Securing 3D Deep Learning Models: Simple and Effective Defense Against Adversarial Attacks
Bibliographic record
Abstract
In recent years, the growing vulnerability of deep neural networks (DNNs) to adversarial attacks has posed sig-nificant challenges in the field of machine learning, particularly in mission-critical applications such as computer vision. As a result, adversarial machine learning has emerged as a crucial research area focused on fortifying deep neural networks against these sophisticated threats. Simultaneously, the use of 3D datasets representing 3D objects has become increasingly important in applications like autonomous driving, robotics, and augmented reality. As the adoption of deep learning networks for processing and classifying both 3D and 2D data grows, concerns over their vulnerability to adversarial attacks have escalated. Despite extensive research on adversarial defense in many areas, adversarial 3D deep learning models remains compara-tively unexplored. This study examines the susceptibility of 3D deep learning to various forms of adversarial attacks, revealing significant weaknesses in current approaches. Our findings demonstrate these attacks, which involve subtle modifications to the data, can significantly degrade the performance of classifiers, leading to misclassification and potentially dangerous outcomes in real-world scenarios. We illustrate that adversarial inputs can compromise prediction performance, with accuracy dropping by over 20%. In response, we propose an efficient defense strategy that does not overburden the learning model with a heavy adversarial training step. Our simplified defense strategy employs only the best classified 3D objects per class and not only restores network classification accuracy to baseline performance but can improve accuracy to above baseline performance in the event of adversarial attacks.
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 | 0.002 | 0.007 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.004 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
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".