Pith. sign in

REVIEW 4 major objections 6 minor 46 references

Exploiting Constraint Reasoning to Build Graphical Explanations for Mixed-Integer Linear Programming

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Contrastive explanations for MILP solutions can be built by extracting a minimal infeasible constraint set and rendering it as a connected graph of natural-language reasons.

desk verdict A novel IIS-based pipeline for contrastive MILP explanations, undone by an inconsistent running example and an unaddressed gap between infeasible and suboptimal queries. read the letter →

arxiv 2507.13007 v1 pith:NGJ6VES7 submitted 2025-07-17 cs.AI

Roger Xavier Lera-Leri , Filippo Bistaffa , Athina Georgara , Juan Antonio Rodriguez-Aguilar This is my paper
classification cs.AI
keywords explainableAImixed-integerlinearprogrammingcontrastiveexplanationsirreducibleinfeasiblesubsystemconstraintreasoninggraphofreasonsuserqueriesoptimizationexplainability
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

The paper proposes X-MILP, a domain-agnostic method for explaining why an optimal solution of a mixed-integer linear program (MILP) is better than an alternative the user might prefer. The key move is to translate the user's 'why not this?' query into extra constraints, form a satisfaction problem that also demands a solution no worse than the optimum, and then extract an irreducible infeasible subsystem (IIS) — a minimal set of constraints that cannot all hold. That IIS is rendered as a connected graph whose nodes are natural-language reasons, so the explanation shows how the user's desired change conflicts with the problem's structure. The authors evaluate the computational cost on scheduling and auction problems and find that, for many instances, the explanation can be computed faster than the original optimization.

What carries the argument

The load-bearing object is the IIS — the irreducible infeasible subsystem — computed by standard deletion-filter or additive algorithms of constraint reasoning. It is the minimal set of constraints that makes the User-Desired Satisfiability Problem infeasible, and it is what turns an unstructured pile of mathematical inequalities into a candidate explanation. The paper augments the IIS with the dual-graph construction: each constraint becomes a node, and two nodes are joined when the constraints share a decision variable. Labeling each node with a natural-language template converts the dual graph into the 'graph of reasons'; Theorem 1 proves this graph is always connected, which the authors take as a structural guarantee that the explanation is not a set of disconnected statements.

What would settle it

Ask users to compare the graph of reasons against a plain list of the same constraints for the same query; if comprehension or acceptance is not better for the graph, the central claim about graphical structure adding value is falsified. A more direct test: find an MILP instance with multiple optimal solutions and pose a query that is satisfied by some other optimum; the UDSP is feasible, so no IIS exists, and the method must either return an empty explanation or silently claim the query is answered—either outcome would show the framework covers only a subset of queries.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that a contrastive explanation for an MILP solution can be produced without re-solving an alternative optimization problem. Given an optimal solution with value $f^*$, the user's query is converted into linear query constraints; adding these to the original constraints together with the minimality constraint $f(x) \le f^*$ yields the User-Desired Satisfiability Problem. If the query cannot yield a solution at least as good as $f^*$, this system is infeasible, and its irreducible infeasible subsystem identifies exactly the constraints responsible. The dual graph of that IIS is connected (Theorem 1), so the resulting graph of reasons is always a single, connected structure, not scattered fragments.

Load-bearing premise

The load-bearing premise is that labeling IIS constraints with natural-language templates yields an explanation a human will actually understand and accept as the reason for the outcome.

Editorial extensions

If this is right

  • Explanations can be computed without solving an additional MILP, unlike earlier contrastive approaches that re-optimize.
  • For the RCPSP benchmark, the median overhead of computing an explanation is 6.7% of the time needed to solve the instance, and over 75% of explanations finish in under a minute.
  • For the combinatorial-auction (WDP) benchmarks, explanations for small and medium instances are computed in under a minute on average, with the scheduling distribution remaining fast even for large instances.
  • A smallest-IIS algorithm yields measurably smaller IIS than an off-the-shelf solver in only about 7% of cases, at an average runtime increase of four orders of magnitude, so using the cheaper solver is the practical choice.
  • Because the dual graph of any IIS is connected, every graph of reasons is a single connected structure, which the paper interprets as a guarantee of coherence.

