Pith. sign in

REVIEW 4 major objections 4 minor 40 references

Decomposed Inductive Procedure Learning: Learning Academic Tasks with Human-Like Data Efficiency

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Three cooperating learning mechanisms let an agent master arithmetic tutor tasks in about 20 problems, where deep RL needs about 10,000.

desk verdict A useful ablation ladder, but the headline claim that mechanism decomposition drives the efficiency gain is confounded with priors and representation; the 2-vs-3 contrast in multi-column addition is the cleanest evidence and is worth engaging. read the letter →

arxiv 2505.10422 v1 pith:WNCY7JZJ submitted 2025-05-15 cs.LG

Daniel Weitekamp , Christopher MacLellan , Erik Harpstead , Kenneth Koedinger This is my paper · ORCID
classification cs.LG
keywords decomposedinductiveprocedurelearningdataefficiencymulti-mechanismintelligenttutoringsystemsreinforcementablationsymbolicruleinductionhuman-likeratehow-where-when
open problems The Measurement Problem
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 tries to establish that human-like data efficiency in learning arithmetic procedures comes not from any single learning algorithm, but from splitting learning into several functionally distinct mechanisms that cooperate. On two intelligent-tutoring tasks, an agent with three such mechanisms—how to compute an action, where an action can apply, and when to apply it—reaches under 10 percent error in about 20 practice problems, whereas deep reinforcement learning needs roughly 10,000 and a decision-tree learner needs thousands. A staged ablation shows each added mechanism yields several orders of magnitude of improvement, and the decomposition helps more than the switch from subsymbolic to symbolic learning alone. If true, this reframes the gap between humans and data-hungry machine learning: the missing ingredient may be architectural decomposition of credit assignment, not a better single learning rule.

What carries the argument

The machinery is a 3-mechanism skill-induction loop. How-learning abductively composes primitive domain-general functions (Add, Multiply, OnesDigit, etc.) to reproduce worked-example actions, retaining the most parsimonious composition and generalizing constants to variables. Where-learning induces argument-variable patterns—a selection variable plus spatial relations—that determine the contexts in which a skill can apply, generalizing across columns or neighboring interface elements. When-learning learns preconditions via decision-tree or relational concept induction over features restated relative to the selection and arguments (relative featurization via shortest paths). Each mechanism is instantiated per skill, so each simplifies the learning problem for the others: how-learning proposes candidate explanations, where-learning sets the matching contexts, and when-learning decides ordering and applicability. The ablation stages are constructed by merging these mechanisms, which is what lets the paper attribute efficiency differences to decomposition.

What would settle it

Give a single-mechanism learner—for example, a decision tree or deep RL agent—the same object-based state representation, the same primitive functions as an action vocabulary, and the same demo annotations, then measure problems to under 10% error on the same two tutor tasks. If it matches DIPL's 20-problem mastery, the decomposition claim is falsified; if it still needs thousands, the decomposition claim is supported.

Watch

Extended reading notes

Core claim

The central claim is that Decomposed Inductive Procedure Learning (DIPL), which induces production-rule skills with separate how-learning, where-learning, and when-learning mechanisms, can learn academic procedures with human-level data efficiency—under 10% error within about 20 problems—while single-mechanism approaches require orders of magnitude more data. On the fractions and multi-column addition tutors, the paper reports DIPL at 20 and 19 problems to mastery, versus 11,315 and 9,496 for deep Q-learning with demos, and 1,944 and 7,816 for a decision tree with demos. The ablation from one to two to three mechanisms shows that each decomposition stage improves efficiency, and the authors argue the key factor is the integration of functionally distinct learning systems—each simplifying error attribution for the others—rather than symbolic representation per se.

Load-bearing premise

The speedup is attributed to mechanism decomposition, but the DIPL agents also receive object-based states, primitive arithmetic functions, and demo argument annotations that the RL and decision-tree agents do not; the conclusion assumes these extra inductive resources are not the true cause.

