Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2609.39300v1 Announce Type: new Abstract: Recent advances in computer vision have made broadcast sports videos increasingly useful for event analysis, performance assessment, and player-safety...
arXiv:2609.39145v1 Announce Type: cross Abstract: Unreliable visual inputs can harm task performance and cause potential physical safety risks for vision-language-action (VLA) models. We analyze how...
arXiv:2609.39429v1 Announce Type: cross Abstract: Uncertainty Quantification (UQ) is a key requirement for trustworthy AI in high-stakes medical image analysis. In this work, we evaluate UQ in a mult...
arXiv:2609.39599v1 Announce Type: cross Abstract: 3D visuomotor policies provide a strong foundation for spatially precise manipulation, yet current text-to-3D policies struggle to follow unseen fine...
arXiv:2609.32923v2 Announce Type: replace Abstract: Wireless digital twins (DTs) rely on 3D environment models to predict radio propagation and support wireless-network decisions, yet these models ar...
arXiv:2605.30454v2 Announce Type: replace-cross Abstract: Prompt-injection benchmarks for LLM agents typically test attacks through a single injection surface and report the resulting attack success...
arXiv:2609.39132v1 Announce Type: new Abstract: We study few-step video generation, i.e., distilling a multi-step video generator, which typically requires tens of sampling steps, incurring substanti...
arXiv:2606.27223v3 Announce Type: replace Abstract: Gaussian Splatting has been recently explored for satellite 3D reconstruction, demonstrating flexibility and efficiency in representing radiometric...
arXiv:2609.39339v1 Announce Type: cross Abstract: Matching problems are ubiquitous in data science as they enable the alignment of structured objects and distributions. While existing solvers are oft...
arXiv:2507.23559v2 Announce Type: replace-cross Abstract: Certain data are naturally modeled by networks or weighted graphs, be they biological networks or mobility networks. When there is no canonic...
The paper develops a statistical theory for minimum‑norm interpolation in high‑dimensional regression, showing how regularization geometry and signal sparsity affect generalization. It identifies regimes where sparsity‑promoting regularizers yield exact interpolation that is far more accurate than approximate fitting, and proves a zero–one generalization law for strongly overparameterized noiseless problems. The authors also characterize training and generalization errors along ρ‑regularization paths when feature dimension and sample size are proportional, demonstrating that generalization improves with more sparsity‑promoting norms and sparser targets, and that small changes in regularization strength can cause large shifts in generalization. whyItMatters":"The work provides a quantitative understanding of delayed generalization (grokking) and reveals a statistical instability in minimum‑norm interpolation, offering insights that could guide the design of regularizers for better generalization in overparameterized models."
arXiv:2609.39640v1 Announce Type: cross Abstract: Cross-lingual transfer describes how knowledge in a source language benefits a target language. Measuring it quantitatively requires broad multilingu...
arXiv:2609.38547v1 Announce Type: cross Abstract: Defining a weighted mean over probability measures under probability metrics is a central tool in probabilistic machine learning. Under the Wasserste...
arXiv:2609.38768v1 Announce Type: cross Abstract: Prior work has shown that neural networks exhibit implicit biases toward low-complexity structure (e.g., spectral bias), memorization dynamics, and c...
arXiv:2609.38860v1 Announce Type: cross Abstract: Learning from human preferences is central to large language model (LLM) alignment, but human preference annotation is costly. Active preference lear...
The paper introduces Manifold Anchored Bilevel Transfer (MABT), a framework that aligns adversarial attack trajectories with the intrinsic data manifold to reduce surrogate-specific overfitting. MABT employs a relaxed manifold-anchoring operator as a semantic rectifier and formulates transfer attack generation as a distributional bilevel optimization problem, learning geometry-aligned initializations that minimize expected transfer risk. A Hessian-free solver with linear-time complexity is developed to efficiently solve the resulting hierarchy, and experiments show improved transferability across 10 attackers, 28 configurations, various victim models, and defense mechanisms.
arXiv:2609.39025v1 Announce Type: cross Abstract: Fairness assessment in algorithmic decisions that affect individuals, such as credit scoring, often relies on parity measures calculated at the aggre...
arXiv:2609.39074v1 Announce Type: cross Abstract: Federated learning (FL) on memory-constrained edge devices faces a dilemma: first-order (FO) optimization (i.e., backpropagation) demands substantial...
arXiv:2609.39408v1 Announce Type: cross Abstract: Population loss can remain nearly constant while a neural network learns a substantially more predictive representation. We establish this separation...
The paper introduces probabilistic adversarial training, a method that enhances robustness by reducing the overlap between a distance-based distribution and a victim-classifier-induced distribution. It derives a KL-based lower bound on probabilistic robustness, which serves as a tractable surrogate objective. Experiments confirm that this approach consistently improves probabilistic robustness and can even boost non-probabilistic adversarial training methods through an induced scaling factor.