Reading between the lines

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

  • A natural next step—not taken in the paper—is a controlled user study measuring whether humans understand these graphs of reasons better than a plain list of constraints; the paper's own experiments measure runtime and size only.
  • The connectedness theorem suggests a quality criterion: any explanation that fragments into disconnected components is likely to be missing constraints, which could be used as a consistency check on other explanation generators.
  • Because the UDSP is infeasible whenever the user's query cannot match the optimal value, the method implicitly assumes the query is 'too demanding'; extending the framework to the feasible case (alternative optima) would require a different notion of explanation, perhaps comparing the two optimal solutions directly.
  • The query classification (enforce/veto) is only instantiated for two problem families; applying X-MILP to a new domain requires an expert to write the natural-language templates, which is the only human-dependent step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes X-MILP, a domain-agnostic method for building contrastive explanations of MILP optimal solutions. User queries are translated into linear constraints; a User-Desired Satisfiability Problem (UDSP) is formed by adding, to the original constraints and the query constraints, a bound requiring an objective value no worse than f*. An Irreducible Infeasible Subsystem (IIS) of the UDSP is computed, and the constraints in the IIS are labeled with natural-language templates and organized as the dual graph of the IIS, called the graph of reasons. The authors prove that the dual graph of an IIS is connected (Theorem 1) and report experiments on RCPSP and WDP instances measuring IIS computation time, overhead relative to MILP solving time, and a comparison between CPLEX's heuristic IIS and the smallest IIS computed by an adaptation of FORQES.

Significance. If the central claim holds, X-MILP offers a principled way to generate contrastive explanations for MILP solutions without re-solving the optimization problem, relying instead on an IIS computation. The paper includes a valid formal result (Theorem 1), a clear domain-agnostic pipeline, a concrete treatment of query translation, and a broad empirical study on two well-known benchmark families, with code released. The main weaknesses are that the explanatory value of the graph of reasons is not evaluated with users, the choice of which IIS to use is left to a solver heuristic without a semantic justification, and the running example appears to contain an inconsistency about the role of the minimality constraint. These gaps directly affect the paper's central claim that the extracted IIS constitutes a correct and useful explanation of suboptimality.

