Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Surrogate Fitness Metrics for Interpretable Reinforcement Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper claims that evolving starting states against a diversity-and-certainty fitness yields demonstrations that expose a reinforcement-learning policy's edge-case behavior better than random or fidelity-optimized baselines.

desk verdict REACT is a plausibly useful tool for diverse trajectory selection, but the continuous-control claim is overreach and the evaluation rests on an unvalidated fidelity proxy. read the letter →

arxiv 2504.14645 v1 pith:A5DZBPPK submitted 2025-04-20 cs.LG cs.AI

Philipp Altmann , Céline Davignon , Maximilian Zorn , Fabian Ritz , Claudia Linnhoff-Popien , Thomas Gabor This is my paper · ORCID
classification cs.LGcs.AI
keywords reinforcementlearninginterpretabilityexplainableevolutionaryoptimizationtrajectoryselectionsurrogatefitnessdemonstrationdiversityfidelity
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

This paper claims that a fixed reinforcement-learning policy can be understood through a small, carefully chosen set of demonstration trajectories, and that the right way to choose them is to evolve starting states against a surrogate fitness function rather than to maximize reward or even the fidelity score directly. The REACT framework combines global diversity among trajectories, local state coverage within a trajectory, and the policy's action certainty into a joint score, then runs a genetic algorithm over encoded initial states. In two gridworld environments, the selected sets reach fidelity IQMs of 0.744 and 2.130, compared with 0.355 and 1.486 for random starting states and 0.255 and 0.291 for fidelity-optimized baselines. In a continuous robotic reaching task, the same approach gives the largest fidelity gain for an early-stage policy, while for a converged policy direct fidelity optimization wins on the score even though REACT's trajectories cover more of the workspace. The result, on the paper's own terms, is a model-agnostic inspection tool that surfaces edge cases and uncertainty in any simulatable policy.

What carries the argument

Three quantities define the objective. Local diversity $D_l(\tau)=|\{s\in\tau\}|/|P|$ is the fraction of the discretized position space the trajectory visits. Certainty $C(\tau)=\frac{1}{|\tau|}\sum_{s,a\in\tau}\pi(a|s)$ is the policy's average confidence in its own actions. Global diversity $D_g(\tau,T)$ is the minimum one-way distance from the candidate to the already selected trajectories, normalized by the maximum state distance. The joint fitness $F(\tau,T)=D_g(\tau,T)+F_l$ combines these through the minimum-distance ratio term $F_l=\min_{t\in T}\left\|\left(\frac{D_l}{C}\right)(\tau)-\left(\frac{D_l}{C}\right)(t)\right\|^2$, which pushes a candidate to be far in the diversity-versus-uncertainty plane from its nearest neighbor in the set. REACT encodes initial states as bit strings, evolves them by tournament selection, single-point crossover, and bit-flip mutation, and prunes the population to keep the highest-fitness individuals, returning the surviving trajectories as the demonstration pool.

What would settle it

Ask human participants to answer behavioral questions about a policy after viewing either REACT-selected or random trajectories; if the sets with higher fidelity IQM are not the ones that enable more accurate answers, the surrogate fidelity measure fails as a stand-in for interpretability.

Watch

Extended reading notes

Core claim

The paper's central claim is that optimizing trajectory selection with the joint surrogate fitness $F(\tau,T)=D_g(\tau,T)+F_l$, where $D_g$ is the normalized one-way distance from the candidate trajectory to the demonstration set and $F_l=\min_{t\in T}\left\|\left(\frac{D_l}{C}\right)(\tau)-\left(\frac{D_l}{C}\right)(t)\right\|^2$ is the minimum squared gap in the local-diversity/certainty ratio, produces demonstration sets that expose the policy's diverse and edge-case behavior better than random initial states or direct optimization of the fidelity score. The evidence is the final fidelity IQMs in Table 1: $0.744$ in FlatGrid11 and $2.130$ in HoleyGrid11, versus $0.355$ and $1.486$ for random states and $0.255$ and $0.291$ for the fidelity-optimized baseline. The authors interpret high fidelity as high explanatory coverage of the policy's decision space. They also report that in continuous control the joint fitness is most useful for early-stage policies, while for a mature policy direct fidelity optimization yields higher scores even though REACT's trajectories show wider spatial coverage.

