This paper studies large-width asymptotics for ReLU neural networks with α-Stable initializations.
problem Characterizing the large-width behavior of ReLU neural networks with α-Stable initializations.
method Analysis of the large-width distributions and training dynamics of ReLU neural networks initialized with α-Stable distributions.
result For ReLU neural networks with α-Stable initializations, the large-width training dynamics achieve zero training error at a linear rate, characterized by a random kernel.
Model selection on validation data is an essential step in machine learning. While the mixing of data between training and validation is considered taboo, practitioners often violate it to increase performance. Here, we offer a simple, practical method for using the validation set for training, which allows for a conti…
Paper proposes HTAF for stable training of binary neural networks.
problem Challenges in training binary neural networks with gradient-based optimization.
method HTAF is a smooth approximation to the Heaviside function that enables stable training.
result HTAF enables stable training of various binary neural networks with gradient-based optimization.
Stable GFlowNets prevent loss spikes and mode collapse in training.
problem Unstable training of GFlowNets leading to loss spikes and mode collapse.
method Assessed sensitivity of GFlowNet objectives, derived loss-to-TV bounds, and proposed Stable GFlowNets.
result Stable GFlowNets improve training behavior and distributional fidelity.
Stability of recurrent models is closely linked with trainability, generalizability and in some applications, safety. Methods that train stable recurrent neural networks, however, do so at a significant cost to expressibility. We propose an implicit model structure that allows for a convex parametrization of stable mod…
Log-Normal Multiplicative Dynamics improves low-precision training of neural networks.
problem Training large neural networks with low precision is unstable.
method Derive a Bayesian learning rule with log-normal posterior distributions and multiplicative updates.
result LMD achieves stable and accurate training for Vision Transformer and GPT-2.
Stable health predictions need deconfounding test set features.
problem Stability of predictions in health machine learning is compromised by selection biases.
method Deconfounding the test set features improves prediction stability across different environments.
result Improved stability achieved by deconfounding test set features.
IMM generates high-quality samples in few steps with stable training.
problem Slow inference and instability in generating high-quality samples using diffusion models and Flow Matching.
method Inductive Moment Matching (IMM) is a new generative model for one- or few-step sampling with a single-stage training procedure.
result IMM achieves state-of-the-art 2-step FID of 1.98 on CIFAR-10 for a model trained from scratch.
Stable training of deep normalizing flows for high-dimensional variational inference.
problem Training deep normalizing flows for high-dimensional posterior distributions is infeasible due to high stochastic gradient variance.
method Proposed a combination of soft-thresholding of scale and bijective soft log transformation to stabilize training.
result Stable training of Real NVPs for posterior distributions with thousands of dimensions is possible.
In many important machine learning applications, the training distribution used to learn a probabilistic classifier differs from the testing distribution on which the classifier will be used to make predictions. Traditional methods correct the distribution shift by reweighting the training data with the ratio of the de…
Stabilizes complex systems using diffusion models trained on Lyapunov functions.
problem Generating stabilizing controllers for complex dynamical systems.
method Trains a diffusion model on pairs of asymptotically stable vector fields and their Lyapunov functions to identify the closest stable field and adjust control functions.
result Efficient and rapid stabilization of unseen systems, showcasing generalizability.
ADMMiRNN solves RNN training issues with stable convergence.
problem Training RNN with stable convergence and avoiding gradient issues.
method Built ADMMiRNN framework on unfolded RNN, providing novel update rules and theoretical analysis.
result ADMMiRNN achieves convergent results and outperforms baselines.
The paper proposes a method to stabilize predictions by identifying causal variables using a seed variable.
problem Stable prediction across unknown test data with potential spurious correlations.
method Conditional independence test based algorithm using a seed variable to separate causal from non-causal variables.
result The algorithm precisely separates causal and non-causal variables for stable prediction across test data.
Gradient descent biases towards stable rank networks for nearly-orthogonal data.
problem Understanding implicit bias in non-smooth neural networks trained by gradient descent.
method Analysis of two-layer ReLU and leaky ReLU networks trained by gradient descent on nearly-orthogonal data.
result Gradient descent biases towards networks with stable rank and uniform margin for nearly-orthogonal data.
Spectral gradient methods outperform Euclidean in certain deep learning scenarios.
problem When do spectral gradient updates outperform Euclidean in deep learning?
method Layerwise condition comparing squared nuclear-to-Frobenius ratio to stable rank of activations.
result Spectral updates can be more effective than Euclidean in deep networks and transformers.
