--Reinforcement learning (RL) policies deployed in safety-critical systems, such as unmanned aerial vehicle (UA V) navigation in dynamic airspace, are vulnerable to out-of-distribution (OOD) adversarial attacks in the observation space. These attacks induce distributional shifts that significantly degrade value estimation, leading to unsafe or suboptimal decision-making rendering the existing policy fragile. T o address this vulnerability, we propose an antifragile RL framework designed to adapt against curriculum of incremental adversarial perturbations. The framework introduces a simulated attacker which incrementally increases the strength of observation-space perturbations which enables the RL agent to adapt and generalize across a wider range of OOD observations and anticipate previously unseen attacks. We begin with a theoretical characterization of fragility, formally defining catastrophic forgetting as a monotonic divergence in value function distributions with increasing perturbation strength. Building on this, we define antifragility as the boundedness of such value shifts and derive adaptation conditions under which forgetting is stabilized. Our method enforces these bounds through iterative expert-guided critic alignment using Wasserstein distance minimization across incrementally perturbed observations. We empirically evaluate the approach in a UA V deconfliction scenario involving dynamic 3D obstacles. Results show that the antifragile policy consistently outperforms standard and robust RL baselines when subjected to both projected gradient descent (PGD) and GPS spoofing attacks, achieving up to 15% higher cumulative reward and over 30% fewer conflict events. These findings demonstrate the practical and theoretical viability of antifragile reinforcement learning for secure and resilient decision-making in environments with evolving threat scenarios. Fragility, robustness, and antifragility represent a continuum of system responses to stress or external perturbations [1, 2].

Recently, multi-objective optimization (MOO) has gained attention for its broad applications in ML, operations research, and engineering. However, MOO algorithm design remains in its infancy and many existing MOO methods suffer from unsatisfactory convergence rate and sample complexity performance. To address this challenge, in this paper, we propose an algorithm called STIMULUS( stochastic path-integrated multi-gradient recursive e\ulstimator), a new and robust approach for solving MOO problems. Different from the traditional methods, STIMULUS introduces a simple yet powerful recursive framework for updating stochastic gradient estimates to improve convergence performance with low sample complexity. In addition, we introduce an enhanced version of STIMULUS, termed STIMULUS-M, which incorporates a momentum term to further expedite convergence. We establish $O(1/T)$ convergence rates of the proposed methods for non-convex settings and $O (\exp{-μT})$ for strongly convex settings, where $T$ is the total number of iteration rounds. Additionally, we achieve the state-of-the-art $O \left(n+\sqrt{n}ε^{-1}\right)$ sample complexities for non-convex settings and $O\left(n+ \sqrt{n} \ln ({μ/ε})\right)$ for strongly convex settings, where $ε>0$ is a desired stationarity error. Moreover, to alleviate the periodic full gradient evaluation requirement in STIMULUS and STIMULUS-M, we further propose enhanced versions with adaptive batching called STIMULUS+/ STIMULUS-M+ and provide their theoretical analysis.

Token embeddings play a crucial role in language modeling but, despite this practical relevance, their theoretical understanding remains limited. Our paper addresses the gap by characterizing the structure of embeddings obtained via gradient descent. Specifically, we consider a one-layer softmax attention model with a linear head for binary classification, i.e., $\texttt{Softmax}( p^\top E_X^\top ) E_X v = \frac{ \sum_{i=1}^T \exp(p^\top E_{x_i}) E_{x_i}^\top v}{\sum_{j=1}^T \exp(p^\top E_{x_{j}}) }$, where $E_X = [ E_{x_1} , \dots, E_{x_T} ]^\top$ contains the embeddings of the input sequence, $p$ is the embedding of the $\mathrm{\langle cls \rangle}$ token and $v$ the output vector. First, we show that, already after a single step of gradient training with the logistic loss, the embeddings $E_X$ capture the importance of tokens in the dataset by aligning with the output vector $v$ proportionally to the frequency with which the corresponding tokens appear in the dataset. Then, after training $p$ via gradient flow until convergence, the softmax selects the important tokens in the sentence (i.e., those that are predictive of the label), and the resulting $\mathrm{\langle cls \rangle}$ embedding maximizes the margin for such a selection. Experiments on real-world datasets (IMDB, Yelp) exhibit a phenomenology close to that unveiled by our theory.