major comments (4)
  1. [§4.2–4.3, Definition 5 and Figure 2] The IIS drawn in Figure 2 is {q, c1, ..., c7} and contains no copy of the minimality constraint f(x) ≤ 43, although the UDSP defined in Example 3 explicitly includes this constraint. If C ∪ CQ is feasible, any IIS of the UDSP must contain f(x) ≤ f*, because removing that constraint would restore feasibility. Conversely, if the displayed set is an IIS, then C ∪ CQ itself is infeasible, which contradicts Example 2's statement that enforcing the query 'would allow to obtain a user-desired solution.' The manuscript must clarify which system is represented in the figure and must show that the IIS used for the explanation is an IIS of the UDSP, not merely of C ∪ CQ; otherwise the explanation concerns infeasibility rather than suboptimality, and the paper's central claim is not supported.
  2. [§4.3 and §5.2] The paper acknowledges that an infeasible system can have exponentially many IISs and that CPLEX returns one by heuristic, but the experimental protocol then uses CPLEX's IIS without any criterion linking the returned IIS to the semantics of the query. Since deletion-filter heuristics depend on the order in which constraints are examined, the resulting graph of reasons may be an artifact of solver tie-breaking rather than a well-defined answer to the user's query. I ask for either (a) a principled selection rule, such as a weighted or lexicographic cost reflecting the query, or (b) a stability analysis showing that the explanation does not change materially across different IIS choices, or (c) a user study establishing that the particular IIS chosen matters for the resulting explanation.
  3. [§4.4 and Conclusions] The claim that the graph of reasons 'helps the user understand the structure among the reasons' is not tested. Section 5 measures only IIS runtime and IIS size; no human evaluation, no task-based accuracy measure, and no comparison against an unstructured list of constraints is reported. Because the added value of the dual-graph representation is a central assertion of the paper, this is a load-bearing missing evaluation: without it, the paper supports only the computational hardness claims, not the explanatory value of the produced artifact.
  4. [§5.1–5.2, Figure 3] The runtime comparison between CPLEX and FORQES is reported as means without confidence intervals or significance tests, and the 10-hour timeout may censor slow instances in a way that underestimates FORQES's average runtime. Please report the number of timed-out instances per method, the median and interquartile range, and a significance test (or justify why the comparison is insensitive to censoring).
minor comments (6)
  1. [Abstract] The phrase 'for trustwor' appears truncated in the abstract; please check the source file.
  2. [Example 2 and Figure 2] Please state explicitly whether the IIS in Figure 2 is for the UDSP of Example 3; if so, the node for f(x) ≤ 43 must appear in the dual graph and graph of reasons, or the example should be revised to a query where C ∪ CQ is feasible.
  3. [§4.3] The sentence 'Following [39], we should aim to compute the smallest possible IIS' cites a metrics-oriented reference; a more direct citation to the smallest-MUS extraction literature (e.g., [31]) would be more precise for this claim.
  4. [Figure 3] The left y-axis is labeled 'IIS size reduction (%)' but the histogram appears to range up to about 60; please clarify the units and specify that the line represents the runtime increase ratio.
  5. [Table 2] The notation EFj and LFj is used in the Q3 and Q4 rows but is not defined in the table or its caption; please define them in the text.
  6. [Theorem 1 proof] The sentence 'since the both subsets of constraints have different variables (different domains)' should read 'since the two subsets have disjoint sets of variables' for clarity.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the X-MILP derivation from query constraints to an IIS-based graph of reasons is self-contained, with only minor non-load-bearing self-citations.

full rationale

The paper's claimed derivation chain is: translate a user query into constraints CQ; form the User-Desired Satisfiability Problem with constraints f(x) ≤ f*, C, and CQ; extract an IIS; label its dual graph to obtain a graph of reasons. Each stage is independently defined. The infeasibility conclusion is a logical consequence of optimality: since f* is the MILP optimum, any solution satisfying C with f ≤ f* is optimal, and infeasibility of the UDSP means no solution satisfies C, CQ, and f ≤ f*, so any query-compliant solution is strictly worse. This is not a restatement of the conclusion, and no parameter is fitted to data. The main methodological choices (IIS, dual graph, NL templates) come from standard constraint-reasoning literature or from earlier non-self work, and the self-citations that do appear ([17] for query classification, and [26-28] as application and future-work pointers) are accompanied by explicit encodings in the paper and are not load-bearing for the central IIS construction. The paper does contain a correctness/consistency concern: the running-example IIS in Figure 2 omits the minimality constraint f ≤ 43 even though Example 2 asserts that C∪CQ is feasible, so the displayed object is not an IIS of the stated UDSP; and the absence of any user evaluation means the explanatory value of a graph of reasons is assumed rather than demonstrated. These are limitations about validity and usefulness, not circularity, and the derivation itself is self-contained against external optimization solvers. Accordingly the circularity score is low.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The approach rests on standard optimization assumptions (feasibility, minimization) and on the logical identity between UDSP infeasibility and the absence of a better solution. No free parameters are fitted to data, and no new entities are postulated.

