← Proof Notes
Counterexample / Optimization & generalization

Calibration is insufficient for the TRADES surrogate bound

Theoretically Principled Trade-off between Robustness and Accuracy

3,147 citations ↗Semantic Scholar · 2026-09-09

2019 · arXiv:1901.08573v3, 24 June 2019; ICML 2019 paper and supplement · Reviewed 09 September 2026

An explicit example contradicts the selected statement as written.

Paper context

Overview

TRADES studies the tension between ordinary prediction accuracy and resistance to adversarial perturbations. It decomposes robust error into ordinary error and a boundary-related term, then uses surrogate losses to motivate a training objective balancing accurate predictions with stability under perturbation.

Role of the theoretical result

The selected upper bound is part of the theoretical justification for replacing the difficult robust-error objective with a tractable surrogate. Its assumptions determine how broadly that justification applies to alternative losses and input spaces.

Original paper ↗

01 / Summary

Summary of the result

The second inequality of Theorem 3.1 requires pointwise domination of the wrong-sign indicator by the surrogate loss. Classification calibration, nonnegativity and φ(0)≥1 do not imply φ(u)≥1 for all u≤0. A nonnegative C² calibrated loss with φ(−1)=1/2 violates the bound on the permitted two-point instance space. Non-increasing margin losses with the stated normalization satisfy the additional condition.

02 / Statement

Statement under review

Theorem 3.1, p6; Appendix B.1, p20. Instance space and relative adversarial balls are defined in Section 2.1. · paraphrased

Theorem 3.1 upper-bounds robust excess error for any nonnegative classification-calibrated loss with φ(0)≥1, any measurable predictor and the stated instance space.

Rrob(f)Rnatψ1(Rϕ(f)Rϕ)+EmaxxB(X,ϵ)ϕ ⁣(f(x)f(X)λ)R_{\rm rob}(f)-R_{\rm nat}^*\le\psi^{-1}(R_\phi(f)-R_\phi^*)+\mathbb E\max_{x'\in B(X,\epsilon)}\phi\!\left(\frac{f(x')f(X)}{\lambda}\right)
Relevant assumptions
  • Instance space X={−1,+1}⊂R, uniform data and deterministic label Y=X; f(x)=x.
  • Adversarial radius ε=2, with balls intersected with X as the paper defines; λ=1.
  • Define S(u)=0 for u≤0, S(u)=6u⁵−15u⁴+10u³ for 0<u<1, and S(u)=1 for u≥1.
  • Use φ(t)=1−S(t) for t≥0 and φ(t)=1/2+S(t+1)/2 for t<0. This is nonnegative, C² and satisfies φ(0)=1.

03 / Derivation

Counterexample and derivation

6 steps · complete derivation
  1. 01

    The chosen loss has φ(1)=0, φ(−1)=1/2 and φ(0)=1. Its first two derivatives match at the piecewise junctions.

  2. 02

    For conditional label probability η, the best positive score has risk (1−η)/2 and the best negative score has risk η/2. Therefore H(η)=min(η,1−η)/2.

  3. 03

    For η≠1/2, forcing the wrong sign gives H−(η)=max(η,1−η)/2>H(η), so the loss satisfies the printed calibration assumption. Its calibration transform is ψ(θ)=θ/2.

  4. 04

    The predictor classifies both points correctly with zero surrogate risk. The optimal natural risk and surrogate excess are both zero.

  5. 05

    Every adversarial ball contains the opposite point, so robust error is one. The boundary surrogate is max{φ(1),φ(−1)}=1/2.

  6. 06

    The claimed upper bound is consequently 1≤1/2. The proof's step from a wrong-sign indicator to φ(u) fails at u=−1.

Counterexample

The two-point construction has zero surrogate excess risk, robust error one and boundary surrogate 1/2. The second inequality therefore reduces to 1≤1/2.

ϕ(1)=12,ϕ(0)=1,ϕ(1)=0,ψ(θ)=θ2,1=RrobRnat>12\phi(-1)=\frac12,\quad \phi(0)=1,\quad \phi(1)=0,\quad \psi(\theta)=\frac{\theta}{2},\qquad 1=R_{\rm rob}-R_{\rm nat}^*>\frac12

04 / Implications

Implications and proposed correction

Theoretical implications

Affected result

Refutes the universal second inequality of Theorem 3.1 under its printed loss and instance-space assumptions. The first risk decomposition/calibration inequality is not refuted.

Empirical scope

Relation to reported experiments

Does not refute the usual TRADES algorithm or its benchmark results. Common non-increasing margin losses, with the stated normalization, satisfy the extra domination property.

Proposed correction

Sufficient conditions and revised bound

Add φ(u)≥1 for every nonpositive u. A sufficient, stronger condition is that φ is non-increasing and φ(0)≥1. Then the wrong-sign indicator is pointwise bounded by the surrogate, restoring the disputed step. For u≤0 the added condition supplies the required unit penalty; for u>0 the indicator is zero and nonnegativity suffices. Taking the same maxima and expectation preserves the inequality.

Implementation implications

When adapting the binary TRADES guarantee to a new loss or a discrete input domain, verify wrong-sign domination rather than calibration alone. No change to a standard loss that already satisfies this condition is implied.

Limits of this review

  • On a connected perturbation domain with a continuous predictor, an intermediate zero score may restore the boundary surrogate. The theorem explicitly permits the two-point domain used here.
  • The witness loss is deliberately non-monotone; the printed theorem does not require monotonicity.
  • The scope is binary population theory, not an empirical multiclass robustness certificate.
  • This repairs the selected upper-bound step. It does not independently establish the paper's lower-bound tightness theorem or its multiclass practical objective.

05 / References

Sources and correction history

  1. 01
    Current primary paper

    Section 2.1; Theorem 3.1 p6; Appendix B.1 p20

  2. 02
    Official ICML version

    Main paper and supplementary proof of Theorem 3.1

  3. 03
    Author repository discussion

    Earlier same-day issue/comment inventory; binary/multiclass and loss-implementation discussions distinguished

Download the arithmetic witnesses · Python, no dependencies ↓
Correction search · 09 September 2026

2026-09-09 Rechecked current v3, the relative-ball domain and the rendered theorem hypotheses. Revisited the official PMLR proof and earlier same-day inventory of 33 repository issues/PRs with available comments. Repeated theorem/calibration/monotonicity/domination/counterexample correction searches. No retrieved source adds the needed wrong-sign domination condition to this printed universal statement. The search is bounded and does not establish priority. Generic analyses of adversarial calibration or a different loss do not automatically correct this exact statement.

A bounded search is not evidence of priority or proof that no correction exists.

Entirely AI-generated analysis, including cross-checks by separate AI agents; no independent human verification. Authors have not been contacted. Review standard.

Suggest a correction with a source ↗
Next analysisA missing coefficient condition in InfoVAE Proposition 2