Pith. sign in

REVIEW 3 major objections 4 minor 39 references

Retraining Seeks Stable Signals

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Retraining provably converges to model-independent stable signals even when performative feedback is arbitrarily strong.

desk verdict Clean, correct affine retraining theory whose abstract overstates the reach: the stable-signal principle holds only under linear performative effects, not for any nonzero stable signal. read the letter →

arxiv 2607.15623 v1 pith:PKGSPOGD submitted 2026-07-17 stat.ML cs.LG

Moritz Hardt This is my paper
classification stat.MLcs.LG MSC 68T0562J05
keywords stablesignalprincipleperformativepredictionrepeatedriskminimizationretrainingdynamicsregularizationfeedbackmodelcollapsedataloops
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 seeks to establish a 'stable signal principle' for repeated risk minimization: whenever the prediction target has even a small model-independent component, suitably regularized or normalized retraining converges to the direction of that component, regardless of how strongly the deployed model shapes future data. The proof centers on an explicit affine recursion for the population retraining update, which yields a unique fixed point along the stable-signal direction and geometric convergence whenever regularization exceeds the performative feedback strength. If correct, the result reframes regularization as a force that controls performativity, and explains why data feedback loops in language modeling can stay stable when a fixed fraction of fresh real data is retained. The analysis extends beyond squared loss to logistic and softmax losses, time-varying operator-valued feedback, normalized deployment, and parametric stochastic-gradient retraining.

What carries the argument

The load-bearing mechanism is the affine population retraining map. Whenever the loss-optimal prediction given a deployed predictor h equals s(x) + (Bh)(x) for some fixed signal s and bounded linear operator B, population retraining reduces to T(h) = s + Bh; for regularized least squares with conditional mean α f* + β h, this becomes the scalar recursion f_{t+1} = (α/(1+δ)) f* + (β/(1+δ)) f_t. The fixed point and geometric error contraction follow directly by subtracting the fixed-point equation. A second mechanism, used for normalization, is the half-angle identity tan(φ_{t+1}/2) ≤ (β/(α+β)) tan(φ_t/2), which yields angular convergence to the stable direction without controlling scale.

What would settle it

In a deployed system where a stable signal can be measured independently (for instance, item quality ratings), fit the conditional mean E[Y|X,h] as a function of the deployed predictor h and test whether it is affine. If a quadratic or threshold term is needed, repeated regularized retraining will have a fixed point that deviates from the f* direction and the error will not decay at the predicted rate |β|/(1+δ). Alternatively, in an LM data loop set λ_t = 1 - 1/t^2 (so Σ(1-λ_t) converges): the paper predicts the loop will not converge to the real-data distribution, so observing the distance to

Watch

Extended reading notes

Core claim

The paper proves that under the conditional-mean decomposition E_{D(h)}[Y|X] = α f*(X) + β h(X), repeated regularized least squares follows f_{t+1} = (α/(1+δ)) f* + (β/(1+δ)) f_t, with unique fixed point f_δ = α/(1+δ-β) f* and error f_t - f_δ = (β/(1+δ))^t (f_0 - f_δ). Geometric convergence holds exactly when |β| < 1+δ; hence any positive feedback strength can be stabilized by choosing δ > β-1. The same affine structure drives the other results: two consecutive updates identify β and recover the unregularized performatively stable point; normalized retraining contracts the half-angle to the stable signal at rate β/(α+β) under arbitrarily strong positive feedback; and in language-model data l

Load-bearing premise

The results all rest on the assumption that the target's conditional mean decomposes additively into a stable model-independent signal plus a term that is linear (affine) in the deployed predictor; if the performative response is nonlinear in h, the fixed point need not lie along the stable-signal direction and the geometric convergence claims do not follow.

Editorial extensions

If this is right

  • If the stable signal principle is right, retraining under strong positive performativity settles on the direction of the model-independent part of outcomes: rankings of items by quality survive even when 90% of clicks are driven by the model's own display rank.
  • Regularization gains a new role: choosing an ℓ2 penalty δ > β-1 makes retraining contractive, so regularization can be tuned as a performativity-control knob rather than only an overfitting remedy.
  • With three consecutive population iterates, the feedback strength β and the unregularized performatively stable point can be recovered in closed form, so one can jump to the stable point after only two retraining steps.
  • Under normalized deployment, the angle to the stable signal contracts geometrically at rate β/(α+β) for any positive feedback, and in high dimension a random initialization is almost surely in the basin of attraction.
  • In language-model data loops, convergence to the real-data distribution is geometric for a fixed fresh-data fraction; a shrinking fresh fraction still guarantees convergence exactly when the series of (1-λ) diverges, and accumulating all past synthetic data slows the rate to polynomial.