assumptions (4)
  • domain assumption The main MILP problem M is feasible.
    Definition 4 states 'Assuming M is a feasible problem.' Without this, there is no optimal solution to explain.
  • standard math The objective function is minimised; maximisation is handled by sign conventions.
    Definition 5 notes 'Without loss of generality, we assume that M is a minimisation problem.' This is a conventional reduction.
  • domain assumption Infeasibility of the UDSP implies that no solution satisfying the user's query can be at least as good as the optimal solution.
    This is the logical bridge between the CSP and the explanation; it is definitional given the constraint f(x) <= f*.
  • domain assumption An IIS is a minimal set of inconsistent constraints, and its constraints are appropriate reasons for the explanation.
    The paper assumes that the constraints in the IIS constitute the answer to the user's query (Section 4.3), and this is not empirically validated for humans.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploiting Constraint Reasoning to Build Graphical Explanations for Mixed-Integer Linear Programming." pith.science (2026). https://pith.science/paper/NGJ6VES7

@misc{pith2026250713007,
  author       = {Pith},
  title        = {Pith review of: Exploiting Constraint Reasoning to Build Graphical Explanations for Mixed-Integer Linear Programming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGJ6VES7}},
  note         = {Machine review of arXiv:2507.13007}
}
read the original abstract

Following the recent push for trustworthy AI, there has been an increasing interest in developing contrastive explanation techniques for optimisation, especially concerning the solution of specific decision-making processes formalised as MILPs. Along these lines, we propose X-MILP, a domain-agnostic approach for building contrastive explanations for MILPs based on constraint reasoning techniques. First, we show how to encode the queries a user makes about the solution of an MILP problem as additional constraints. Then, we determine the reasons that constitute the answer to the user's query by computing the Irreducible Infeasible Subsystem (IIS) of the newly obtained set of constraints. Finally, we represent our explanation as a "graph of reasons" constructed from the IIS, which helps the user understand the structure among the reasons that answer their query. We test our method on instances of well-known optimisation problems to evaluate the empirical hardness of computing explanations.

Figures

Figures reproduced from arXiv: 2507.13007 by the authors.

Figure 1
Figure 1. Overview of our approach. See Section 4 for the details. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Process to build a graph-based explanation applied to our running ex [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Distribution of IIS size reduction between those computed by FORQES [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Runtimes to solve MILP problems and compute IISs (left y-axis), and [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Average IIS computation time (lines, left y-axis) and overhead (dots, right y-axis) for different query types for the RCPSP. 5.2 Results Smallest IIS We compute the IISs resulting from all generated queries with both algorithms (i.e., CPLEX and FORQES), and we compare …
Figure 6
Figure 6. Figure 6: Average IIS runtimes (lines) and overhead (dots) for different WDP in￾stance distributions: scheduling, matching, paths, and regions. Time overhead when computing explanations RCPSP [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 42 canonical work pages

  1. [1]

    IEEE access6, 52138–52160 (2018)

    Adadi, A., Berrada, M.: Peeking inside the black-box: a survey on explainable artificial intelligence (xai). IEEE access6, 52138–52160 (2018)

    work page 2018

  2. [2]

    Knowledge-Based Systems 182, 104799 (2019)

    Andrejczuk, E., Bistaffa, F., Blum, C., Rodríguez-Aguilar, J.A., Sierra, C.: Syn- ergistic team composition: A computational approach to foster diversity in teams. Knowledge-Based Systems 182, 104799 (2019)

    work page 2019

  3. [3]

    Information fusion58, 82–115 (2020)

    Arrieta, A.B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., García, S., Gil-López, S., Molina, D., Benjamins, R., et al.: Explainable artifi- cial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai. Information fusion58, 82–115 (2020)

    2020

  4. [4]

    IEEE Transactions on Intelligent Transportation Systems22(1), 119–130 (2019)

    Bistaffa, F., Blum, C., Cerquides, J., Farinelli, A., Rodríguez-Aguilar, J.A.: A computational approach to quantify the benefits of ridesharing for policy makers and travellers. IEEE Transactions on Intelligent Transportation Systems22(1), 119–130 (2019)

    work page 2019

  5. [5]

    Artificial Intelligence300, 103550 (2021)