Editorial extensions

If this is right

  • A machine learner that separates action generation, applicability matching, and action sequencing can reach under 10% error on these tutor tasks in roughly 20 problems, within a small constant of human mastery rates.
  • Ablating one mechanism at a time costs several orders of magnitude in data efficiency, even when the learner is symbolic; decomposition, not symbol use, produces the jump.
  • Adding demo worked examples helps, but not enough: deep RL with demos still needs about 10,000 problems.
  • Relative featurization—restating state features relative to the selection and arguments—contributes 13–19 problems' worth of efficiency in these tasks.
  • Adding a fourth mechanism, such as hierarchical task-network induction, may push efficiency beyond what 3-mechanism DIPL achieves.

Reading between the lines

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

  • If the confound is resolved and decomposition still drives the gain, the result suggests a design rule for sample-efficient agents: localize credit assignment into separate mechanisms rather than tuning a single global policy.
  • A natural extension is to let each mechanism operate over neural representations—for example, learned visual features for where-learning—which could test whether the decomposition principle survives when perception is not hand-coded.
  • The comparison leaves open whether the efficiency gap is primarily a representation gap; the headline mechanism claim should not be treated as established until a matched-representation ablation is run.
  • The same decomposition could be tested on less structured procedural domains, such as algebraic manipulation or data wrangling, where primitive functions are harder to enumerate; failure there would mark the boundary of the approach.
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 / 4 minor

Summary. The paper proposes Decomposed Inductive Procedure Learning (DIPL), a multi-mechanism symbolic learning approach inspired by prior simulated-learner architectures, and compares it against single-mechanism baselines (PPO, DQN, decision trees) on two intelligent tutoring system tasks: fraction arithmetic and multi-column addition. The central claim is that decomposing learning into distinct how-, where-, and when-learning mechanisms is the key factor enabling human-like data efficiency, with DIPL reaching under 10% error in roughly 20 problems while the reinforcement learning baselines require thousands to tens of thousands. The paper reports an ablation from 1-mechanism learning through 2-mechanism (how-learning plus a single left-hand-side classifier) to 3-mechanism DIPL, and includes human learning curves for the fraction task.

Significance. If the attribution were cleanly established, the result would be significant for both cognitive science and machine learning: it would provide concrete evidence that modular, mechanism-decomposed learning can close the data-efficiency gap between human learners and deep RL on structured academic tasks. The paper has clear strengths: it uses two realistic ITS environments, compares against multiple baseline families, reports convergence failures honestly, and builds on an established cognitive architecture. The ablation framing is novel and addresses a question that is often discussed only qualitatively. However, the central claim is currently not cleanly supported because the efficiency difference between the 1-mechanism baselines and DIPL is confounded with differences in representation, prior knowledge, action space, and demo annotations. The significance is therefore conditional on additional matched experiments that isolate the number of learning mechanisms.

