task. This reverses the intended effect of the constraint in Eq. (6), and consequently the results in Tables 1 and 2 may be artifacts of rewarding constraint violation rather than evidence for the proposed method.
  2. [Section 4, Eq. (6)] The reformulation in Eq. (6) is presented as being 'equivalent' to Eq. (4), but no derivation or formal statement is given. Eq. (4) minimizes the classification loss alone, while Eq. (6) minimizes Omega(m) subject to a loss bound l + epsilon. The equivalence would require that the set of masks with loss below l + epsilon coincides with the relevant minimizers of Eq. (4), and that l = L(f_{omega*}(x), y) is the right benchmark; neither is established. The threshold l + epsilon is hand-set to 0.1 (Toy) and 0.6 (HateXplain) with no sensitivity analysis, so a near-empty mask may satisfy the constraint vacuously. This affects the central claim that GenSPP optimizes the disjoint formulation and hence that it is interlocking-free in the intended sense.
  3. [Section 7, Table 1] Table 1 reports that GenSPP selects on average 0.75 tokens on HateXplain (selection rate 6.51%), whereas baselines select about 3 tokens. The paper interprets this as a learned preference to select no highlight for negative examples, but with the inverted fitness this is also consistent with degenerate near-empty selections that still achieve Clf-F1 around 69.7. The manuscript does not report the distribution of selection sizes, the number of examples with zero selected tokens, or the held-out Lt relative to the threshold l + epsilon. Without such evidence, the high Hl-F1 and comparable Clf-F1 do not demonstrate that meaningful rationales are learned, rather than a trivial bias exploited through empty inputs.
minor comments (5)
  1. [Abstract and Section 8] The phrase 'does not require any learning overhead' is misleading given the roughly 5-10x wall-clock increase reported in Section 8 and Table 9; clarify that it refers to additional modules or regularizers, not computational cost.
  2. [Section 4, Eq. (5)] omega* is not defined; specify that it is the predictor trained on the unmasked input and describe how it is obtained in practice.
  3. [Section 2.1, Eq. (2)] The summation for Ls runs from i=0 while Lc runs from i=1; align the indexing with the token definition x_1..x_n.
  4. [Appendix A.2] The number of predictor training epochs during evaluation is set to 3 without sensitivity analysis; since fitness scores depend critically on this choice, a small study of its effect would strengthen the paper.
  5. [Section 7] The Wilcoxon test is reported only against the 'best baseline'; state whether multiple-comparison correction was applied and report the exact test statistic or p-value range.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; GenSPP's derivation is self-contained and its empirical evaluation uses external held-out labels and ground-truth highlights.

full rationale

The paper's central claims are not circular. The 'interlocking-free' property is asserted as a design consequence of disjoint training of the generator and predictor, not derived from a fitted parameter or from the paper's own output. The reformulation in Eq. 6 is a stated optimization surrogate: it defines l as the loss of a predictor trained on the unmasked input and then constrains the masked-input loss to be within a hand-set tolerance l+epsilon. This is a hyperparameter choice, not a fitted quantity renamed as a prediction, and no experimental result is claimed to be a prediction of l or of the tolerance. The reported comparisons in Tables 1 and 2 use external held-out test labels and human ground-truth highlights, so the headline improvements in Hl-F1 and Clf-F1 are not forced by construction. The paper contains no load-bearing self-citations: the cited interlocking analysis is due to Yu et al. (2021), and no uniqueness theorem or prior result by the same authors is invoked to rule out alternatives. The genetic-algorithm objective in Eqs. 7-8 is an internal training objective; even if the apparent inversion of the fitness direction were a real flaw, it would be a correctness or optimization error, not a circularity, because it does not reduce a reported result to its own input. The Limitations section acknowledges computational overhead, dataset scope, and lack of parallel evaluation, but none of these admissions indicates that a derivation reduces to a fit or to a self-citation chain. Overall, the derivation chain is self-contained and externally evaluated, so the appropriate circularity score is 0.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The method rests on assumptions about the equivalence of the constrained reformulation, the reliability of 3-epoch predictor training for fitness, and the global-search power of the GA. The hand-set genetic hyperparameters and evaluation threshold are free parameters; no new entities are introduced.