    Bogaerts, B., Gamba, E., Guns, T.: A framework for step-wise explaining how to solve constraint satisfaction problems. Artificial Intelligence300, 103550 (2021)

    work page 2021

  6. [6]

    Cambridge University Press (2004)

    Boyd, S., Vandenberghe, L.: Convex Optimization. Cambridge University Press (2004)

    2004

  7. [7]

    European Journal of Operational Research73(1), 139–143 (1994)

    Chakravarti, N.: Some results concerning post-infeasibility analysis. European Journal of Operational Research73(1), 139–143 (1994)

    work page 1994

  8. [8]

    Computational Opti- mization and Applications4, 167–179 (1995)

    Chinneck, J.W.: Analyzing infeasible nonlinear programs. Computational Opti- mization and Applications4, 167–179 (1995)

    work page 1995

Show all 46 references
  1. [9]

    INFORMS Journal on Computing9(2), 164–174 (1997)

    Chinneck, J.W.: Finding a useful subset of constraints for analysis in an infeasible linear program. INFORMS Journal on Computing9(2), 164–174 (1997)

    1997

  2. [10]

    Chinneck, J.W.: Feasibility and Infeasibility in Optimization:: Algorithms and Computational Methods, vol. 118. Springer Science & Business Media (2007)

    2007

  3. [11]

    ORSA Journal on Computing3(2), 157–168 (1991)

    Chinneck, J.W., Dravnieks, E.W.: Locating minimal infeasible constraint sets in linear programs. ORSA Journal on Computing3(2), 157–168 (1991)

    1991

  4. [12]

    https: //digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai

    European Commission: Ethics guidelines for trustworthy ai. https: //digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai. Online (2019)

    2019

  5. [13]

    INFORMS Journal on computing 15(3), 284–309 (2003)

    De Vries, S., Vohra, R.V.: Combinatorial auctions: A survey. INFORMS Journal on computing 15(3), 284–309 (2003)

    2003

  6. [14]

    Springer Nature (2022)

    Dechter, R.: Reasoning with probabilistic and deterministic graphical models: Ex- act algorithms. Springer Nature (2022)

    2022

  7. [15]

    Artificial Intelligence328, 104064 (2024)

    Fenoy, A., Bistaffa, F., Farinelli, A.: An attention model for the formation of col- lectives in real-world domains. Artificial Intelligence328, 104064 (2024)

    2024

  8. [16]

    Journal of Artificial Intelligence Research78, 709–746 (2023)

    Gamba, E., Bogaerts, B., Guns, T.: Efficiently explaining csps with unsatisfiable subset optimization. Journal of Artificial Intelligence Research78, 709–746 (2023)

    2023

  9. [17]

    In: Proceedings of the 21st International Conference on Autonomous Agents and Multiagent Systems

    Georgara, A., Rodriguez Aguilar, J.A., Sierra, C.: Building contrastive explana- tions for multi-agent team formation. In: Proceedings of the 21st International Conference on Autonomous Agents and Multiagent Systems. pp. 516–524 (2022)

    2022

  10. [18]

    ORSA Journal on Computing2(1), 61–63 (1990)

    Gleeson, J., Ryan, J.: Identifying minimally infeasible subsystems of inequalities. ORSA Journal on Computing2(1), 61–63 (1990)

    1990

  11. [19]

    INFORMS Journal on Computing11(1), 63–77 (1999)

    Guieu, O., Chinneck, J.W.: Analyzing infeasible mixed-integer and integer linear programs. INFORMS Journal on Computing11(1), 63–77 (1999)

    1999

  12. [20]

    European Journal of Operational Research 297(1) (02 2022) 18 R

    Hartmann, S., Briskorn, D.: An updated survey of variants and extensions of the resource-constrained project scheduling problem. European Journal of Operational Research 297(1) (02 2022) 18 R. X. Lera-Leri et al

    2022

  13. [21]

    In: International Conference on Principles and Practice of Constraint Programming