Load-bearing premise

The load-bearing premise is that the fidelity score used to quantify demonstration quality actually measures how well humans can understand the policy; the paper does not validate this link, so if score and understanding come apart, the main claim collapses.

Editorial extensions

If this is right

  • For partially trained or intermediate policies, REACT-selected demonstrations achieve higher fidelity IQMs than random starting states and than optimizing fidelity directly, so the method can reveal what an agent has learned mid-training.
  • In the two gridworld environments, the full joint fitness beats each of its components (certainty, local diversity, global diversity) and a simple sum of them, showing the minimum-distance ratio formulation is doing real work.
  • In continuous control, the benefit of diversity-based selection shrinks as the policy converges; for mature policies, direct fidelity optimization is the better surrogate, and REACT's advantage becomes qualitative (wider spatial coverage) rather than quantitative.
  • Because the method only needs to sample trajectories from a fixed policy, it is model-agnostic and can be applied to any RL agent whose behavior can be simulated.

Reading between the lines

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

  • A direct test of the paper's interpretability claim would replace the fidelity score with human ratings of explanation quality and check whether the two rankings agree across REACT, random, and fidelity-optimized sets.
  • The joint fitness could be turned into an adversarial curriculum by feeding REACT's diverse trajectories back into training, hardening the policy against out-of-distribution starting states; the paper mentions this direction but does not test it.
  • If local diversity were generalized beyond positions to object layouts or task parameters, the same evolutionary loop could interrogate policies over a much larger space of novel situations.
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

4 major / 5 minor

Summary. The paper introduces REACT, an evolutionary framework that perturbs a policy's initial states and selects a small set of demonstration trajectories intended to expose diverse, edge-case, or uncertain behaviors of a fixed reinforcement learning policy. Trajectories are scored by a joint surrogate fitness F (Eq. 5) combining global diversity, local diversity, and policy certainty, and the resulting demonstration set is evaluated by a reward-spread fidelity statistic S (Eq. 7). Experiments compare REACT against random initial states, the training initial state, a fidelity-optimized baseline, and ablated fitness components in FlatGrid11, HoleyGrid11, and FetchReach at three SAC training stages. The authors report higher fidelity IQMs for REACT over random and fidelity baselines in the two gridworlds and in early-stage FetchReach, and they argue for qualitative interpretability gains in continuous control through trajectory spread.

Significance. If the central claim held, REACT would be a useful model-agnostic tool for post-hoc policy analysis: it requires no gradient access, produces a compact set of user-reviewable demonstrations, and the authors release code and videos. A genuine strength is that the optimized joint fitness F (Eq. 5) is not identical to the evaluation metric S (Eq. 7), so the discrete results are not forced by construction, and the fitness function contains no fitted constants. The hyperparameter study and ablations also provide useful practical guidance. However, the paper's significance is currently capped by three issues: the evaluation metric S is asserted, not validated, as a measure of interpretability; the continuous results in Table 1 contradict the abstract's claim of improvement in both discrete and continuous environments; and the word 'significantly' is used without any significance tests. These problems are fixable by reframing claims and adding validation, but as written the headline contribution is stronger than the evidence.