major comments (4)
  1. [Task Domains / Ablation Analysis, Table 1] The headline comparison between 1-mechanism learners and DIPL is confounded across multiple dimensions. The RL and decision-tree baselines receive one-hot state vectors and predefined flat action spaces, while DIPL agents receive object-based states, no predefined action space, and domain-specific primitive functions (Add, Multiply, OnesDigit, TensDigit, Add3) plus demo argument annotations in multi-column addition. The orders-of-magnitude gap in Table 1 could therefore be driven by these extra inductive resources rather than by the number of learning mechanisms. The abstract's claim that decomposition is 'the key factor' requires matched conditions, for example RL with object-based states and primitive actions, or DIPL with one-hot states and no priors.
  2. [Results, Table 1] The claim that 'each stage of ablation yields several orders of magnitude of learning efficiency improvement' is not supported by the 2-mechanism versus 3-mechanism contrast. In fractions, the 2-mechanism How+LHS model reaches <10% error in 17 problems, faster than DIPL's 20; in multi-column addition, the improvement from 270 to 19 is large, but the DIPL condition includes relative featurization, an additional representational resource, and the version without relative featurization still improves from 270 to 38. These numbers suggest the main efficiency gain comes from how-learning plus a single LHS classifier, not from the full three-way decomposition.
  3. [Results, human data] The 6-problem shift applied to the human learning curves is a post hoc adjustment, and the paper provides no criterion or sensitivity analysis for choosing this offset. The claim that DIPL is 'in line with human learning' depends directly on this adjustment. Please report the unshifted comparison, justify the offset with an explicit model of unobserved practice, and show how the mastery intercept changes under a range of plausible offsets.
  4. [Discussion] The claim that RL agents with domain-specific primitive actions still require thousands of episodes is delegated to an external citation (MacLellan & Gupta, 2021) rather than being demonstrated in the present ablation. Since this point is directly relevant to the confound between mechanism decomposition and prior knowledge, the paper should either include such a condition in the reported experiments or clearly state that this aspect of the central claim rests on prior work.
minor comments (4)
  1. [Related Work] The text contains a typo: 'domain specifc' should be 'domain-specific'.
  2. [Discussion] The phrase 'about as data-efficient as human learning' is used in the Introduction but the operational comparison with human data appears only in the Results; consider clarifying the definition of data efficiency when humans are the baseline.
  3. [Figure 6] The caption describes human curves as 'offset' but does not state the offset value; please add this detail to the caption.
  4. [Introduction] The paper defines DIPL as encompassing both 3-mechanism and 4-mechanism approaches, but the experiments only instantiate the 3-mechanism variant; the text should state this limitation at first use to avoid overgeneralization.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DIPL's data-efficiency results are empirical ablation outcomes, not derivations from their own inputs.

full rationale

The paper's central claim is an empirical ablation result, not a derivation that reduces to its inputs. DIPL, the 2-mechanism model, decision trees, DQN, and PPO are independently specified learners, and the problem counts in Table 1 are measured simulation outcomes rather than fitted constants. The primitive prior-knowledge functions, demo annotations, and object-based representations given to DIPL are inputs to the systems, but the learning curves are not constructed from those inputs by definition; they come from running the learners. The human-data comparison is weakened by the post hoc 6-problem shift used to align initial error rates, but this is an analytic adjustment of the human curve, not a fitted parameter that is later called a prediction, and the DIPL curve itself is not fit to human data. The main self-citations, such as MacLellan and Gupta (2021) for RL with primitive actions, are supporting prior results rather than the load-bearing derivation of the ablation. Concerns about confounded resources between DIPL and RL baselines are validity threats about attribution of the efficiency gain, not circularity in the sense of output being equivalent to input by construction. No equation or definition in the paper makes the measured performance equal to the experimental setup.

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

The central empirical comparison is between systems that differ along multiple dimensions at once: mechanism count, prior knowledge, state representation, and action-space specification. The listed free parameters and assumptions are the most consequential of those differences for interpreting the paper's main claim.

free parameters (4)
  • Human data shift offset = 6 problems
    Applied to the human fractions learning curve to align initial 30 percent error with the simulated learner; the resulting 9-14 problem mastery estimate depends on this hand-chosen offset.
  • Primitive function sets per domain = Fractions: Add, Multiply, Equals; Multi-column addition: Add, Add3, OnesDigit, TensDigit
    Hand-selected so how-learning can compose the target procedures; RL agents are not given these building blocks, confounding the mechanism-decomposition comparison.
  • How-learning search depth limit = 1 to 3
    Composition search is capped at depths 1 to 3, constraining the space of learnable skill definitions and affecting measured efficiency.
  • Mastery threshold = Average error below 10 percent
    Used to convert learning curves to a single data-efficiency number; comparisons can shift if this threshold changes.
assumptions (4)
  • domain assumption The hand-coded primitive functions Add, Multiply, OnesDigit, TensDigit, Add3, and Equals are sufficient to compose correct how-parts for the target tasks.
    The paper supplies these functions to DIPL agents in the Task Domains section; the RL agents do not receive them, so the ablation does not control for this prior.
  • domain assumption A decision tree classifier is an adequate implementation of when-learning and of the LHS mechanism for measuring the effect of decomposition.
    Both the 2-mechanism and 3-mechanism agents use a decision tree; DIPL's advantage over the 2-mechanism agent in multi-column addition depends on this specific implementation.
  • domain assumption The two RL baselines, DQN and PPO trained for 500,000 timesteps, fairly represent the data efficiency of data-driven deep learning.
    The paper generalizes from two algorithms to 'data-driven deep learning' without a broader survey or sample-efficiency baselines.
  • domain assumption The human tutoring data from Patel, Liu, and Koedinger (2016) is comparable to the simulated fractions environment after a 6-problem shift.
    The human-like efficiency claim relies on this shifted comparison, but environment details and practice protocols may differ.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decomposed Inductive Procedure Learning: Learning Academic Tasks with Human-Like Data Efficiency." pith.science (2026). https://pith.science/paper/WNCY7JZJ

