A recent line of work focused on making adversarial training computationally efficient for deep learning models. In particular, Wong et al. (2020) showed that $\ell_\infty$-adversarial training with fast gradient sign method (FGSM) can fail due to a phenomenon called catastrophic overfitting, when the model quickly loses its robustness over a single epoch of training. We show that adding a random step to FGSM, as proposed in Wong et al. (2020), does not prevent catastrophic overfitting, and that randomness is not important per se --- its main role being simply to reduce the magnitude of the perturbation. Moreover, we show that catastrophic overfitting is not inherent to deep and overparametrized networks, but can occur in a single-layer convolutional network with a few filters. In an extreme case, even a single filter can make the network highly non-linear locally, which is the main reason why FGSM training fails. Based on this observation, we propose a new regularization method, GradAlign, that prevents catastrophic overfitting by explicitly maximizing the gradient alignment inside the perturbation set and improves the quality of the FGSM solution. As a result, GradAlign allows to successfully apply FGSM training also for larger $\ell_\infty$-perturbations and reduce the gap to multi-step adversarial training.

Study of the nonlinear evolution deep neural network (DNN) parameters undergo during training has uncovered regimes of distinct dynamical behavior. While a detailed understanding of these phenomena has the potential to advance improvements in training efficiency and robustness, the lack of methods for identifying when DNN models have equivalent dynamics limits the insight that can be gained from prior work. Topological conjugacy, a notion from dynamical systems theory, provides a precise definition of dynamical equivalence, offering a possible route to address this need. However, topological conjugacies have historically been challenging to compute. By leveraging advances in Koopman operator theory, we develop a framework for identifying conjugate and non-conjugate training dynamics. To validate our approach, we demonstrate that comparing Koopman eigenvalues can correctly identify a known equivalence between online mirror descent and online gradient descent. We then utilize our approach to: (a) identify non-conjugate training dynamics between shallow and wide fully connected neural networks; (b) characterize the early phase of training dynamics in convolutional neural networks; (c) uncover non-conjugate training dynamics in Transformers that do and do not undergo grokking. Our results, across a range of DNN architectures, illustrate the flexibility of our framework and highlight its potential for shedding new light on training dynamics.

He has said privately the deal was'non-binding'. He has said privately the deal was'non-binding'. What does the disappearance of a $100bn deal mean for the AI economy? Apparent collapse of Nvidia-OpenAI tie-up raises questions about circular funding and who will bear the cost of AI's expansion Did the circular AI economy just wobble? Last week it was reported that a much-discussed $100bn deal - announced last September - between Nvidia and OpenAI might not be happening at all.

Recent works (e.g., (Li \& Arora, 2020)) suggest that the use of popular normalization schemes (including Batch Normalization) in today's deep learning can move it far from a traditional optimization viewpoint, e.g., use of exponentially increasing learning rates. The current paper highlights other ways in which behavior of normalized nets departs from traditional viewpoints, and then initiates a formal framework for studying their mathematics via suitable adaptation of the conventional framework namely, modeling SGD-induced training trajectory via a suitable stochastic differential equation (SDE) with a noise term that captures gradient noise. This yields: (a) A new \textquotedblleft intrinsic learning rate\textquotedblright\ parameter that is the product of the normal learning rate $\eta$ and weight decay factor $\lambda$. Analysis of the SDE shows how the effective speed of learning varies and equilibrates over time under the control of intrinsic LR.

Alphabet reports $34.5bn profit and revenue soars 48% in recent quarter as it plans a sharp increase in AI spending Google's parent company, Alphabet, beat Wall Street expectations on Wednesday, and is planning a sharp increase in capital spending in 2026 as it continues to invest deeply in AI infrastructure. Alphabet on Wednesday reported profit of $34.5bn in the recently ended quarter, as revenue from cloud computing soared 48%. In an earnings call, investors pressed Alphabet's chief executive, Sundar Pichai, on the significant increase. "We've been supply constrained, even as we've been ramping up our capacity. Obviously, our CapEx spend this year is an eye towards the future," Pichai said, in response.

Training modern large language models (LLMs) has become a veritable smorgasbord of algorithms and datasets designed to elicit particular behaviors, making it critical to develop techniques to understand the effects of datasets on the model's properties. This is exacerbated by recent experiments that show datasets can transmit signals that are not directly observable from individual datapoints, posing a conceptual challenge for dataset-centric understandings of LLM training and suggesting a missing fundamental account of such phenomena. Towards understanding such effects, inspired by recent work on the linear structure of LLMs, we uncover a general mechanism through which hidden subtexts can arise in generic datasets. We introduce Logit-Linear-Selection (LLS), a method that prescribes how to select subsets of a generic preference dataset to elicit a wide range of hidden effects. We apply LLS to discover subsets of real-world datasets so that models trained on them exhibit behaviors ranging from having specific preferences, to responding to prompts in a different language not present in the dataset, to taking on a different persona. Crucially, the effect persists for the selected subset, across models with varying architectures, supporting its generality and universality.