Reading between the lines

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

  • A testable extension: if the conditional mean is nonlinear in the deployed predictor, the stable-signal principle predicts the fixed point will drift off the f* direction, so measuring this nonlinearity in a deployed system can separate regimes where the theorem applies from regimes needing a different theory.
  • The theory's 'direction, not magnitude' conclusion suggests that under strong performativity absolute predicted values are not recoverable, but relative orderings are; systems that only need rankings may therefore be inherently more stable than systems that require calibrated scores.
  • For language modeling, the paper's mixture analysis implies a practical monitoring rule: track the cumulative sum of fresh-data fractions; if it diverges, the loop should converge to real-data statistics, and if it converges, drift toward the synthetic distribution is to be expected even in the population limit.
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

3 major / 4 minor

Summary. The paper proposes a 'stable signal principle' for repeated risk minimization under performativity: if the prediction target has a model-independent component f*, then suitably regularized or normalized retraining converges to the direction of this stable signal, even when the model's own influence on the target is arbitrarily strong. The main technical setting is population least squares with the conditional-mean assumption E_{D(h)}[Y|X] = α f*(X) + β h(X) (Section 3, Eq. (1)). Under this assumption, the paper derives the affine update f_{t+1} = (α/(1+δ)) f* + (β/(1+δ)) f_t, proves geometric convergence when |β| < 1+δ, and gives a two-step identification of β and of the unregularized performatively stable point. It then extends the affine theory to heterogeneous and time-varying coefficients, to general affine retraining operators, to logistic and multiclass losses with affine logits, and to nonlinear response links whose minimizer remains affine in the deployed predictor. A separate normalization argument (Section 5) shows angular contraction via the power-method-like bound tan(φ_t/2) ≤ (β/(α+β))^t tan(φ_0/2). The final sections apply the stable-signal perspective to language-model data feedback loops and to parametric stochastic-gradient retraining.

Significance. If the results are taken as applying to the stated linear/affine performative models, the paper makes a useful and elegant contribution. The recursions are closed-form and the proofs are short and easily checkable; the two-step decoding of the performative feedback coefficient is a neat identification result, and the normalization argument gives a clean geometric intuition for why direction can survive strong positive feedback. The paper also connects the stable-signal perspective to existing observations about language-model self-training, showing when initialization effects decay under mixture and accumulating-pool feedback. However, the broad 'stable signal principle' advertised in the abstract and conclusion is not established beyond the affine-in-h conditional-mean assumption; the paper's own technical sections are clear about this assumption, but the framing overclaims its scope. The value of the work is therefore conditional on the linearity of performative dependence, which is an extra structural assumption rather than a consequence of the existence of a stable signal.