major comments (4)
  1. [§6.1, Eq. (7)] The fidelity metric S is the quantitative backbone for the interpretability claim, but the paper never validates S as a measure of interpretability. Section 6.1 states that 'a higher fidelity score indicates that the selected demonstrations provide a more representative summary' and calls this 'intuitive,' yet no evidence links S to human understanding, explanation quality, or coverage of decision-relevant state-action space. Since the abstract's headline claim is that REACT 'significantly improves interpretability,' this unvalidated proxy is load-bearing and needs either external validation (e.g., a human study or correlation with an established interpretability measure) or a substantial weakening of the claim to 'diversity of generated trajectories.'
  2. [Table 1 and Abstract] The abstract states that optimizing trajectory selection 'significantly improves interpretability of RL policies in both discrete and continuous environments,' but Table 1 shows that in FetchReach at SAC-100k and SAC-150k, REACT's final fidelity IQM (1.373 and 0.137) is below the Random baseline (1.681 and 0.173). Section 6.5 acknowledges this and falls back to qualitative trajectory spread, but no quantitative measure of that spread is reported, so the continuous half of the headline claim is unsupported by the paper's primary metric. The abstract and the 'Discussion and Summary' section should be revised to state that REACT helps in early-stage continuous policies and that fidelity-based optimization is more effective for mature policies.
  3. [§6.3, Table 1] The fidelity-optimized baseline performs worse than Random under the fidelity metric it directly optimizes: in FlatGrid11 it achieves 0.255 versus Random's 0.355, and in HoleyGrid11 0.291 versus 1.486. If the reported S is the same objective being optimized, evolutionary search should not monotonically decrease it below a random initialization unless the optimization is stuck, the baseline is mis-specified, or the reported fidelity is computed differently from the optimized fidelity. The paper's explanation that 'directly maximizing fidelity does not necessarily yield the most informative demonstrations' addresses interpretability, not why the numerical metric drops; this discrepancy needs to be resolved before S can be trusted as the paper's main evaluation statistic.
  4. [§6.1, Sections 6.3-6.5] The paper repeatedly uses the word 'significantly' (abstract, Sections 6.3 and 6.4) but reports no significance tests, confidence intervals, or paired comparisons. Section 6.1 says results are 'averaged over ten random seeds to ensure statistical significance,' but averaging alone does not establish significance, and with 10 seeds the reported standard deviations in Table 1 overlap for several comparisons (e.g., FlatGrid11 REACT versus Random, and FetchReach SAC-150k REACT versus Random). The authors should report the rliable IQM confidence intervals or equivalent pairwise tests for the headline comparisons, and should remove the word 'significantly' where no test supports it.
minor comments (5)
  1. [§6.5 heading] The heading 'Continous Robotic Control' contains a typo; it should read 'Continuous Robotic Control.'
  2. [Eq. (7)] The notation |T| is overloaded: it denotes the size of the demonstration set in the weighting factor and the total trajectory length in the denominator. Using a distinct symbol, such as L_T, for the summed trajectory length would avoid ambiguity.
  3. [Eq. (2)] The normalization factor ⌈P⌉ is not defined precisely. The text says it is the 'maximum state distance' max_{s,s'} ||s - s'||_2, but it is unclear why the ceiling notation is used or how this quantity is computed in continuous state spaces; please define it explicitly.
  4. [§6.2] The hyperparameter choices (encoding length, population size, crossover and mutation probabilities) are selected based on experiments in the same environments that are later used for the main evaluation, with no held-out validation or correction for selection bias. This should be stated as a limitation, or the final comparisons should use a separate validation split.
  5. [Alg. 1, line 8] The notation 'children_{pc,F(P)}' in Algorithm 1 is unclear; it should state explicitly how crossover uses the fitness values and what the offspring size is relative to the population size.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: REACT's selection fitness F (Eq. 5) and the evaluation statistic S (Eq. 7) are distinct, and the paper's own negative continuous-control numbers show they are not aligned by construction.

full rationale

The claimed derivation chain is not circular. The objective actually optimized by REACT is the joint surrogate fitness F(τ,T)=Dg(τ,T)+Fl (Eq. 5), whose terms are state-space distances (Eqs. 2-3), visited-state coverage (Eq. 1), and policy probabilities (Eq. 4). The headline evidence is the fidelity IQM computed from Eq. 7, S=Σ(|τ|/|T|)|R̄−rτ|, a length-weighted dispersion of cumulative rewards. F does not contain S, S does not contain F, and no fitted constants bridge them. The paper's own results demonstrate the two quantities are not equivalent by construction: in both gridworlds REACT (joint fitness) beats a baseline that directly maximizes S, while in FetchReach SAC-100k and SAC-150k the fidelity-optimized baseline beats REACT (Table 1); a forced alignment would not produce this crossover. The self-citation to the earlier REACT paper [5] in Algorithm 1 is provenance, not load-bearing evidence; all equations and hyperparameters are specified in this manuscript. The fidelity metric is adapted from external work [15] and IQM from the rliable framework [1]. The genuine weaknesses—S is asserted to measure interpretability without external human validation, and hyperparameters are tuned on the same evaluation environments—are construct-validity and evaluation-tuning concerns, not circular derivation. Section 6.5 itself concedes 'REACT offers limited quantitative gains in fidelity at this stage,' further showing the evaluation metric is not rigged to favor the proposed fitness.