Stochastic Gradient Descent (SGD) and its variants, such as ADAM, are foundational to deep learning optimization, adjusting model parameters through fixed or adaptive learning rates based on loss function gradients. However, these methods often struggle to balance adaptability and efficiency in high-dimensional, non-convex settings. This paper introduces AYLA, a novel optimization framework that enhances training dynamics via loss function transformation. AYLA applies a tunable power-law transformation to the loss, preserving critical points while scaling loss values to amplify gradient sensitivity and accelerate convergence. Additionally, we propose an effective learning rate that dynamically adapts to the transformed loss, further improving optimization efficiency. Empirical evaluations on minimizing a synthetic non-convex polynomial, solving a non-convex curve-fitting task, and performing digit classification (MNIST) and image recognition (CIFAR-100) demonstrate that AYLA consistently outperforms SGD and ADAM in both convergence speed and training stability. By reshaping the loss landscape, AYLA provides a model-agnostic enhancement to existing optimization methods, offering a promising advancement in deep neural network training.

Companion paper [118] developed a powerful \emph{Random duality theory} (RDT) based analytical program to statistically characterize performance of \emph{descending} phase retrieval algorithms (dPR) (these include all variants of gradient descents and among them widely popular Wirtinger flows). We here generalize the program and show how it can be utilized to handle rank $d$ positive definite phase retrieval (PR) measurements (with special cases $d=1$ and $d=2$ serving as emulations of the real and complex phase retrievals, respectively). In particular, we observe that the minimal sample complexity ratio (number of measurements scaled by the dimension of the unknown signal) which ensures dPR's success exhibits a phase transition (PT) phenomenon. For both plain and lifted RDT we determine phase transitions locations. To complement theoretical results we implement a log barrier gradient descent variant and observe that, even in small dimensional scenarios (with problem sizes on the order of 100), the simulated phase transitions are in an excellent agreement with the theoretical predictions.

Stochastic Gradient Descent (SGD) is fundamental for training deep neural networks, especially in non-convex settings. Understanding SGD's generalization properties is crucial for ensuring robust model performance on unseen data. In this paper, we analyze the generalization error bounds of SGD for non-convex learning by introducing the Type II perturbed SGD (T2pm-SGD), which accommodates both sub-Gaussian and bounded loss functions. The generalization error bound is decomposed into two components: the trajectory term and the flatness term. Our analysis improves the trajectory term to $O(n^{-1})$, significantly enhancing the previous $O((nb)^{-1/2})$ bound for bounded losses, where n is the number of training samples and b is the batch size. By selecting an optimal variance for the perturbation noise, the overall bound is further refined to $O(n^{-2/3})$. For sub-Gaussian loss functions, a tighter trajectory term is also achieved. In both cases, the flatness term remains stable across iterations and is smaller than those reported in previous literature, which increase with iterations. This stability, ensured by T2pm-SGD, leads to tighter generalization error bounds for both loss function types. Our theoretical results are validated through extensive experiments on benchmark datasets, including MNIST and CIFAR-10, demonstrating the effectiveness of T2pm-SGD in establishing tighter generalization bounds.