@misc{pith2026250510422,
  author       = {Pith},
  title        = {Pith review of: Decomposed Inductive Procedure Learning: Learning Academic Tasks with Human-Like Data Efficiency},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNCY7JZJ}},
  note         = {Machine review of arXiv:2505.10422}
}
read the original abstract

Human learning relies on specialization -- distinct cognitive mechanisms working together to enable rapid learning. In contrast, most modern neural networks rely on a single mechanism: gradient descent over an objective function. This raises the question: might human learners' relatively rapid learning from just tens of examples instead of tens of thousands in data-driven deep learning arise from our ability to use multiple specialized mechanisms of learning in combination? We investigate this question through an ablation analysis of inductive human learning simulations in online tutoring environments. Comparing reinforcement learning to a more data-efficient 3-mechanism symbolic rule induction approach, we find that decomposing learning into multiple distinct mechanisms significantly improves data efficiency, bringing it in line with human learning. Furthermore, we show that this decomposition has a greater impact on efficiency than the distinction between symbolic and subsymbolic learning alone. Efforts to align data-driven machine learning with human learning often overlook the stark difference in learning efficiency. Our findings suggest that integrating multiple specialized learning mechanisms may be key to bridging this gap.

Figures

Figures reproduced from arXiv: 2505.10422 by the authors.

Figure 1
Figure 1. Decomposition from 1-mechanism learning, like [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. An example of a where-part pattern generalized to act across columns in multi-column addition. where-learning generalizes from. How-learning attempts to explain each new worked example using existing skills’ how￾part compositions. If there are any candidate explanations, the one with arguments that would make the minimal change to an existing skill (quantified by a score that measures struc￾ture similarity) is used … view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Fraction arithmetic tutoring system for teaching [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Log-scale x-axis learning curves, for DQN-Demos, [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 30 canonical work pages

  1. [1]

    write newline

    " write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...

  2. [2]

    \ Schunn, C D

    anderson2013implications APACrefauthors Anderson, J R. \ Schunn, C D. APACrefauthors \ 2013 . Implications of the ACT-R learning theory: No magic bullets Implications of the act-r learning theory: No magic bullets . Advances in instructional Psychology, Volume 5 Advances in instructional psychology, volume 5 \ ( \ 1--33). Routledge

    work page 2013

  3. [3]

    APACrefauthors \ 1958

    bellman1958routing APACrefauthors Bellman, R. APACrefauthors \ 1958 . On a routing problem On a routing problem . Quarterly of applied mathematics 16 1 87--90

    work page 1958

  4. [4]

    , Chandrasekaran, V

    bubeck2023sparks APACrefauthors Bubeck, S. , Chandrasekaran, V. , Eldan, R. , Gehrke, J. , Horvitz, E. , Kamar, E. others APACrefauthors \ 2023 . Sparks of artificial general intelligence: Early experiments with gpt-4 Sparks of artificial general intelligence: Early experiments with gpt-4 . arXiv preprint arXiv:2303.12712

    arXiv 2023

  5. [5]

    , Kosaraju, V

    cobbe2021training APACrefauthors Cobbe, K. , Kosaraju, V. , Bavarian, M. , Chen, M. , Jun, H. , Kaiser, L. others APACrefauthors \ 2021 . Training verifiers to solve math word problems, 2021 Training verifiers to solve math word problems, 2021 . URL https://arxiv. org/abs/2110.14168

    arXiv 2021

  6. [6]