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

The method introduces no free parameters inside the fitness function itself, but the reported final results depend on several hyperparameters hand-selected on the same evaluation environments, and on two domain assumptions about how distances and fidelity relate to interpretability.

free parameters (5)
  • population_size = 10
    Hand-picked from Fig. 3 as a compromise between diversity and computational cost; the selection is made on the same gridworld environment used for the main evaluation, so it is a form of in-sample tuning.
  • generations = 40
    Chosen based on observed convergence within about 20 generations in FlatGrid11, so the maximum is set with knowledge of the evaluation environment.
  • crossover_probability = 0.75
    Selected by comparing a small grid of settings (Fig. 4) on FlatGrid11 with no held-out validation.
  • mutation_probability = 0.5
    Selected from the same grid (Fig. 4) as a tradeoff between exploration and exploitation.
  • state_encoding_length = 6 bits per dimension recommended; 9 bits per dimension in FetchReach
    Recommended based on the encoding-bias analysis in Section 6.2; the main gridworld experiments do not clearly state which length was used, which is an ambiguity.
assumptions (5)
  • domain assumption A policy trained from a single deterministic initial state can be meaningfully evaluated from other initial states to reveal its behavior.
    Algorithm 1 requires a policy trained with a single initial state; the experiments train PPO from one starting cell (Fig. 5a) and then vary the initial state. If the policy's behavior under distribution shift is not indicative of its general competence, the interpretability claims lose their force.
  • domain assumption The Euclidean distance between positions, and the one-way trajectory distance of Eq. 3, capture the behavioral difference between trajectories.
    Used in the global and local diversity terms (Eqs. 2-3) and in the inverse-normalization mapping; the paper acknowledges in Section 3.2 that other distance metrics would be needed for non-positional deviations.
  • ad hoc to paper The reward-spread fidelity S of Eq. 7 is a valid operationalization of how interpretable a demonstration set is.
    S is defined in this paper, adapted from Guo et al. [15] without derivation; all quantitative claims about interpretability rest on this measure, which is never validated against human judgment or a downstream explanation task.
  • standard math Standard evolutionary operators (tournament selection, single-point crossover, bit-flip mutation) together with the stated hyperparameters find good optima of the joint fitness.
    This is a standard assumption of evolutionary computation, used in Algorithm 1.
  • domain assumption The trained policies (PPO, SAC) are representative RL agents and their action probabilities are meaningful certainty signals.
    Certainty in Eq. 4 uses π(a|s); for the discrete gridworld this is the categorical action probability, and for continuous SAC this is a density, so the interpretation of certainty differs between environments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Surrogate Fitness Metrics for Interpretable Reinforcement Learning." pith.science (2026). https://pith.science/paper/A5DZBPPK

@misc{pith2026250414645,
  author       = {Pith},
  title        = {Pith review of: Surrogate Fitness Metrics for Interpretable Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A5DZBPPK}},
  note         = {Machine review of arXiv:2504.14645}
}
read the original abstract

We employ an evolutionary optimization framework that perturbs initial states to generate informative and diverse policy demonstrations. A joint surrogate fitness function guides the optimization by combining local diversity, behavioral certainty, and global population diversity. To assess demonstration quality, we apply a set of evaluation metrics, including the reward-based optimality gap, fidelity interquartile means (IQMs), fitness composition analysis, and trajectory visualizations. Hyperparameter sensitivity is also examined to better understand the dynamics of trajectory optimization. Our findings demonstrate that optimizing trajectory selection via surrogate fitness metrics significantly improves interpretability of RL policies in both discrete and continuous environments. In gridworld domains, evaluations reveal significantly enhanced demonstration fidelities compared to random and ablated baselines. In continuous control, the proposed framework offers valuable insights, particularly for early-stage policies, while fidelity-based optimization proves more effective for mature policies. By refining and systematically analyzing surrogate fitness functions, this study advances the interpretability of RL models. The proposed improvements provide deeper insights into RL decision-making, benefiting applications in safety-critical and explainability-focused domains.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Interpret Policies in Deep Reinforcement Learning using SILVER with RL-Guided Labeling: A Model-level Approach to High-dimensional and Multi-action Environments

    cs.LG 2025-10 reject novelty 4.0 of 10

    SILVER with RL-guided labeling: SHAP plus clustering plus policy-query labels plus decision trees or regression to interpret multi-action Atari policies.