free parameters (9)
  • evaluation tolerance l + epsilon = 0.1 (Toy), 0.6 (HateXplain)
    Hand-set threshold in Eq. 7 and Eq. 8 that decides whether an individual's Lt is acceptable; controls the trade-off between classification and rationalization and is dataset-specific.
  • population size I = 50
    Number of generator individuals in the genetic search; larger I increases search coverage but also compute cost.
  • generations G = 100 (150 in skew experiment)
    Number of genetic search iterations; Table 2 shows results improve with G=150, so the headline numbers depend on this budget.
  • mutation probability pm = 1.0
    Every generated individual is mutated; a genetic algorithm hyperparameter with direct effect on local exploration.
  • crossover probability pc = 1.0
    All selected pairs undergo one-point crossover; affects global exploration.
  • selection and survival probabilities psl, psu = 0.5
    Controls roulette-wheel selection and half-elitism composition of the next population.
  • mutation noise standard deviation = 0.05
    Gaussian noise added to generator parameters during mutation; determines the step size of local search.
  • predictor training epochs during evaluation = 3
    Each individual's predictor is trained only 3 epochs; this is a major approximation of Lt and directly affects fitness.
  • predictor learning rate during evaluation = 1e-2
    Used with Adam to train f_omega for 3 epochs per individual; contributes to the approximate Lt estimate.
assumptions (5)
  • ad hoc to paper The constrained reformulation Eq. 6 is equivalent to the original joint minimization Eq. 4, so any solution of Eq. 6 is a near-optimal solution of Eq. 4 without interlocking.
    Stated in Section 4 as 'This formulation is equivalent...' without proof; the equivalence hinges on l being the loss of an unconstrained predictor on x, but Eq. 6 does not guarantee the selected mask is the ground-truth highlight.
  • domain assumption Genetic search over raw parameter vectors with one-point crossover and Gaussian mutation explores the generator parameter space well enough to find high-quality masks without gradient information.
    Relies on general GA convergence properties cited from Katoch et al. 2021 and Whitelam et al. 2021; no proof is given for this specific discrete-mask objective.
  • ad hoc to paper Training a predictor from scratch for 3 epochs yields an Lt estimate accurate enough for fitness evaluation and threshold decisions.
    The per-individual evaluation budget is fixed at 3 epochs in Appendix A.2; if Lt is far from converged, the fitness threshold and individual ranking are unreliable.
  • domain assumption Performing disjoint training makes interlocking impossible by construction, because only one module is trained at a time.
    Section 5.4 'Disjoint Training' argues that f_omega depends on g_theta and not vice versa; this avoids the feedback loop but does not prove global optimality or the absence of other suboptimal equilibria.
  • ad hoc to paper The objectives Lt and Omega(m) in Eq. 7 are normalized to comparable scales, so their product behaves as a balanced fitness.
    The paper states 'learning objectives are normalized and equally important' but never defines the normalization; raw lambda_s and lambda_c remain dataset-specific hyperparameters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interlocking-free Selective Rationalization Through Genetic-based Learning." pith.science (2026). https://pith.science/paper/UW3ZBJDM

@misc{pith2026241210312,
  author       = {Pith},
  title        = {Pith review of: Interlocking-free Selective Rationalization Through Genetic-based Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UW3ZBJDM}},
  note         = {Machine review of arXiv:2412.10312}
}
read the original abstract

A popular end-to-end architecture for selective rationalization is the select-then-predict pipeline, comprising a generator to extract highlights fed to a predictor. Such a cooperative system suffers from suboptimal equilibrium minima due to the dominance of one of the two modules, a phenomenon known as interlocking. While several contributions aimed at addressing interlocking, they only mitigate its effect, often by introducing feature-based heuristics, sampling, and ad-hoc regularizations. We present GenSPP, the first interlocking-free architecture for selective rationalization that does not require any learning overhead, as the above-mentioned. GenSPP avoids interlocking by performing disjoint training of the generator and predictor via genetic global search. Experiments on a synthetic and a real-world benchmark show that our model outperforms several state-of-the-art competitors.