Deep learning has achieved remarkable success across a wide range of domains, significantly expanding the frontiers of what is achievable in artificial intelligence. Yet, despite these advances, critical challenges remain -- most notably, ensuring robustness to small input perturbations and generalization to out-of-distribution data. These critical challenges underscore the need to understand the underlying fundamental principles that govern robustness and generalization. Among the theoretical tools available, Lipschitz continuity plays a pivotal role in governing the fundamental properties of neural networks related to robustness and generalization. It quantifies the worst-case sensitivity of network's outputs to small input perturbations. While its importance is widely acknowledged, prior research has predominantly focused on empirical regularization approaches based on Lipschitz constraints, leaving the underlying principles less explored. This thesis seeks to advance a principled understanding of the principles of Lipschitz continuity in neural networks within the paradigm of machine learning, examined from two complementary perspectives: an internal perspective -- focusing on the temporal evolution of Lipschitz continuity in neural networks during training (i.e., training dynamics); and an external perspective -- investigating how Lipschitz continuity modulates the behavior of neural networks with respect to features in the input data, particularly its role in governing frequency signal propagation (i.e., modulation of frequency signal propagation).

Setting the learning rate for a deep learning model is a critical part of successful training, yet choosing this hyperparameter is often done empirically with trial and error. In this work, we explore a solvable model of optimal learning rate schedules for a powerlaw random feature model trained with stochastic gradient descent (SGD). We consider the optimal schedule $η_T^\star(t)$ where $t$ is the current iterate and $T$ is the total training horizon. This schedule is computed both numerically and analytically (when possible) using optimal control methods. Our analysis reveals two regimes which we term the easy phase and hard phase. In the easy phase the optimal schedule is a polynomial decay $η_T^\star(t) \simeq T^{-ξ} (1-t/T)^δ$ where $ξ$ and $δ$ depend on the properties of the features and task. In the hard phase, the optimal schedule resembles warmup-stable-decay with constant (in $T$) initial learning rate and annealing performed over a vanishing (in $T$) fraction of training steps. We investigate joint optimization of learning rate and batch size, identifying a degenerate optimality condition. Our model also predicts the compute-optimal scaling laws (where model size and training steps are chosen optimally) in both easy and hard regimes. Going beyond SGD, we consider optimal schedules for the momentum $β(t)$, where speedups in the hard phase are possible. We compare our optimal schedule to various benchmarks in our task including (1) optimal constant learning rates $η_T(t) \sim T^{-ξ}$ (2) optimal power laws $η_T(t) \sim T^{-ξ} t^{-χ}$, finding that our schedule achieves better rates than either of these. Our theory suggests that learning rate transfer across training horizon depends on the structure of the model and task. We explore these ideas in simple experimental pretraining setups.

BA YESIAN PINNS FOR UNCERT AINTY-A W ARE INVERSE PROBLEMS (BPINN-IP) Ali MOHAMMAD-DJAF ARI ISCT, Bures-sur-Y vette, France Institute of Digital T win (IDT), EIT, Ningbo, China Dept. of Statistics, Central South University, Changcha, China ABSTRACT The main contribution of this paper is to develop a hierarchical Bayesian formulation of PINNs for linear inverse problems, which is called BPINN-IP . The proposed methodology extends PINN to account for prior knowledge on the nature of the expected NN output, as well as its weights. Also, as we can have access to the posterior probability distributions, naturally uncertainties can be quantified. Also, variational inference and Monte Carlo dropout are employed to provide predictive means and variances for reconstructed images. Un example of applications to deconvolution and super-resolution is considered, details of the different steps of implementations are given, and some preliminary results are presented.

Information Bottleneck (IB) is widely used, but in deep learning, it is usually implemented through tractable surrogates, such as variational bounds or neural mutual information (MI) estimators, rather than directly controlling the MI I(X;Z) itself. The looseness and estimator-dependent bias can make IB "compression" only indirectly controlled and optimization fragile. We revisit the IB problem through the lens of information geometry and propose a \textbf{Geo}metric \textbf{I}nformation \textbf{B}ottleneck (\textbf{GeoIB}) that dispenses with mutual information (MI) estimation. We show that I(X;Z) and I(Z;Y) admit exact projection forms as minimal Kullback-Leibler (KL) distances from the joint distributions to their respective independence manifolds. Guided by this view, GeoIB controls information compression with two complementary terms: (i) a distribution-level Fisher-Rao (FR) discrepancy, which matches KL to second order and is reparameterization-invariant; and (ii) a geometry-level Jacobian-Frobenius (JF) term that provides a local capacity-type upper bound on I(Z;X) by penalizing pullback volume expansion of the encoder. We further derive a natural-gradient optimizer consistent with the FR metric and prove that the standard additive natural-gradient step is first-order equivalent to the geodesic update. We conducted extensive experiments and observed that the GeoIB achieves a better trade-off between prediction accuracy and compression ratio in the information plane than the mainstream IB baselines on popular datasets. GeoIB improves invariance and optimization stability by unifying distributional and geometric regularization under a single bottleneck multiplier. The source code of GeoIB is released at "https://anonymous.4open.science/r/G-IB-0569".