Reference graph

Works this paper leans on

44 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    Advances in neural information processing systems 34:29304–29320

    Agarwal R, Schwarzer M, Castro PS, et al (2021) Deep reinforcement learning at the edge of the statistical precipice. Advances in neural information processing systems 34:29304–29320

    work page 2021

  2. [2]

    IEEE Access 8:171058–171077

    Alharin A, Doan TN, Sartipi M (2020) Reinforcement learning interpretation meth- ods: A survey. IEEE Access 8:171058–171077. https://doi.org/10.1109/ACCESS.2020. 3023394

    doi:10.1109/access.2020 2020

  3. [3]

    https://github.com/philippaltmann/hyphi-gym/

    Altmann P (2023) hyphi gym. https://github.com/philippaltmann/hyphi-gym/

    work page 2023

  4. [4]

    Altmann P, Ritz F, Feuchtinger L, et al (2023) Crop: towards distributional-shift robust reinforcement learning using compact reshaped observation processing. In: Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI ’23, https://doi.org/10.24963/ijcai.2023/380, URL https://doi.org/10.24963/ijcai.2023/ 380

    work page doi:10.24963/ijcai.2023/380 2023

  5. [5]

    In: Proceedings of the 16th International Joint Conference on Computational Intelligence

    Altmann P, Davignon C, Zorn M, et al (2024) React: Revealing evolutionary action consequence trajectories for interpretable reinforcement learning. In: Proceedings of the 16th International Joint Conference on Computational Intelligence. SciTePress, IJCCI ’24, pp 127–138, https://doi.org/10.5220/0013005900003837

    work page doi:10.5220/0013005900003837 2024

  6. [6]

    Neural Computing and Applications pp 1–17

    Altmann P, Ritz F, Zorn M, et al (2024) Discriminative reward co-training. Neural Computing and Applications pp 1–17. https://doi.org/10.1007/s00521-024-10512-8

    work page doi:10.1007/s00521-024-10512-8 2024

  7. [7]

    In: Adap- tive Agents and Multi-Agent Systems, URL https://api.semanticscholar.org/CorpusID: 21755369

    Amir D, Amir O (2018) Highlights: Summarizing agent behavior to people. In: Adap- tive Agents and Multi-Agent Systems, URL https://api.semanticscholar.org/CorpusID: 21755369

    work page 2018

  8. [8]

    Sports medicine 53(1):7–31

    Behrens M, Gube M, Chaabene H, et al (2023) Fatigue and human performance: an updated framework. Sports medicine 53(1):7–31

    work page 2023

Show all 44 references
  1. [9]

    Advances in Neural Information Processing Systems 35:37762–37777

    Bhatt V , Tjanaka B, Fontaine M, et al (2022) Deep surrogate assisted generation of environments. Advances in Neural Information Processing Systems 35:37762–37777

    2022

  2. [10]

    In: International conference on machine learning, PMLR, pp 2048–2056

    Cobbe K, Hesse C, Hilton J, et al (2020) Leveraging procedural generation to benchmark reinforcement learning. In: International conference on machine learning, PMLR, pp 2048–2056

    2020

  3. [11]

    John Wiley & Sons

    Fogel DB (2006) Evolutionary computation: toward a new philosophy of machine intelligence. John Wiley & Sons

    2006

  4. [12]

    In: Proceedings of the Genetic and Evolutionary Computation Conference Companion

    Gabor T, Altmann P (2019) Benchmarking surrogate-assisted genetic recommender systems. In: Proceedings of the Genetic and Evolutionary Computation Conference Companion. Association for Computing Machinery, New York, NY , USA, GECCO ’19, p 1568–1575, https://doi.org/10.1145/331...