Figures

Figures reproduced from arXiv: 2412.10312 by the authors.

Figure 1
Figure 1. Loss landscape comparison between our fit [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Number of contiguous highlights (i.e., con [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 26 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Bushra Alhijawi and Arafat Awajan. 2024. https://doi.org/10.1007/S12065-023-00822-6 Genetic algorithms: theory, genetic operators, solutions, and applications . Evol. Intell., 17(3):1245--1256

    work page doi:10.1007/s12065-023-00822-6 2024

  4. [4]

    Diego Antognini and Boi Faltings. 2021. https://doi.org/10.18653/V1/2021.FINDINGS-ACL.68 Rationalization through concepts . In Findings of the Association for Computational Linguistics: ACL/IJCNLP 2021, Online Event, August 1-6, 2021 , volume ACL/IJCNLP 2021 of Findings of ACL , pages 761--775. Association for Computational Linguistics

    work page doi:10.18653/v1/2021.findings-acl.68 2021

  5. [5]

    Diego Antognini, Claudiu Musat, and Boi Faltings. 2021. https://doi.org/10.1609/AAAI.V35I14.17483 Multi-dimensional explanation of target variables from documents . In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educa...

    work page doi:10.1609/aaai.v35i14.17483 2021

  6. [6]

    Lei Jimmy Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. 2016. https://arxiv.org/abs/1607.06450 Layer normalization . CoRR, abs/1607.06450

    arXiv 2016

  7. [7]

    Yujia Bao, Shiyu Chang, Mo Yu, and Regina Barzilay. 2018. https://doi.org/10.18653/v1/D18-1216 Deriving machine attention from human rationales . In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1903--1913, Brussels, Belgium. Association for Computational Linguistics

    work page doi:10.18653/v1/d18-1216 2018

  8. [8]

    Jasmijn Bastings, Wilker Aziz, and Ivan Titov. 2019. https://doi.org/10.18653/v1/P19-1284 Interpretable neural predictions with differentiable binary variables . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2963--2977, Florence, Italy. Association for Computational Linguistics

    doi:10.18653/v1/p19-1284 2019

Show all 53 references
  1. [9]

    Goldberg

    Erick Cantú-Paz and David E. Goldberg. 2000. https://doi.org/10.1016/S0045-7825(99)00385-0 Efficient parallel genetic algorithms: theory and practice . Computer Methods in Applied Mechanics and Engineering, 186(2):221--238

    2000 doi

  2. [10]

    Jaakkola

    Shiyu Chang, Yang Zhang, Mo Yu, and Tommi S. Jaakkola. 2019. https://proceedings.neurips.cc/paper/2019/hash/5ad742cd15633b26fdce1b80f7b39f7c-Abstract.html A game theoretic approach to class-wise selective rationalization . In Advances in Neural Information Processing Systems 3...

    2019

  3. [11]

    Jaakkola

    Shiyu Chang, Yang Zhang, Mo Yu, and Tommi S. Jaakkola. 2020. http://proceedings.mlr.press/v119/chang20c.html Invariant rationalization . In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event , volume 119 of Proceedin...

    2020

  4. [12]

    Stanley, and Jeff Clune

    Edoardo Conti, Vashisht Madhavan, Felipe Petroski Such, Joel Lehman, Kenneth O. Stanley, and Jeff Clune. 2018. https://proceedings.neurips.cc/paper/2018/hash/b1301141feffabac455e1f90a7de2054-Abstract.html Improving exploration in evolution strategies for deep reinforcement lea...