    , Hendler, J A

    erol1994semantics APACrefauthors Erol, K. , Hendler, J A. \ Nau, D S. APACrefauthors \ 1994 . Semantics for hierarchical task-network planning Semantics for hierarchical task-network planning . Citeseer

    work page 1994

  7. [7]

    , Burns, C

    hendrycks2021measuring APACrefauthors Hendrycks, D. , Burns, C. , Kadavath, S. , Arora, A. , Basart, S. , Tang, E. Steinhardt, J. APACrefauthors \ 2021 . Measuring mathematical problem solving with the math dataset Measuring mathematical problem solving with the math dataset . arXiv preprint arXiv:2103.03874

    arXiv 2021

  8. [8]

    , Urban, J

    kaliszyk2018reinforcement APACrefauthors Kaliszyk, C. , Urban, J. , Michalewski, H. \ Ol s \'a k, M. APACrefauthors \ 2018 . Reinforcement learning of theorem proving Reinforcement learning of theorem proving . Advances in Neural Information Processing Systems 31

    work page 2018

Show all 40 references
  1. [9]

    \ Alexandron, G

    kaser2024simulated APACrefauthors K \"a ser, T. \ Alexandron, G. APACrefauthors \ 2024 . Simulated learners in educational technology: A systematic literature review and a turing-like test Simulated learners in educational technology: A systematic literature review and a turin...

    2024

  2. [10]

    , Baker, R S

    koedinger2010data APACrefauthors Koedinger, K R. , Baker, R S. , Cunningham, K. , Skogsholm, A. , Leber, B. \ Stamper, J. APACrefauthors \ 2010 . A data repository for the EDM community: The PSLC DataShop A data repository for the edm community: The pslc datashop . Handbook of...

    2010

  3. [11]

    , Carvalho, P F

    koedinger2023astonishing APACrefauthors Koedinger, K R. , Carvalho, P F. , Liu, R. \ McLaughlin, E A. APACrefauthors \ 2023 . An astonishing regularity in student learning rate An astonishing regularity in student learning rate . Proceedings of the National Academy of Sciences...

    2023

  4. [12]

    , Corbett, A T

    koedinger2012knowledge APACrefauthors Koedinger, K R. , Corbett, A T. \ Perfetti, C. APACrefauthors \ 2012 . The Knowledge-Learning-Instruction framework: Bridging the science-practice chasm to enhance robust student learning The knowledge-learning-instruction framework: Bridg...

    2012

  5. [13]

    \ Tsitsiklis, J

    konda1999actor APACrefauthors Konda, V. \ Tsitsiklis, J. APACrefauthors \ 1999 . Actor-critic algorithms Actor-critic algorithms . Advances in neural information processing systems 12

    1999

  6. [14]

    APACrefauthors \ 2019

    laird2019soar APACrefauthors Laird, J E. APACrefauthors \ 2019 . The Soar cognitive architecture The soar cognitive architecture . MIT press

    2019

  7. [15]

    , Lebiere, C

    laird2017standard APACrefauthors Laird, J E. , Lebiere, C. \ Rosenbloom, P S. APACrefauthors \ 2017 . A standard model of the mind: Toward a common computational framework across artificial intelligence, cognitive science, neuroscience, and robotics A standard model of the min...

    2017

  8. [16]

    , Matsuda, N

    li2015integrating APACrefauthors Li, N. , Matsuda, N. , Cohen, W W. \ Koedinger, K R. APACrefauthors \ 2015 . Integrating representation learning and skill learning in a human-like intelligent agent Integrating representation learning and skill learning in a human-like intelli...

    2015

  9. [17]

    , Kosaraju, V

    lightman2023let APACrefauthors Lightman, H. , Kosaraju, V. , Burda, Y. , Edwards, H. , Baker, B. , Lee, T. Cobbe, K. APACrefauthors \ 2023 . Let's Verify Step by Step Let's verify step by step . arXiv preprint arXiv:2305.20050

    2023 arXiv

  10. [18]

    \ Gupta, A

    maclellan2021EDM APACrefauthors MacLellan, C J. \ Gupta, A. APACrefauthors \ 2021 . Learning Expert Models for Educationally Relevant Tasks Using Reinforcement Learning. Learning expert models for educationally relevant tasks using reinforcement learning. International Educati...