Lipschitz continuity characterizes the worst-case sensitivity of neural networks to small input perturbations; yet its dynamics (i.e. temporal evolution) during training remains under-explored. We present a rigorous mathematical framework to model the temporal evolution of Lipschitz continuity during training with stochastic gradient descent (SGD). This framework leverages a system of stochastic differential equations (SDEs) to capture both deterministic and stochastic forces. Our theoretical analysis identifies three principal factors driving the evolution: (i) the projection of gradient flows, induced by the optimization dynamics, onto the operator-norm Jacobian of parameter matrices; (ii) the projection of gradient noise, arising from the randomness in mini-batch sampling, onto the operator-norm Jacobian; and (iii) the projection of the gradient noise onto the operator-norm Hessian of parameter matrices. Furthermore, our theoretical framework sheds light on such as how noisy supervision, parameter initialization, batch size, and mini-batch sampling trajectories, among other factors, shape the evolution of the Lipschitz continuity of neural networks. Our experimental results demonstrate strong agreement between the theoretical implications and the observed behaviors.

Federated learning (FL) has emerged as a groundbreaking paradigm in machine learning (ML), offering privacy-preserving collaborative model training across diverse datasets. Despite its promise, FL faces significant hurdles in non-identically and independently distributed (non-IID) data scenarios, where most existing methods often struggle with data heterogeneity and lack robustness in performance. This paper introduces Federated Loss Exploration (FedLEx), an innovative approach specifically designed to tackle these challenges. FedLEx distinctively addresses the shortcomings of existing FL methods in non-IID settings by optimizing its learning behavior for scenarios in which assumptions about data heterogeneity are impractical or unknown. It employs a federated loss exploration technique, where clients contribute to a global guidance matrix by calculating gradient deviations for model parameters. This matrix serves as a strategic compass to guide clients' gradient updates in subsequent FL rounds, thereby fostering optimal parameter updates for the global model. FedLEx effectively navigates the complex loss surfaces inherent in non-IID data, enhancing knowledge transfer in an efficient manner, since only a small number of epochs and small amount of data are required to build a strong global guidance matrix that can achieve model convergence without the need for additional data sharing or data distribution statics in a large client scenario. Our extensive experiments with state-of-the art FL algorithms demonstrate significant improvements in performance, particularly under realistic non-IID conditions, thus highlighting FedLEx's potential to overcome critical barriers in diverse FL applications.

Low - rank representat i on learn ing ha s emerged as a powerful tool for recoverin g missing values i n power load data due to its ability to exploit the inherent low - dimensional structures of spatiotemporal measurements. Among various techniques, low - rank factorization models are f a vou red f o r t he ir efficiency and interpretability . Howeve r, their performan ce is highly sensitive to the choice of regularization parameter s, which are typically fixed or manually tuned, resulting in limited generalization capability or slow convergenc e in pra ctica l sc en arios. In this paper, we propo se a Regular ization - optimized Low - Rank Factorization, which introduces a Proportional - Integral - Derivative controller to adaptively adju st the regularization coefficient . Furthe rmore, we provide a detailed algori t hmi c com plex i t y analysis, showing that our method preser ves the computatio nal efficiency of stochastic gradient descent while improving ad aptivity. Experimental results on real - world power load datasets validate the superiority of our method in both imput a tio n acc urac y and training efficiency compared to existi ng baselines.

Differentially private (DP) image synthesis aims to generate synthetic images from a sensitive dataset, alleviating the privacy leakage concerns of organizations sharing and utilizing synthetic images. Although previous methods have significantly progressed, especially in training diffusion models on sensitive images with DP Stochastic Gradient Descent (DP-SGD), they still suffer from unsatisfactory performance. In this work, inspired by curriculum learning, we propose a two-stage DP image synthesis framework, where diffusion models learn to generate DP synthetic images from easy to hard. Unlike existing methods that directly use DP-SGD to train diffusion models, we propose an easy stage in the beginning, where diffusion models learn simple features of the sensitive images. To facilitate this easy stage, we propose to use `central images', simply aggregations of random samples of the sensitive dataset. Intuitively, although those central images do not show details, they demonstrate useful characteristics of all images and only incur minimal privacy costs, thus helping early-phase model training. We conduct experiments to present that on the average of four investigated image datasets, the fidelity and utility metrics of our synthetic images are 33.1% and 2.1% better than the state-of-the-art method.