    2018

  5. [13]

    Jay DeYoung, Sarthak Jain, Nazneen Fatema Rajani, Eric Lehman, Caiming Xiong, Richard Socher, and Byron C. Wallace. 2020. https://doi.org/10.18653/v1/2020.acl-main.408 ERASER : A benchmark to evaluate rationalized NLP models . In Proceedings of the 58th Annual Meeting of the A...

    2020 doi

  6. [14]

    William Falcon and The PyTorch Lightning team . 2019. https://doi.org/10.5281/zenodo.3828935 PyTorch Lightning

    2019 doi

  7. [15]

    Edgar Galv \' a n and Peter Mooney. 2021. https://doi.org/10.1109/TAI.2021.3067574 Neuroevolution in deep neural networks: Current trends and future challenges . IEEE Trans. Artif. Intell. , 2(6):476--493

    2021

  8. [16]

    Nikolaus Hansen and Andreas Ostermeier. 2001. https://doi.org/10.1162/106365601750190398 Completely derandomized self-adaptation in evolution strategies . Evolutionary Computation

    2001 doi

  9. [17]

    Elize Herrewijnen, Dong Nguyen, Floris Bex, and Kees van Deemter. 2024. https://doi.org/10.3389/FRAI.2024.1260952 Human-annotated rationales and explainable text classification: a survey . Frontiers Artif. Intell., 7

    2024

  10. [18]

    Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dryden, and Alexandra Peste. 2021. http://jmlr.org/papers/v22/21-0366.html Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks . Journal of Machine Learning Research, 22(241):1--124

    2021

  11. [19]

    Shuaibo Hu and Kui Yu. 2024. https://doi.org/10.1609/AAAI.V38I16.29783 Learning robust rationales for model explainability: A guidance-based approach . In Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative Applications of...

    2024 doi

  12. [20]

    Xuming Hu, Zhijiang Guo, GuanYu Wu, Aiwei Liu, Lijie Wen, and Philip Yu. 2022. https://doi.org/10.18653/v1/2022.naacl-main.246 CHEF : A pilot C hinese dataset for evidence-based fact-checking . In Proceedings of the 2022 Conference of the North American Chapter of the Associat...

    2022 doi

  13. [21]

    Yongfeng Huang, Yujun Chen, Yulun Du, and Zhilin Yang. 2021. https://doi.org/10.1609/AAAI.V35I14.17547 Distribution matching for rationalization . In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artif...

    2021 doi

  14. [22]

    Sarthak Jain, Sarah Wiegreffe, Yuval Pinter, and Byron C. Wallace. 2020. https://doi.org/10.18653/v1/2020.acl-main.409 L earning to faithfully rationalize by construction . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4459--...

    2020 doi

  15. [23]

    Eric Jang, Shixiang Gu, and Ben Poole. 2017. Categorical reparameterization with gumbel-softmax. In ICLR

    2017

  16. [25]

    Sourabh Katoch, Sumit Singh Chauhan, and Vijay Kumar. 2021 b . https://doi.org/10.1007/S11042-020-10139-6 A review on genetic algorithm: past, present, and future . Multim. Tools Appl., 80(5):8091--8126

    2021 doi

  17. [26]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2015. http://arxiv.org/abs/1412.6980 Adam: A method for stochastic optimization . In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings

    2015 arXiv

  18. [27]

    Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2016. https://doi.org/10.18653/v1/D16-1011 Rationalizing neural predictions . In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 107--117, Austin, Texas. Association for Computational L...