    2021

  11. [19]

    , Harpstead, E

    maclellan2016apprentice APACrefauthors Maclellan, C J. , Harpstead, E. , Patel, R. \ Koedinger, K R. APACrefauthors \ 2016 . The Apprentice Learner Architecture: Closing the Loop between Learning Theory and Educational Data. The apprentice learner architecture: Closing the loo...

    2016

  12. [20]

    \ Koedinger, K R

    maclellan2020domain APACrefauthors MacLellan, C J. \ Koedinger, K R. APACrefauthors \ 2020 . Domain-General Tutor Authoring with Apprentice Learner Models Domain-general tutor authoring with apprentice learner models . International Journal of Artificial Intelligence in Educat...

    2020

  13. [21]

    , Dumancic, S

    manhaeve2018deepproblog APACrefauthors Manhaeve, R. , Dumancic, S. , Kimmig, A. , Demeester, T. \ De Raedt, L. APACrefauthors \ 2018 . Deepproblog: Neural probabilistic logic programming Deepproblog: Neural probabilistic logic programming . Advances in neural information proce...

    2018

  14. [22]

    , Cohen, W W

    matsuda2015teaching APACrefauthors Matsuda, N. , Cohen, W W. \ Koedinger, K R. APACrefauthors \ 2015 . Teaching the Teacher: Tutoring SimStudent Leads to More Effective Cognitive Tutor Authoring Teaching the teacher: Tutoring simstudent leads to more effective cognitive tutor ...

    2015

  15. [23]

    \ Cohen, N J

    mccloskey1989catastrophic APACrefauthors McCloskey, M. \ Cohen, N J. APACrefauthors \ 1989 . Catastrophic interference in connectionist networks: The sequential learning problem Catastrophic interference in connectionist networks: The sequential learning problem . Psychology o...

    1989

  16. [24]

    , Alizadeh, K

    mirzadeh2024gsm APACrefauthors Mirzadeh, I. , Alizadeh, K. , Shahrokhi, H. , Tuzel, O. , Bengio, S. \ Farajtabar, M. APACrefauthors \ 2024 . Gsm-symbolic: Understanding the limitations of mathematical reasoning in large language models Gsm-symbolic: Understanding the limitatio...

    2024 arXiv

  17. [25]

    , Kavukcuoglu, K

    mnih2013playing APACrefauthors Mnih, V. , Kavukcuoglu, K. , Silver, D. , Graves, A. , Antonoglou, I. , Wierstra, D. \ Riedmiller, M. APACrefauthors \ 2013 . Playing atari with deep reinforcement learning Playing atari with deep reinforcement learning . arXiv preprint arXiv:1312.5602

    2013 arXiv

  18. [26]

    , Kavukcuoglu, K

    mnih2015dqn APACrefauthors Mnih, V. , Kavukcuoglu, K. , Silver, D. , Rusu, A A. , Veness, J. , Bellemare, M G. others APACrefauthors \ 2015 . Human-level control through deep reinforcement learning Human-level control through deep reinforcement learning . nature 518 7540 529--533

    2015

  19. [27]

    APACrefauthors \ 1985

    neves1985learning APACrefauthors Neves, D M. APACrefauthors \ 1985 . Learning Procedures from Examples and by Doing. Learning procedures from examples and by doing. IJCAI Ijcai \ ( \ 624--630)

    1985

  20. [28]

    , Liu, R

    patel2016block APACrefauthors Patel, R. , Liu, R. \ Koedinger, K R. APACrefauthors \ 2016 . When to Block versus Interleave Practice? Evidence Against Teaching Fraction Addition before Fraction Multiplication. When to block versus interleave practice? evidence against teaching...

