- Rating
- 3
- Confidence
- 2
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
This paper addresses a failure mode in scientific surrogates: small errors in a complete physical response can produce large errors in downstream metrics governed by cancellation-sensitive residuals. The authors propose residual-aware output representations that explicitly encode the relevant residual, reconstruct the complete response through an invertible transformation, and retain the original response-space training objective. The accompanying analysis characterizes the directional amplification of prediction errors near physical boundaries. For RF passives, the method exposes dissipation through impedance-based coordinates and reduces relative quality-factor errors by approximately 3.3–6.8× while maintaining comparable scattering-response accuracy. A nanophotonic experiment demonstrates an approximately twofold reduction in relative absorptance error. These results support output parameterization as a practical way to improve derived-metric accuracy without replacing complete-response prediction with task-specific regression.
Critical strengths
- Clear and practically relevant problem. The directional error analysis goes beyond observing a mismatch between response and task accuracy. The measured reduction in radial prediction error provides useful empirical support for the proposed mechanism.
- Simple, physically motivated formulation. The invertible representation preserves the complete response and its consistency with derived metrics, while avoiding additional task supervision in the main experiments.
- Meaningful downstream validation. Selected circuit designs are reevaluated using solver responses, and improvements remain after adding passivity filtering to the baseline. This strengthens the practical relevance of the prediction gains.
Critical weaknesses
The link between sensitivity analysis and learning remains incomplete. Equations (8) and (10) use different coordinate-error metrics. Their comparison does not establish reduced sensitivity under identical physical perturbations, nor explain why the unchanged response-space loss produces more favorable learned error directions. The main source of improvement deserves clearer emphasis. The direct (R,X) ablation nearly matches (log R,X), suggesting that explicit dissipation coordinates contribute more than the logarithmic parameterization. This distinction is important given the emphasis on logarithmic coordinates in the general formulation.
Questions for authors
How does the theoretical coordinate-error metric relate to the error distribution actually learned under the unchanged response-space objective? Does the strong performance of direct (R,X) prediction extend to the other RF templates and downstream circuit tasks? How does a direct-S baseline with comparable scalar contractivity constraints perform?
4 comments
3. Direct with a scalar contractivity constraint. We trained two constrained componentwise- baselines on the differential inductor. The training protocols and loss definition here match those of the componentwise- row of Table 1. The only change is a map applied to the predicted differential-mode reflection , the mode from which is extracted, before the complete response is reconstructed.
- (a) Radial projection: The network predicts as in the baseline, and a prediction outside the unit circle is moved back onto that circle along the radial direction, while a prediction inside it is left untouched. This is the baseline plus the constraint .
- (b) Smooth radial map: The network's raw output is not used as directly. It is first squashed: its angle is kept, and its magnitude is replaced by , which is below for any . Every prediction therefore lies inside the unit circle.
| MAE$(S)$ | rel. error | |
|---|---|---|
| componentwise | 0.0062 | 0.172 |
| + radial projection (a) | 0.0060 | 0.184 |
| + radius (b) | 0.0071 | 0.071 |
| 0.0065 | 0.053 |
The radial projection (a) does not change error, so we can conclude that enforcing contractivity is not the mechanism. The smooth radial map (b) is changing coordinates: near the boundary the network's radial output equals up to a constant, which is already a residual-exposing coordinate, and as a result it recovers most of the gain in accuracy. Both results support the paper's claim that the improvement comes from residual-aware output coordinates.
Thanks for your reply. Basically, I think the motivation is critical and practical. Many scenes, such as Battery Life Prediction, Machine Learning Interatomic Potential, face the problem demonstrated in this paper. The model's performance is always extremely high in metrics, while can only provide limited screening or ranking ability, since the residual error is inherently small. Despite this, since I am not familiar with RF passive components, I can't further increase my score to four. That's why I give a weak accept with confidence 2. Best luck to the authors, I hope the paper can be accepted. By the way, I would appreciate it if the author could extend this paper to other scientific applications.
Thank you for your time and for your supportive comments. We agree that the same structure appears well beyond RF, including in battery-life prediction and interatomic potentials: a model can be accurate on the complete response and still unreliable for the small residual quantity which determines the quality of designs. The atmospheric heating-rate case added in this rebuttal is one of our first steps in that direction.
Thank you for the careful reading and informative feedback, and we answer your three questions below, two of them with new experiments, and we will revise the manuscript accordingly.
1. Sensitivity analysis and error. Whichever representation does not change the physical sensitivity of Q to S; it changes the direction of the network's own error. Under an identical perturbation of S, ΔQ is the same for every representation, because Q is computed from S mathematically. Eqs. 8 and 10 compare a different quantity: the Q error produced by an error of a given size in the quantities the network outputs. When the network outputs S, its error is set against the small dissipative margin ϵ, and a radial error K gives a relative Q error of about 2K/ϵ (Eq. 7). When it outputs logR, an error of the same size is a relative error in R, and the relative Q error has that same size (Eq. 10).
To compare the two models (same backbone) with one ruler, we measure their learned errors in the physical S plane (Sec. 4.2, Fig. 2b). The mean radial error falls from 8.3×10−3 to 1.5×10−3, which is a 5.5× reduction, while the mean tangential error stays at the same scale (4.1×10−3 and 5.9×10−3). As a result, K/T moves from 2.08 to 0.26. The training loss (MAE of S) is the same in both cases.
2. Source of the improvement; (R,X) on the other templates and circuit tasks. Direct (R,X) is an instance of the proposed representation. R=Zdϵ/∣1−s∣2 vanishes with the dissipative margin, so R is the residual, and (R,X) gives it its own output coordinate with ρ(m)=m in place of ρ(m)=logm. Its performance therefore supports the main claim: the gain comes from exposing the residual as an output coordinate. We trained (R,X) on the other two templates and re-ran both circuit tasks (based on the protocol in Sec. 4.4, pooled over three seeds).
Both residual-exposing representations reduce Q error by 5–7× on both templates at similar MAE of (S).
The logarithm achieves clearly better result in circuit synthesis. It guarantees R>0 for any network output, leading to a passive reconstructed response (Proposition 1). In transformer matching-network selection, where the design is chosen by minimizing the predicted insertion loss, (logR,X) halves the median insertion-loss violation (0.11 against 0.23 dB), lowers its 95th percentile from 1.22 to 0.26 dB and that of the optimality gap from 0.92 to 0.20 dB, and every selected design is passive (while (R,X) coordinate has 1% non-passivity).