Generative adversarial networks (GANs) have received a tremendous amount of attention in the past few years, and have inspired applications addressing a wide range of problems. Despite its great potential, GANs are difficult to train. Recently, a series of papers (Arjovsky & Bottou, 2017a; Arjovsky et al. 2017b; and Gu…
Collaborative filtering (CF) is a popular technique in today's recommender systems, and matrix approximation-based CF methods have achieved great success in both rating prediction and top-N recommendation tasks. However, real-world user-item rating matrices are typically sparse, incomplete and noisy, which introduce ch…
Any endomorphism of a finitely generated free group naturally descends to an injective endomorphism of its stable quotient. In this paper, we prove a geometric incarnation of this phenomenon: namely, that every expanding irreducible train track map inducing an endomorphism of the fundamental group gives rise to an expa…
Stability is a fundamental property of dynamical systems, yet to this date it has had little bearing on the practice of recurrent neural networks. In this work, we conduct a thorough investigation of stable recurrent models. Theoretically, we prove stable recurrent neural networks are well approximated by feed-forward …
Batch Normalization (BatchNorm) is a widely adopted technique that enables faster and more stable training of deep neural networks (DNNs). Despite its pervasiveness, the exact reasons for BatchNorm's effectiveness are still poorly understood. The popular belief is that this effectiveness stems from controlling the chan…
Retraining stabilizes model influence on data.
problem Performativity in predictive models leads to feedback loops.
method Developed the stable signal principle to address retraining dynamics.
result Repeated risk minimization converges geometrically to stable signal direction.
New ODE solvers improve training efficiency and accuracy.
problem Training Neural ODEs requires efficient and accurate gradient calculation.
method Presented algebraically reversible ODE solvers that are time and memory efficient, calculate exact gradients, and are numerically stable.
result Reversible solvers strictly improve upon previous architectures in efficiency and accuracy.
FGSM is more stable in adversarially robust transfer learning than PGD.
problem Computational efficiency in adversarially robust transfer learning.
method Revisited use of FGSM in adversarial fine-tuning.
result FGSM is more stable and efficient in adversarial fine-tuning.
ESPO optimizes LLMs for complex tasks by balancing fine-grained updates and stability.
problem Gradient underutilization in sequence-level optimization.
method Entropy Grouping Importance Sampling and Entropy Adaptive Clipping.
result ESPO accelerates convergence and achieves state-of-the-art performance.
Generative model solves financial market equilibria with stable reinforcement learning.
problem Financial market equilibria under realistic frictions and multiple agents.
method Generative adversarial reinforcement learning with decoupling feedback.
result Algorithm learns and predicts asset returns and volatilities.
Recurrent neural networks trained on regular languages exhibit stable states that can recover from noise.
problem Stability of internal states in recurrent neural networks trained on regular languages.
method Empirical study with analysis of network activation and transitions between states.
result Recurrent neural networks trained on regular languages can recover from random perturbations and maintain stable states.
Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often du…
Generative Adversarial Networks (GANs) are powerful models for learning complex distributions. Stable training of GANs has been addressed in many recent works which explore different metrics between distributions. In this paper we introduce Fisher GAN which fits within the Integral Probability Metrics (IPM) framework f…
We present an efficient technique, which allows to train classification networks which are verifiably robust against norm-bounded adversarial attacks. This framework is built upon the work of Gowal et al., who applies the interval arithmetic to bound the activations at each layer and keeps the prediction invariant to t…
Enhances OOD detection using latent diffusion for more robust and efficient training.
problem Improving reliability of machine learning models in real-world scenarios.
method Proposes Outlier-Aware Learning (OAL) framework that generates synthetic OOD data in latent space and uses MICL and KD modules.
result Demonstrates superior performance on benchmark datasets.
Default-ERM shortcut learning persists even without additional information.
problem Default-ERM shortcut learning in perception tasks despite stable feature sufficiency.
method Studied linear perception task; developed margin control (MARG-CTRL) loss functions.
result Margin control mitigates shortcut learning on various tasks.