major comments (3)
  1. [Abstract & §3, Eq. (1)] The headline claim—that a nonzero stable signal guarantees geometric convergence to its direction under arbitrarily strong performative effects—is proven only under the affine conditional-mean ansatz E_{D(h)}[Y|X] = αf*(X) + βh(X). This linearity in h is what makes the recursion affine and its fixed point collinear with f*. If the performative dependence is nonlinear, e.g. E[Y|X] = αf* + βh + γh^2, then the induced update f_{t+1} = αf* + βf_t + γf_t^2 has fixed points that are not proportional to f* (for X∈{0,1}, f*=(1,2), α=0.1, β=0.5, γ=0.1). Thus the abstract's unconditional phrasing overstates what is proved. The abstract and conclusion should be qualified to linear/affine performative effects, or the paper should give additional conditions under which a stable signal alone (without linearity) guarantees attraction.
  2. [§4.1, Corollaries 8–10] The heading 'Nonlinear performative effects' is potentially misleading. The corollaries allow nonlinear response functions ψ (e.g., sigmoid or σ(a+γ sin a)), but the loss-optimal prediction remains an affine function s(x)+(Bh)(x) of the deployed predictor, as required by Eq. (7). The dependence of the induced distribution on h is still linear through the logits or scores. These results therefore do not address nonlinear dependence of the conditional mean on h; the counterexample in the first major comment remains outside their scope. Please clarify that 'nonlinear' refers to the link function, not to the performative map h ↦ conditional mean.
  3. [§6, Props. 14–15] The language-modeling section is explicitly at the population level, but the practical conclusions—geometric vs. polynomial convergence—are presented in a way that could be read as a statement about actual LLM retraining. Real training is finite-sample, parametric, and optimization-constrained; the identity P_{t+1}=Q_t and the mixture equations hold only for population cross-entropy minimization. The paper should state more prominently that these results describe an idealized population loop and are not direct guarantees for finite-sample, gradient-based LLM training. This is a limitation of scope rather than a technical error, but it is load-bearing for the section's claims about 'stability of language model training.'
minor comments (4)
  1. [§3.1, Prop. 4] In Eq. (3), the decoded stable point f_PS is undefined when 1−(1+δ)b_t = 0, which corresponds to β=1. This case is mentioned in the text, but the proposition statement could explicitly exclude it to avoid division-by-zero ambiguity.
  2. [§5, Eq. (15)] The notation for f_t in the normalization section is a little unclear: f_t is the unnormalized iterate and z_t = f_t/∥f_t∥ is the deployed predictor, but the recursion is written only for z_{t+1}. Briefly defining f_t as the pre-normalization iterate would improve readability.
  3. [§6.2, Prop. 15] The assumption that the initial pool and all batches share the same context distribution is stated before the proposition, but it deserves more emphasis, since real data pools typically have shifting context distributions; a sentence discussing the dependence on this assumption would be helpful.
  4. [Global] There are a few places where the phrase 'direction of the stable signal' is used loosely. In the heterogeneous case (Section 3.2) the limit f_δ(x)=s(x)/(1+δ−β(x)) is not generally a global scalar multiple of a single f*, so 'direction' is only meaningful pointwise. Please add a short clarification to prevent misreading.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; all theorems are direct consequences of explicitly stated structural assumptions.

full rationale

The paper's central results are derived from explicitly stated modeling assumptions—the affine conditional-mean decomposition E[Y|X]=αf*+βh (Section 3), the general affine conditional-risk minimizer condition (Prop. 6, Eq. 7), the normalization dynamics (Eq. 15), and the mixture response model for language data (Eq. 18). The abstract's language about 'stable signals' is a label for the model-independent component f*, but f* is not defined as the limit of retraining; convergence to f* is proven from the recursions. Prop. 3, Thm. 11, Props. 14 and 15 are self-contained algebraic derivations, not restatements of hidden definitions. Prop. 4 is an exact identifiability result given the model, not a statistical fit or prediction tested against data. Self-citations are background and not load-bearing for any theorem. The main weakness—that the headline conclusion is only as general as the affine/mixture assumptions, which are stronger than the abstract's generic 'model-independent component'—is a scope/overclaim issue, not a circularity of the derivation chain. No equation is used to prove what it itself assumes, and no fitted parameter is re-presented as a prediction. Therefore the paper's derivation is self-contained and non-circular.

Assumptions & free parameters 5 free parameters · 8 assumptions · 1 invented entities

The paper's central claim rests on a stipulated affine decomposition of the conditional mean into a stable component αf* and a performative component linear in the deployed predictor (βh or B_t h). This ansatz, rather than measured data, supplies the stable signal: it is the entity the paper introduces and the reason its conclusions hold. The main free ingredients are the feedback coefficient β (or operators B_t), the stable-signal strength α, the regularization choice δ, and — in the LM section — mixture fractions λ_t and data-addition schedules; none are fitted, but the theory is fully contingent on their assumed values. The remaining axioms are common-support conditions, eigenvector coupling of B_t, standard SGD moment bounds, and classical analysis facts.