    2019

  5. [13]

    In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 841–848

    Gabor T, Belzner L, Linnhoff-Popien C (2018) Inheritance-based diversity measures for explicit convergence control in evolutionary algorithms. In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 841–848

    2018

  6. [14]

    In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 898–906

    Gabor T, Sedlmeier A, Kiermeier M, et al (2019) Scenario co-evolution for reinforce- ment learning on a grid world smart factory domain. In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 898–906

    2019

  7. [15]

    Advances in Neural Information Processing Systems 34:12222–12236

    Guo W, Wu X, Khan U, et al (2021) Edge: Explaining deep reinforcement learning policies. Advances in Neural Information Processing Systems 34:12222–12236

    2021

  8. [16]

    CoRR abs/1801.01290

    Haarnoja T, Zhou A, Abbeel P, et al (2018) Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. CoRR abs/1801.01290. URL http://arxiv.org/abs/1801.01290, 1801.01290

    2018 arXiv

  9. [17]

    Knowledge-Based Systems 214:106685

    Heuillet A, Couthouis F, D ´ıaz-Rodr´ıguez N (2021) Explainability in deep rein- forcement learning. Knowledge-Based Systems 214:106685. https://doi.org/https: //doi.org/10.1016/j.knosys.2020.106685, URL https://www.sciencedirect.com/science/ article/pii/S0950705120308145

    2021

  10. [18]

    CoRR abs/1702.03465

    Huang SH, Held D, Abbeel P, et al (2017) Enabling robots to communicate their objectives. CoRR abs/1702.03465. URL http://arxiv.org/abs/1702.03465, 1702.03465

    2017 arXiv

  11. [19]

    CoRR abs/1810.08174

    Huang SH, Bhatia K, Abbeel P, et al (2018) Establishing appropriate trust via critical states. CoRR abs/1810.08174. URL http://arxiv.org/abs/1810.08174, 1810.08174

    2018 arXiv

  12. [20]

    In: 2008 IEEE congress on evolutionary computation (IEEE world congress on computational intelligence), IEEE, pp 2419–2426

    Ishibuchi H, Tsukamoto N, Nojima Y (2008) Evolutionary many-objective optimiza- tion: A short review. In: 2008 IEEE congress on evolutionary computation (IEEE world congress on computational intelligence), IEEE, pp 2419–2426

    2008

  13. [21]

    CoRR abs/1805.07917

    Khadka S, Tumer K (2018) Evolutionary reinforcement learning. CoRR abs/1805.07917. URL http://arxiv.org/abs/1805.07917, 1805.07917

    2018 arXiv

  14. [22]

    In: International conference on machine learning, PMLR, pp 1885–1894

    Koh PW, Liang P (2017) Understanding black-box predictions via influence functions. In: International conference on machine learning, PMLR, pp 1885–1894

    2017

  15. [23]

    CoRR abs/1905.13271

    Lage I, Lifschitz D, Doshi-Velez F, et al (2019) Exploring computational user models for agent policy summarization. CoRR abs/1905.13271. URL http://arxiv.org/abs/1905. 13271, 1905.13271

    2019 arXiv

  16. [24]

    Evolutionary computation 19(2):189–223

    Lehman J, Stanley KO (2011) Abandoning objectives: Evolution through the search for novelty alone. Evolutionary computation 19(2):189–223

    2011

  17. [25]

    Knowledge and Information Systems

    Li X, Xiong H, Li X, et al (2022) Interpretable deep learning: interpretation, inter- pretability, trustworthiness, and beyond. Knowledge and Information Systems

    2022

  18. [26]

    GeoInformatica 12:117–142

    Lin B, Su J (2008) One way distance: For shape based similarity search of moving object trajectories. GeoInformatica 12:117–142. URL https://api.semanticscholar.org/ 28 CorpusID:5279325

    2008

  19. [27]

    Advances in neural information processing systems 30

    Lundberg SM, Lee SI (2017) A unified approach to interpreting model predictions. Advances in neural information processing systems 30

    2017

  20. [28]

    Complex systems 9(3):193–212

    Miller BL, Goldberg DE, et al (1995) Genetic algorithms, tournament selection, and the effects of noise. Complex systems 9(3):193–212

    1995

  21. [29]

    In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 837–845

    Neumann A, Gao W, Wagner M, et al (2019) Evolutionary diversity optimization using multi-objective indicators. In: Proceedings of the Genetic and Evolutionary Computation Conference, pp 837–845