    2016

  21. [29]

    , Dong, W

    poesia2021contrastive APACrefauthors Poesia, G. , Dong, W. \ Goodman, N. APACrefauthors \ 2021 . Contrastive reinforcement learning of symbolic reasoning domains Contrastive reinforcement learning of symbolic reasoning domains . Advances in neural information processing system...

    2021

  22. [30]

    \ Cameron-Jones, R M

    quinlan1995induction APACrefauthors Quinlan, J R. \ Cameron-Jones, R M. APACrefauthors \ 1995 . Induction of logic programs: FOIL and related systems Induction of logic programs: Foil and related systems . New Generation Computing 13 287--312

    1995

  23. [31]

    , Tehranchi, F

    ritter2019act APACrefauthors Ritter, F E. , Tehranchi, F. \ Oury, J D. APACrefauthors \ 2019 . ACT-R: A cognitive architecture for modeling cognition Act-r: A cognitive architecture for modeling cognition . Wiley Interdisciplinary Reviews: Cognitive Science 10 3 e1488

    2019

  24. [32]

    , Wolski, F

    schulman2017proximal APACrefauthors Schulman, J. , Wolski, F. , Dhariwal, P. , Radford, A. \ Klimov, O. APACrefauthors \ 2017 . Proximal policy optimization algorithms Proximal policy optimization algorithms . arXiv preprint arXiv:1707.06347

    2017 arXiv

  25. [33]

    , Georgiev, P

    team2024gemini APACrefauthors Team, G. , Georgiev, P. , Lei, V I. , Burnell, R. , Bai, L. , Gulati, A. others APACrefauthors \ 2024 . Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context Gemini 1.5: Unlocking multimodal understanding across milli...

    2024 arXiv

  26. [34]

    APACrefauthors \ 1990

    vanlehn1990mind APACrefauthors VanLehn, K. APACrefauthors \ 1990 . Mind bugs: The origins of procedural misconceptions Mind bugs: The origins of procedural misconceptions . MIT press

    1990

  27. [35]

    , Ohlsson, S

    vanlehn1994applications APACrefauthors VanLehn, K. , Ohlsson, S. \ Nason, R. APACrefauthors \ 1994 . Applications of simulated students: An exploration Applications of simulated students: An exploration . Journal of artificial intelligence in education 5 135--135

    1994

  28. [36]

    , Harpstead, E

    weitekamp2024ai2t APACrefauthors Weitekamp, D. , Harpstead, E. \ Koedinger, K. APACrefauthors \ 2024 . AI2T: Building Trustable AI Tutors by Interactively Teaching a Self-Aware Learning Agent Ai2t: Building trustable ai tutors by interactively teaching a self-aware learning ag...

    2024 arXiv

  29. [37]

    , Harpstead, E

    weitekamp2019toward APACrefauthors Weitekamp, D. , Harpstead, E. , MacLellan, C J. , Rachatasumrit, N. \ Koedinger, K R. APACrefauthors \ 2019 . Toward Near Zero-Parameter Prediction Using a Computational Model of Student Learning. Toward near zero-parameter prediction using a...

    2019

  30. [38]

    , Rachatasumrit, N

    weitekamp2023simulating APACrefauthors Weitekamp, D. , Rachatasumrit, N. , Wei, R. , Harpstead, E. \ Koedinger, K. APACrefauthors \ 2023 . Simulating Learning from Language and Examples Simulating learning from language and examples . International Conference on Artificial Int...

    2023

  31. [39]

    weitekamp2020investigating APACrefauthors Weitekamp, D. , Ye, Z. , Rachatasumrit, N. , Harpstead, E. \ Koedinger, K. APACrefauthors \ 2020 . Investigating Differential Error Types Between Human and Simulated Learners Investigating differential error types between human and sim...

    2020

  32. [40]

    \ Zhang, D

    xiao2023deep APACrefauthors Xiao, Z. \ Zhang, D. APACrefauthors \ 2023 . A deep reinforcement learning agent for geometry online tutoring A deep reinforcement learning agent for geometry online tutoring . Knowledge and Information Systems 65 4 1611--1625

    2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.