free parameters (5)
  • β (performative feedback coefficient)
    Assumed constant in §3, pointwise function β(x) in §3.2, operator B_t in §4, scalar in §7, and common eigenvalue β_t in Thm. 12. Sets the contraction threshold δ > β-1 and, with α, the fixed point's magnitude. Not fitted in the theory; Prop. 4 identifies it from three population iterates.
  • α (stable-signal strength)
    Assumed > 0; the paper's central postulate. Scales the attracting fixed point (α/(1+δ-β))f*. In §3.2 only the product s(x) = α(x)f*(x) is identifiable without additional normalization.
  • δ (regularization penalty) = δ = 2β (suggested)
    Design choice that makes retraining contractive for β ≥ 0; the paper explicitly frames regularization as the control knob for performativity. All convergence conditions are stated as functions of δ, and §3.1 shows how to decode the unregularized stable point from the regularized one.
  • λ_t (synthetic-data fraction)
    Exogenous modeling sequence in the LM mixture model (Eq. 18); the convergence criterion is ∏λ_s = 0 iff ∑(1-λ_s) = ∞ (Prop. 14). Not estimated from data.
  • n_real^t, n_gen^t (per-round data additions)
    Schedules of real and generated data added to the training pool (Prop. 15); Eq. (23) gives the convergence criterion ∑ n_real^t/N_{t+1} = ∞ and Remark 16 the polynomial rate t^{-r/(r+g)}. Chosen exogenously.
assumptions (8)
  • domain assumption Conditional mean under deployment is affine in the deployed predictor with an additive stable component: E_{D(h)}[Y|X] = αf*(X) + βh(X)
    Load-bearing premise of the paper (Section 3, first display; generalized in Prop. 6, Eq. (7)). Everything downstream — the affine recursion (1), the fixed point along f*, the decoding — follows from this additivity.
  • domain assumption A nonzero stable signal exists: α > 0 and f* ≠ 0
    The abstract's central condition. If α = 0 the fixed-point equation degenerates (e.g., β = 1 leaves no stable point), and there is nothing for retraining to converge to.
  • domain assumption The feature marginal D_X(h) shares the same null sets as D_0 (common support)
    Used in the proofs of Claim 2 and Prop. 6 to argue that marginal reweighting w_t(x) > 0 does not change the pointwise minimizer, so 'arbitrary model-induced feature changes' do not affect the update rule.
  • domain assumption Conditional risk minimizer is affine: argmin_{a∈R} r_{t,h}(a,x) = s_t(x) + (B_t h)(x) (Eq. 7)
    The generalized assumption behind Section 4; Corollaries 8-10 verify it for logistic, softmax, and injective-link losses but always keep the dependence on h affine through the logit or score.
  • domain assumption In Thm. 12, the stable signal u is a common left and right eigenvector of B_t (Eq. 16): B_t u = β_t u and B_t* u = β_t u
    Restricts operator-valued feedback; the angular contraction q_t = M_t/(α_t + β_t) depends on this eigencoupling to keep B_t v_t orthogonal to u.
  • domain assumption For Prop. 17: uniform eigenvalue lower bound Σ(θ) ⪰ κI, a.s. norm bound ‖X‖ ≤ R, and bounded noise variance E[ξ²|X,θ] ≤ ν²
    Standard stochastic-approximation conditions required for the mean-square convergence of one-sample SGD retraining under the 'Greedy Deploy' protocol.
  • domain assumption Population cross-entropy retraining makes the next model equal to the conditional distribution of the training mix: P_{t+1} = Q_t
    The LM section's idealization (Eq. 18 and the proof of Prop. 14). Ignores finite samples, parametric constraints, and optimization error of real LLM training; the paper explicitly calls it the population version.
  • standard math Standard analytic facts: geometric convergence of linear recurrences; contraction arguments; ∏λ_s = 0 iff ∑(1-λ_s) = ∞ for λ_s ∈ (0,1); power-method half-angle dynamics
    Background used without proof in Props. 3, 5, 14, 15 and Thm. 11.
invented entities (1)
  • Stable signal f* (u in §5; P* in §6)
    purpose: Model-independent component of the prediction target that serves as the attractor of retraining; e.g., 'intrinsic quality' of an item, or the real-data conditional distribution in LM loops.
    Postulated by the affine decomposition E[Y|X] = αf* + βh; no measurement or falsifiable prediction outside the framework is provided. The two-step decoding recovers αf* only under the very model that defines it; the internal consistency check (decoded β stable across steps) is not independent evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Retraining Seeks Stable Signals." pith.science (2026). https://pith.science/paper/PKGSPOGD

@misc{pith2026260715623,
  author       = {Pith},
  title        = {Pith review of: Retraining Seeks Stable Signals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PKGSPOGD}},
  note         = {Machine review of arXiv:2607.15623}
}
read the original abstract