    2019

  22. [30]

    In: Proceedings of the 31st ACM SIGSOFT International Symposium on Software Testing and Analysis, pp 378–390

    Pang Q, Yuan Y , Wang S (2022) Mdpfuzz: testing models solving markov decision processes. In: Proceedings of the 31st ACM SIGSOFT International Symposium on Software Testing and Analysis, pp 378–390

    2022

  23. [31]

    CoRR abs/2002.00632

    Parker-Holder J, Pacchiano A, Choromanski K, et al (2020) Effective diversity in population-based reinforcement learning. CoRR abs/2002.00632. URL https://arxiv.org/ abs/2002.00632, 2002.00632

    2020 arXiv

  24. [32]

    In: International Conference on Machine Learning, PMLR, pp 17473–17498

    Parker-Holder J, Jiang M, Dennis M, et al (2022) Evolving curricula with regret-based environment design. In: International Conference on Machine Learning, PMLR, pp 17473–17498

    2022

  25. [33]

    Advances in Neural Information Processing Systems 33:17044–17056

    Pleiss G, Zhang T, Elenberg E, et al (2020) Identifying mislabeled data using the area under the margin ranking. Advances in Neural Information Processing Systems 33:17044–17056

    2020

  26. [34]

    In: Inter- national cross-domain conference for machine learning and knowledge extraction, Springer, pp 77–95

    Puiutta E, Veith EM (2020) Explainable reinforcement learning: A survey. In: Inter- national cross-domain conference for machine learning and knowledge extraction, Springer, pp 77–95

    2020

  27. [35]

    Handbooks in operations research and management science 2:331–434

    Puterman ML (1990) Markov decision processes. Handbooks in operations research and management science 2:331–434

    1990

  28. [36]

    Progress in aerospace sciences 41(1):1–28

    Queipo NV , Haftka RT, Shyy W, et al (2005) Surrogate-based analysis and optimization. Progress in aerospace sciences 41(1):1–28

    2005

  29. [37]

    Journal of Machine Learning Research 22(268):1–8

    Raffin A, Hill A, Gleave A, et al (2021) Stable-baselines3: Reliable reinforcement learning implementations. Journal of Machine Learning Research 22(268):1–8. URL http://jmlr.org/papers/v22/20-1364.html

    2021

  30. [38]

    In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pp 1135–1144

    Ribeiro MT, Singh S, Guestrin C (2016) ” 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 mining, pp 1135–1144

    2016

  31. [39]

    Sutton AGB (2014, 2015) Reinforcement Learning: An Introduction, 2nd edn

    Richard S. Sutton AGB (2014, 2015) Reinforcement Learning: An Introduction, 2nd edn. The MIT Press, Cambridge, Massachusetts, London, England 29

    2014

  32. [40]

    1707.06347

    Schulman J, Wolski F, Dhariwal P, et al (2017) Proximal policy optimization algorithms. 1707.06347

    2017 arXiv

  33. [41]

    Artificial Intelligence 288:103367

    Sequeira P, Gervasio M (2020) Interestingness elements for explainable reinforce- ment learning: Understanding agents’ capabilities and limitations. Artificial Intelligence 288:103367. https://doi.org/https://doi.org/10.1016/j.artint.2020.103367, URL https:// www.sciencedirect...

    2020

  34. [42]

    arXiv preprint arXiv:220504887

    Tappler M, C ´ordoba FC, Aichernig BK, et al (2022) Search-based testing of reinforce- ment learning. arXiv preprint arXiv:220504887

    2022

  35. [43]

    In: The Eleventh International Conference on Learning Representations, URL https://openreview.net/forum?id=bLmSMXbqXr

    Wu S, Yao J, Fu H, et al (2023) Quality-similar diversity via population based reinforce- ment learning. In: The Eleventh International Conference on Learning Representations, URL https://openreview.net/forum?id=bLmSMXbqXr

    2023

  36. [44]

    IEEE Transactions on Software Engineering 49(7):3715–3735 30

    Zolfagharian A, Abdellatif M, Briand LC, et al (2023) A search-based testing approach for deep reinforcement learning agents. IEEE Transactions on Software Engineering 49(7):3715–3735 30

    2023

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.