    Ignatiev, A., Previti, A., Liffiton, M., Marques-Silva, J.: Smallest mus extraction with minimal hitting set dualization. In: International Conference on Principles and Practice of Constraint Programming. pp. 173–182. Springer (2015)

    2015

  14. [22]

    In: Proceedings of the 19th national conference on Artifical intelligence

    Junker, U.: Quickxplain: Preferred explanations and relaxations for over- constrained problems. In: Proceedings of the 19th national conference on Artifical intelligence. pp. 167–172 (2004)

    2004

  15. [23]

    European journal of opera- tional research 96(1), 205–216 (1997)

    Kolisch, R., Sprecher, A.: Psplib-a project scheduling problem library: Or software- orsep operations research software exchange program. European journal of opera- tional research 96(1), 205–216 (1997)

    1997

  16. [24]

    In: ICAPS 2021 Work- shop on Explainable AI Planning

    Korikov, A., Shleyfman, A., Beck, C.: Counterfactual explanations for optimization-based decisions in the context of the gdpr. In: ICAPS 2021 Work- shop on Explainable AI Planning. pp. 4097–4103 (2021)

    2021

  17. [25]

    arXiv preprint arXiv:2202.01875 (2022)

    Lakkaraju, H., Slack, D., Chen, Y., Tan, C., Singh, S.: Rethinking explainability as a dialogue: A practitioner’s perspective. arXiv preprint arXiv:2202.01875 (2022)

    2022 arXiv

  18. [26]

    Annals of Op- erations Research pp

    Lera-Leri, R.X., Bistaffa, F., Trescak, T., Rodríguez-Aguilar, J.A.: Computing job- tailored degree plans towards the acquisition of professional skills. Annals of Op- erations Research pp. 1–34 (2025)

    2025

  19. [27]

    Knowledge-Based Systems287, 111453 (2024)

    Lera-Leri, R.X., Liscio, E., Bistaffa, F., Jonker, C.M., Lopez-Sanchez, M., Mu- rukannaiah, P.K., Rodriguez-Aguilar, J.A., Salas-Molina, F.: Aggregating value systems for decision support. Knowledge-Based Systems287, 111453 (2024)

    2024

  20. [28]

    In: ECAI 2024, pp

    Lera-Leri, R.X., Bistaffa, F.: A robust and scalable approach to meet user pref- erences in research project planning. In: ECAI 2024, pp. 4516–4523. IOS Press (2024)

    2024

  21. [29]

    In: Principles and Practice of Constraint Programming-CP 2002: 8th International Conference, CP 2002 Ithaca, NY, USA, September 9–13, 2002 Proceedings 8

    Leyton-Brown, K., Nudelman, E., Shoham, Y.: Learning the empirical hardness of optimization problems: The case of combinatorial auctions. In: Principles and Practice of Constraint Programming-CP 2002: 8th International Conference, CP 2002 Ithaca, NY, USA, September 9–13, 2002 ...

    2002

  22. [30]

    In: Proceedings of the 2nd ACM conference on Electronic commerce

    Leyton-Brown, K., Pearson, M., Shoham, Y.: Towards a universal test suite for combinatorial auction algorithms. In: Proceedings of the 2nd ACM conference on Electronic commerce. pp. 66–76 (2000)

    2000

  23. [31]

    Constraints 14, 415–442 (2009)

    Liffiton, M., Mneimneh, M., Lynce, I., Andraus, Z., Marques-Silva, J., Sakallah, K.: A branch and bound algorithm for extracting smallest minimal unsatisfiable subformulas. Constraints 14, 415–442 (2009)

    2009

  24. [32]

    In: International conference on theory and applications of satisfiability testing

    Liffiton, M.H., Sakallah, K.A.: On finding all minimally unsatisfiable subformulas. In: International conference on theory and applications of satisfiability testing. pp. 173–186. Springer (2005)

    2005

  25. [33]