    2016 doi

  19. [28]

    Shuangqi Li, Diego Antognini, and Boi Faltings. 2022. https://doi.org/10.48550/ARXIV.2205.06756 Interlock-free multi-aspect rationalization for text classification . CoRR, abs/2205.06756

    2022 doi arXiv

  20. [29]

    Adam Lipowski and Dorota Lipowska. 2012. https://doi.org/10.1016/j.physa.2011.12.004 Roulette-wheel selection via stochastic acceptance . Physica A: Statistical Mechanics and its Applications, 391(6):2193--2196

    2012 doi

  21. [30]

    Wei Liu, Zhiying Deng, Zhongyu Niu, Jun Wang, Haozhao Wang, YuanKai Zhang, and Ruixuan Li. 2024. https://proceedings.neurips.cc/paper_files/paper/2024/file/d53d51e88d92d3723755f6d425bc513b-Paper-Conference.pdf Is the mmi criterion necessary for interpretability? degenerating n...

    2024

  22. [31]

    Wei Liu, Haozhao Wang, Jun Wang, Zhiying Deng, Yuankai Zhang, Cheng Wang, and Ruixuan Li. 2023 a . https://doi.org/10.48550/ARXIV.2312.04103 Enhancing the rationale-input alignment for self-explaining rationalization . CoRR, abs/2312.04103

    2023 doi arXiv

  23. [32]

    Wei Liu, Haozhao Wang, Jun Wang, Ruixuan Li, Xinyang Li, Yuankai Zhang, and Yang Qiu. 2023 b . https://doi.org/10.18653/V1/2023.ACL-LONG.715 MGR: multi-generator based rationalization . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics ...

    2023 doi

  24. [33]

    Wei Liu, Haozhao Wang, Jun Wang, Ruixuan Li, Chao Yue, and Yuankai Zhang. 2022. http://papers.nips.cc/paper\_files/paper/2022/hash/2e0bd92a1d3600d4288df51ac5e6be5f-Abstract-Conference.html FR: folded rationalization with a unified encoder . In Advances in Neural Information Pr...

    2022

  25. [34]

    Wei Liu, Jun Wang, Haozhao Wang, Ruixuan Li, Zhiying Deng, Yuankai Zhang, and Yang Qiu. 2023 c . http://papers.nips.cc/paper\_files/paper/2023/hash/87e82678c0d6e5b729398426f82e9af6-Abstract-Conference.html D-separation for causal self-explanation . In Advances in Neural Inform...

    2023

  26. [35]

    Wei Liu, Jun Wang, Haozhao Wang, Ruixuan Li, Yang Qiu, Yuankai Zhang, Jie Han, and Yixiong Zou. 2023 d . https://doi.org/10.1145/3580305.3599299 Decoupled rationalization with asymmetric learning rates: A flexible lipschitz restraint . In Proceedings of the 29th ACM SIGKDD Con...

    2023

  27. [36]

    Binny Mathew, Punyajoy Saha, Seid Muhie Yimam, Chris Biemann, Pawan Goyal, and Animesh Mukherjee. 2021. https://doi.org/10.1609/AAAI.V35I17.17745 Hatexplain: A benchmark dataset for explainable hate speech detection . In Thirty-Fifth AAAI Conference on Artificial Intelligence,...

    2021 doi

  28. [37]

    Julian McAuley, Jure Leskovec, and Dan Jurafsky. 2012. https://doi.org/10.1109/ICDM.2012.110 Learning attitudes and attributes from multi-aspect reviews . In 2012 IEEE 12th International Conference on Data Mining, pages 1020--1025

    2012 doi

  29. [38]

    Zbigniew Michalewicz. 1996. https://doi.org/10.1007/978-3-662-03315-9 Genetic Algorithms + Data Structures = Evolution Programs, Third Revised and Extended Edition . Springer

    1996 doi

  30. [39]

    Risto Miikkulainen, Jason Liang, Elliot Meyerson, Aditya Rawal, Daniel Fink, Olivier Francon, Bala Raju, Hormoz Shahrzad, Arshak Navruzyan, Nigel Duffy, and Babak Hodjat. 2019. https://doi.org/10.1016/B978-0-12-815480-9.00015-3 Chapter 15 - evolving deep neural networks . In R...

