Investigating the Reliability of Machine Learning Algorithms for Inverse Design: Polarization Rotator Unit-Cell Model
Bibliographic record
Abstract
In this article, we implement three types of data-driven algorithms for the inverse design of the polarization rotator (PR) unit cell. Toward this end, a novel configuration pattern of the PR unit cell is proposed, which is reshapable to different geometries with a large number of design variables and recasts as a regression problem. The state-of-the-art algorithms including the neural network (NN), the deep neural networks (DNNs) with multiple hidden layers, and the support vector regression (SVR) are experimented with cross validations for ensuring the prediction generalizability. Averaged over all the experiments with competitive performances, the highest prediction accuracy about 95.23% was achieved for the SVR algorithm. This demonstrates the enormous capability of the data-driven algorithms in the geometrical dimension prediction of the unit cells for any given frequency band designated in the radar range (X, Ku, K, and Ka). The proposed inverse design procedure can expedite, facilitate, and to some extent replace the conventional and time-consuming manual design approaches with electromagnetic (EM) simulation software. Although these models can be very efficient in practice, they might be vulnerable against adversarial attacks that craft fake inputs to purposely fool the victim regressors toward adversary's wishes. This poses security concerns for the learning-based algorithms and might negatively affect their prediction reliabilities in runtime. In this article, we only characterize the existence of the adversarial attacks for the regression models using the fast gradient sign method. Our conducted experiments uncover that the fooling rate of all the aforementioned cutting-edge NN and DNN-based regressors is above 98% and this rate for the SVR model is about 11% better than other models.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.017 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".