    In: Foundations of Artificial Intelligence, vol

    Meseguer, P., Rossi, F., Schiex, T.: Soft constraints. In: Foundations of Artificial Intelligence, vol. 2, pp. 281–328. Elsevier (2006)

    2006

  26. [34]

    Artificial intelligence 267, 1–38 (2019)

    Miller, T.: Explanation in artificial intelligence: Insights from the social sciences. Artificial intelligence 267, 1–38 (2019)

    2019

  27. [35]

    In: Proceedings of the 21st International Con- ference on Autonomous Agents and Multiagent Systems

    Nardi, O., Boixel, A., Endriss, U.: A graph-based algorithm for the automated justification of collective decisions. In: Proceedings of the 21st International Con- ference on Autonomous Agents and Multiagent Systems. pp. 935–943 (2022)

    2022

  28. [36]

    Annals of Mathematics and Artificial Intelligence17(1), 107– 126 (1996)

    Parker, M., Ryan, J.: Finding the minimum weight iis cover of an infeasible system of linear inequalities. Annals of Mathematics and Artificial Intelligence17(1), 107– 126 (1996)

    1996

  29. [37]

    In: Proceedings of the Inter- national Conference on Automated Planning and Scheduling

    Pozanco, A., Mosca, F., Zehtabi, P., Magazzeni, D., Kraus, S.: Explaining preference-driven schedules: the expres framework. In: Proceedings of the Inter- national Conference on Automated Planning and Scheduling. vol. 32, pp. 710–718 (2022) Graphical Explanations for MILP 19

    2022

  30. [38]

    Journal of the Academy of Marketing Science 48(1), 137–141 (2020)

    Rai, A.: Explainable ai: From black box to glass box. Journal of the Academy of Marketing Science 48(1), 137–141 (2020)

    2020

  31. [39]

    In: Proceedings of the 20th international conference on autonomous agents and mul- tiagent systems

    Rosenfeld, A.: Better metrics for evaluating explainable artificial intelligence. In: Proceedings of the 20th international conference on autonomous agents and mul- tiagent systems. pp. 45–50 (2021)

    2021

  32. [40]

    In: Explain- able AI: interpreting, explaining and visualizing deep learning, pp

    Samek, W., Müller, K.R.: Towards explainable artificial intelligence. In: Explain- able AI: interpreting, explaining and visualizing deep learning, pp. 5–22. Springer (2019)

    2019

  33. [41]

    In: ISMVL

    Silva, J.P.M.: Minimal unsatisfiability: Models, algorithms and applications (in- vited paper). In: ISMVL. vol. 10, pp. 9–14 (2010)

    2010

  34. [42]

    Computers & operations research23(2), 113–119 (1996)

    Tamiz, M., Mardle, S.J., Jones, D.F.: Detecting iis in infeasible linear programmes using techniques from goal programming. Computers & operations research23(2), 113–119 (1996)

    1996

  35. [43]

    European Journal of Operational Research8(3), 283–288 (1981)

    Van Loon, J.: Irreducibly inconsistent systems of linear inequalities. European Journal of Operational Research8(3), 283–288 (1981)

    1981

  36. [44]

    John Wiley & Sons (2020)

    Wolsey, L.A.: Integer programming. John Wiley & Sons (2020)

    2020

  37. [45]

    arXiv preprint arXiv:2110.02395 (2021)

    Zečević, M., Dhami, D.S., Rothkopf, C.A., Kersting, K.: Causal explanations of structural causal models. arXiv preprint arXiv:2110.02395 (2021)

    2021 arXiv

  38. [46]

    In: Proceedings of the Inter- national Conference on Automated Planning and Scheduling

    Zehtabi, P., Pozanco, A., Bolch, A., Borrajo, D., Kraus, S.: Contrastive explana- tions of centralized multi-agent optimization solutions. In: Proceedings of the Inter- national Conference on Automated Planning and Scheduling. vol. 34, pp. 671–679 (2024)

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.