    2019 doi

  31. [40]

    Bhargavi Paranjape, Mandar Joshi, John Thickstun, Hannaneh Hajishirzi, and Luke Zettlemoyer. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.153 An information bottleneck approach for controlling conciseness in rationale extraction . In Proceedings of the 2020 Conference on ...

    2020 doi

  32. [41]

    Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K \" o pf, Edward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Stei...

    2019

  33. [42]

    Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. https://doi.org/10.3115/V1/D14-1162 Glove: Global vectors for word representation . In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, EMNLP 2014, October 25-29, 2014,...

    2014 doi

  34. [43]

    Riccardo Poli and W. B. Langdon. 1998. Genetic programming with one-point crossover. In Soft Computing in Engineering Design and Manufacturing, pages 180--189, London. Springer London

    1998

  35. [44]

    why should I trust you?

    Marco T \' u lio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. https://doi.org/10.1145/2939672.2939778 "why should I trust you?": Explaining the predictions of any classifier . In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data M...

    2016

  36. [45]

    Tim Salimans, Jonathan Ho, Xi Chen, and Ilya Sutskever. 2017. https://arxiv.org/abs/1703.03864 Evolution strategies as a scalable alternative to reinforcement learning . CoRR, abs/1703.03864

    2017 arXiv

  37. [46]

    Maarten Sap, Swabha Swayamdipta, Laura Vianna, Xuhui Zhou, Yejin Choi, and Noah A. Smith. 2022. https://doi.org/10.18653/v1/2022.naacl-main.431 Annotators with attitudes: How annotator beliefs and identities bias toxic language detection . In Proceedings of the 2022 Conference...

    2022 doi

  38. [47]

    Lei Sha, Oana - Maria Camburu, and Thomas Lukasiewicz. 2023. https://doi.org/10.1016/J.ARTINT.2022.103828 Rationalizing predictions by adversarial information calibration . Artif. Intell., 315:103828

    2023

  39. [48]

    Hongning Wang, Yue Lu, and Chengxiang Zhai. 2010. https://doi.org/10.1145/1835804.1835903 Latent aspect rating analysis on review text data: a rating regression approach . In Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Wa...

    2010

  40. [49]

    Zeerak Waseem. 2016. https://doi.org/10.18653/v1/W16-5618 Are you a racist or am I seeing things? annotator influence on hate speech detection on T witter . In Proceedings of the First Workshop on NLP and Computational Social Science , pages 138--142, Austin, Texas. Associatio...

    2016 doi

  41. [50]

    Stephen Whitelam, Viktor Selin, Sang-Won Park, and Isaac Tamblyn. 2021. https://doi.org/10.1038/s41467-021-26568-2 Correspondence between neuroevolution and gradient descent . Nature Communications

    2021 doi

  42. [51]

    Sarah Wiegreffe and Ana Marasovic. 2021. https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/698d51a19d8a121ce581499d7b701668-Abstract-round1.html Teach me to explain: A review of datasets for explainable natural language processing . In Proceedings of the Neura...

    2021

  43. [52]

    Williams

    Ronald J. Williams. 1992. https://doi.org/10.1007/BF00992696 Simple statistical gradient-following algorithms for connectionist reinforcement learning . Mach. Learn., 8:229--256

    1992 doi

  44. [53]

    Jaakkola

    Mo Yu, Yang Zhang, Shiyu Chang, and Tommi S. Jaakkola. 2021. https://proceedings.neurips.cc/paper/2021/hash/6a711a119a8a7a9f877b5f379bfe9ea2-Abstract.html Understanding interlocking dynamics of cooperative rationalization . In Advances in Neural Information Processing Systems ...

    2021

  45. [54]

    Linan Yue, Qi Liu, Yichao Du, Yanqing An, Li Wang, and Enhong Chen. 2022. http://papers.nips.cc/paper\_files/paper/2022/hash/a9a67d9309a28372dde3de2a1c837390-Abstract-Conference.html DARE: disentanglement-augmented rationale extraction . In Advances in Neural Information Proce...

    2022

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.