Predictive models deployed at scale influence future data, a phenomenon called performativity. And there is always one way to cope: Train the model on new data, deploy it again, and repeat. This process, called retraining or repeated risk minimization, creates a feedback loop between model and data that real-world learning systems can't avoid. Results on performative prediction shed light on this dynamic: If the model's influence on the data is small, retraining reaches a fixed point. What remains open is why fixed points should naturally exist, and what governs retraining when the model's influence is strong. In this work we develop a new perspective on retraining -- the stable signal principle -- that addresses these questions. We start from the assumption that the prediction target has at least some small model-independent component, a stable signal, such as the intrinsic quality of an item. We prove that when a nonzero stable signal exists, repeated risk minimization, suitably regularized, converges geometrically to the direction of this stable signal. This is true even if the model's influence on the target is arbitrarily large relative to the stable signal. Regularization emerges naturally as a force to control performativity, rather than to promote generalization, revealing a new facet of an old concept. We extend the analysis to a broad family of affine retraining operators under arbitrary model-induced feature changes, heterogeneous time-varying effects, and nonlinear responses. The stable signal perspective also applies to data feedback loops in language modeling, providing new explanations for the stability of language model training from model-generated data.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 2 linked inside Pith

  1. [1]

    Perdomo and Tijana Zrnic and Celestine Mendler-D

    Juan C. Perdomo and Tijana Zrnic and Celestine Mendler-D. Performative prediction , booktitle =

  2. [2]

    Oskar Morgenstern , title =

  3. [3]

    Merton , title =

    Robert K. Merton , title =. The Antioch Review , volume =

  4. [4]

    Journal of Political Economy , volume =

    Emile Grunberg and Franco Modigliani , title =. Journal of Political Economy , volume =

  5. [5]

    Simon , title =

    Herbert A. Simon , title =. The Public Opinion Quarterly , volume =

  6. [6]

    A survey on concept drift adaptation , journal =

    Jo. A survey on concept drift adaptation , journal =

  7. [7]

    Innovations in Theoretical Computer Science , pages =

    Moritz Hardt and Nimrod Megiddo and Christos Papadimitriou and Mary Wootters , title =. Innovations in Theoretical Computer Science , pages =

  8. [8]

    Kim and Juan C

    Michael P. Kim and Juan C. Perdomo , title =. Innovations in Theoretical Computer Science , year =

Show all 39 references
  1. [9]

    Performative prediction:

    Moritz Hardt and Celestine Mendler-D. Performative prediction:. Statistical Science , volume =

  2. [10]

    ACM Computing Surveys , volume =

    Thomas Kehrenberg and Francisco Javier Sanguino Bautiste and Jose Lozano and Novi Quadrianto , title =. ACM Computing Surveys , volume =

  3. [11]

    International Conference on Machine Learning , pages =

    Tatsunori Hashimoto and Megha Srivastava and Hongseok Namkoong and Percy Liang , title =. International Conference on Machine Learning , pages =

  4. [12]

    Stochastic optimization for performative prediction , booktitle =

    Celestine Mendler-D. Stochastic optimization for performative prediction , booktitle =

  5. [13]

    SIAM Review , volume =

    Optimization Methods for Large-Scale Machine Learning , author =. SIAM Review , volume =

  6. [14]

    Mathematics of Operations Research , volume =

    Dmitriy Drusvyatskiy and Lin Xiao , title =. Mathematics of Operations Research , volume =

  7. [15]

    International Conference on Artificial Intelligence and Statistics , pages =

    Gavin Brown and Shlomi Hod and Iden Kalemaj , title =. International Conference on Artificial Intelligence and Statistics , pages =

  8. [16]

    International Conference on Artificial Intelligence and Statistics , pages =

    Zachary Izzo and James Zou and Lexing Ying , title =. International Conference on Artificial Intelligence and Statistics , pages =

  9. [17]

    International Conference on Artificial Intelligence and Statistics , pages =

    Qiang Li and Hoi-To Wai , title =. International Conference on Artificial Intelligence and Statistics , pages =

  10. [18]

    Ratliff and Dmitriy Drusvyatskiy and Maryam Fazel , title =

    Mitas Ray and Lillian J. Ratliff and Dmitriy Drusvyatskiy and Maryam Fazel , title =. AAAI Conference on Artificial Intelligence , pages =

  11. [19]