We study whether a neural network optimizes to the same, linearly connected minimum under different samples of SGD noise (e.g., random data order and augmentation). We find that standard vision models become stable to SGD noise in this way early in training. From then on, the outcome of optimization is determined to a …
Generative adversarial networks have been very successful in generative modeling, however they remain relatively challenging to train compared to standard deep neural networks. In this paper, we propose new visualization techniques for the optimization landscapes of GANs that enable us to study the game vector field re…
We observe that several existing policy gradient methods (such as vanilla policy gradient, PPO, A2C) may suffer from overly large gradients when the current policy is close to deterministic (even in some very simple environments), leading to an unstable training process. To address this issue, we propose a new method, …
Proposes BSSP to stabilize predictions in biased data.
problem Distribution shift between training and test data causes prediction instability.
method Balance-subsampled stable prediction (BSSP) algorithm based on fractional factorial design.
result Significantly improves prediction stability across unknown test data.
Unsupervised Domain Adaptation (DA) is used to automatize the task of labeling data: an unlabeled dataset (target) is annotated using a labeled dataset (source) from a related domain. We cast domain adaptation as the problem of finding stable labels for target examples. A new definition of label stability is proposed, …
New text-to-image diffusion models improve scene understanding for AI agents.
problem Fine-grained scene understanding for AI agents from text and images.
method Pre-trained text-to-image diffusion models optimized for generating images from text prompts.
result Policies learned with Stable Control Representations outperform state-of-the-art approaches on various control tasks.
Stable long-term predictions for fluid flows using neural networks.
problem Predicting complex dynamics of fluid flows with high temporal stability.
method End-to-end trained neural network architecture combining CNN for spatial compression and LSTM for temporal prediction.
result Novel latent space subdivision (LSS) allows stable and controllable long-term predictions.
This paper analyzes deep Stable neural networks, showing convergence rates under different growth settings.
problem Analyzing the behavior of deep Stable neural networks as width increases.
method Large-width asymptotic analysis and convergence rates for fully connected feed-forward deep Stable NNs.
result The rescaled deep Stable NN converges weakly to a Stable SP under joint growth, with sup-norm convergence rates established.
Divide-and-conquer framework speeds up black-box inference for large data.
problem Computational intractability of uncertainty quantification for expensive data simulation.
method Divide data into partitions, train on a subset, bootstrap on partitions, combine results.
result Feasibility of estimating max-stable process parameters with tens of thousands of locations.
New framework improves model robustness by focusing on stable relations across environments.
problem Standard supervised learning fails under data distribution shift.
method Gradient-based learning framework derived from the principle of independent causal mechanisms (ICM).
result Models generalize well to unseen scenarios, ignoring unstable relations.
New method shows random, diverse initializations are not essential for deep neural networks.
problem The necessity of random, diverse initializations in deep neural networks.
method Constructed a deep convolutional network with identical features by initializing weights to 0, enabling signal propagation and stable gradients.
result Random, diverse initializations are not necessary for training neural networks.
We introduce new families of Integral Probability Metrics (IPM) for training Generative Adversarial Networks (GAN). Our IPMs are based on matching statistics of distributions embedded in a finite dimensional feature space. Mean and covariance feature matching IPMs allow for stable training of GANs, which we will call M…
This note gives a brief survey of the minimum dilatation problem for pseudo-Anosov mapping classes, and the first explicit train track description of an infinite family of pseudo-Anosov mapping classes with orientable stable foliations and the conjectural minimum dilatation for closed surfaces of even genus g≥2.
WSqD extends learning rate schedules for large model training without fixed horizons.
problem Fixed learning rate schedules limit training horizon extension.
method WSqD replaces constant stable phase with a shifted inverse-square-root base, retaining linear cooldown.
result WSqD achieves minimax-optimal convergence rate and horizon-independence.
GradPower speeds up language model training with minimal code changes.
problem Slower training of large language models.
method Elementwise sign-power transformation applied to gradients.
result Consistently lower terminal loss across various models and datasets.
Math proves deep learning unstable, despite stable neural networks existing.
problem Unstable neural networks in deep learning despite stable ones existing.
method Mathematical proof showing instability of current training procedures.
result Proven existence of stable and accurate neural networks with variable dimensions, but current algorithms cannot compute them.
Proposes a stable classifier using inflated argmax for multiclass classification.
problem Inherent instability of taking the maximizer in multiclass classification.
method Bagging for stable continuous scores, inflated argmax for stable labels.
result Inflated argmax provides necessary protection against unstable classifiers without loss of accuracy.