    Ratliff , title =

    Adhyyan Narang and Evan Faulkner and Dmitriy Drusvyatskiy and Maryam Fazel and Lillian J. Ratliff , title =. Journal of Machine Learning Research , volume =

  12. [20]

    International Conference on Artificial Intelligence and Statistics , pages =

    Mehrnaz Mofakhami and Ioannis Mitliagkas and Gauthier Gidel , title =. International Conference on Artificial Intelligence and Statistics , pages =

  13. [21]

    arXiv preprint arXiv:2606.07890 , year =

    Jaewook Lee and Tijana Zrnic , title =. arXiv preprint arXiv:2606.07890 , year =

    arXiv

  14. [22]

    arXiv preprint arXiv:2602.24207 , year =

    Gabriele Farina and Juan Carlos Perdomo , title =. arXiv preprint arXiv:2602.24207 , year =

    arXiv

  15. [23]

    International Conference on Machine Learning , pages =

    Zachary Izzo and Lexing Ying and James Zou , title =. International Conference on Machine Learning , pages =

  16. [24]

    Miller and Juan C

    John P. Miller and Juan C. Perdomo and Tijana Zrnic , title =. International Conference on Machine Learning , pages =

  17. [25]

    Regret minimization with performative feedback , booktitle =

    Meena Jagadeesan and Tijana Zrnic and Celestine Mendler-D. Regret minimization with performative feedback , booktitle =

  18. [26]

    International Conference on Machine Learning , pages =

    Licong Lin and Tijana Zrnic , title =. International Conference on Machine Learning , pages =

  19. [27]

    International Conference on Machine Learning , pages =

    Juan Carlos Perdomo , title =. International Conference on Machine Learning , pages =

  20. [28]

    Friedler and Scott Neville and Carlos Scheidegger and Suresh Venkatasubramanian , title =

    Danielle Ensign and Sorelle A. Friedler and Scott Neville and Carlos Scheidegger and Suresh Venkatasubramanian , title =. Conference on Fairness, Accountability and Transparency , pages =

  21. [29]

    Degenerate feedback loops in recommender systems , booktitle =

    Ray Jiang and Silvia Chiappa and Tor Lattimore and Andr. Degenerate feedback loops in recommender systems , booktitle =

  22. [30]

    International Conference on Machine Learning , pages =

    Rohan Taori and Tatsunori Hashimoto , title =. International Conference on Machine Learning , pages =

  23. [31]

    Nature , volume =

    Ilia Shumailov and Zakhar Shumaylov and Yiren Zhao and Nicolas Papernot and Ross Anderson and Yarin Gal , title =. Nature , volume =

  24. [32]

    Baraniuk , title =

    Sina Alemohammad and Josue Casco-Rodriguez and Lorenzo Luzi and Ahmed Imtiaz Humayun and Hossein Babaei and Daniel LeJeune and Ali Siahkoohi and Richard G. Baraniuk , title =. International Conference on Learning Representations , year =

  25. [33]

    A tale of tails:

    Elvis Dohmatob and Yunzhen Feng and Pu Yang and Fran. A tale of tails:. International Conference on Machine Learning , pages =

  26. [34]

    International Conference on Learning Representations , year =

    Quentin Bertrand and Avishek Joey Bose and Alexandre Duplessis and Marco Jiralerspong and Gauthier Gidel , title =. International Conference on Learning Representations , year =

  27. [35]

    Roberts and Diyi Yang and David L

    Matthias Gerstgrasser and Rylan Schaeffer and Apratim Dey and Rafael Rafailov and Henry Sleight and John Hughes and Tomasz Korbak and Rajashree Agrawal and Dhruv Pai and Andrey Gromov and Daniel A. Roberts and Diyi Yang and David L. Donoho and Sanmi Koyejo , title =. First Con...

  28. [36]

    International Conference on Machine Learning , pages=

    Delayed impact of fair machine learning , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  29. [37]

    ACM SIGKDD , pages=

    Stackelberg games for adversarial prediction problems , author=. ACM SIGKDD , pages=

  30. [38]

    Advances in Neural Information Processing Systems , year=

    An engine not a camera: Measuring performative power of online search , author=. Advances in Neural Information Processing Systems , year=

  31. [39]

    2022 , publisher =

    Ward, Jacob , title =. 2022 , publisher =

    2022

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.