A new method trains deep networks by separating weight locations from values.
problem Training deep networks efficiently and effectively.
method Lookahead Permutation (LaPerm) to train DNNs by reconnecting weights.
result LaPerm can train DNNs with random and dense, sparse, or single-valued initial weights.
AWP improves robustness by flattening weight loss landscape.
problem Improving robustness of deep neural networks against adversarial examples.
method Explicitly regularizes the flatness of weight loss landscape through adversarial weight perturbation.
result AWP forms a double-perturbation mechanism in adversarial training, leading to flatter weight loss landscape.
Adversarial training makes logistic regression weight loss landscapes sharper.
problem Understanding why adversarial training sharpens the weight loss landscape in logistic regression.
method Theoretical analysis of linear logistic regression model with L2 norm constraints, and experiments on ResNet18.
result Adversarial training sharpens the weight loss landscape in linear logistic regression models.
New methods train neural networks without changing weights, achieving similar or higher performance.
problem Training neural networks efficiently with randomly initialized weights.
method Switching connections on and off, flipping weights' signs, minimizing changed connections.
result Achieves similar or higher performance with less computational cost than training all weights.
Researchers develop method to protect against 'weight poisoning' attacks on pre-trained models.
problem The security threat of downloading untrusted pre-trained weights that can be manipulated after fine-tuning.
method RIPPLe regularization method and Embedding Surgery initialization procedure.
result Demonstrated that weight poisoning attacks are possible even with limited knowledge of the dataset and fine-tuning procedure.
Improved LLM pre-training performance through better weight and variance control.
problem Improper weight and variance control in LLM pre-training affects downstream task performance.
method Introduced Layer Index Rescaling (LIR) and Target Variance Rescaling (TVR) techniques.
result Substantial improvements in downstream task performance (up to 4.6%) and reduced extreme activation values.
Random weights in GNNs match learned weights in performance.
problem Feature rank collapse in GNNs.
method Replacing learned weights with random weights.
result Random weights achieve comparable performance to learned weights, reducing training time and memory usage.
This paper optimizes deep learning training by efficiently sharding weight updates across replicas.
problem Redundant weight update computation on all replicas in data-parallel training.
method Automatic sharding of weight updates using static analysis and transformations on the training graph.
result Substantial speedups achieved on large-scale models using Cloud TPUs.
Weight decay stabilizes training dynamics by slowing progressive sharpening.
problem Understanding how weight decay affects training stability in deep learning models.
method Analyzing weight decay effects at the Edge of Stability, developing a mathematical framework.
result Weight decay dampens oscillations and stabilizes sharpness in CNNs, causing a phase transition in MLPs.
This paper reveals periodic behavior in neural network training with BN and weight decay.
problem Understanding the dynamics of neural network training with BN and weight decay.
method Rigorous investigation of empirical and theoretical mechanisms.
result Periodic behavior in training is a generalization of previously opposing perspectives.
MLDS dataset reveals hidden model behavior via weight-space analysis.
problem Neural networks' opacity makes them hard to evaluate.
method Presented MLDS dataset of trained neural networks.
result Weight-space analysis reveals meaningful divergence with small changes in training data.
Paper tackles model collapse in recursive generative models using a weighted training scheme.
problem Model collapse in recursive generative models trained on synthetic data.
method Iteratively trains models on real and synthetic data, evaluates weighted training schemes.
result Optimal weighting scheme for synthetic data follows a unified expression, revealing a trade-off with model performance.
Optimizes weights for better model performance in shifting data.
problem Improper importance weighting leads to poor model performance in data shifts.
method Interprets weights as a bias-variance trade-off and optimizes them simultaneously with model parameters.
result Optimizing weights significantly improves model generalization performance.
Two retraining techniques outperform fine-tuning in neural network pruning.
problem Improving accuracy and compression in neural network pruning.
method Weight rewinding and learning rate rewinding compared to fine-tuning.
result Rewinding techniques outperform fine-tuning in accuracy and compression.
Proposes QDF to improve multi-step time-series forecasting.
problem Ignoring label autocorrelation and unequal task weights in training objectives.
method Quadratic-form weighted training objective and QDF learning algorithm.
result Improves performance of various forecast models, achieving state-of-the-art results.
Weight Decay induces low-rank weight matrices in neural networks, improving generalization.
problem Improving generalization in neural networks.
method Training ReLU NN with Weight Decay and Stochastic Gradient Descent.
result The weight matrix of a trained NN is approximately rank-two.
Neural network training is computationally and memory intensive. Sparse training can reduce the burden on emerging hardware platforms designed to accelerate sparse computations, but it can affect network convergence. In this work, we propose a novel CNN training algorithm Sparse Weight Activation Training (SWAT). SWAT …
Recent breakthroughs in computer vision make use of large deep neural networks, utilizing the substantial speedup offered by GPUs. For applications running on limited hardware, however, high precision real-time processing can still be a challenge. One approach to solving this problem is training networks with binary or…
Neural networks' weights don't converge to stationary points but training loss stabilizes.
problem The disconnect between theoretical analyses and neural network training practice.
method An invariant measure perspective inspired by ergodic theory of dynamical systems.
result The distribution of weights converges to an approximate invariant measure, explaining loss stabilization.
Method measures weight similarity in neural networks using normalization and statistical inference.
problem Quantifying weight similarity in non-convex neural networks.
method Chain normalization rule and hypothesis-training-testing statistical inference.
result Weights of identical neural networks converge to similar local solutions.
WASH trains ensembles with shuffled weights to improve accuracy and reduce communication.
problem Training ensembles for weight averaging leads to models converging to different loss basins.
method WASH randomly shuffles a small percentage of weights during training to keep models within the same basin.
result WASH achieves state-of-the-art image classification accuracy with lower communication costs.
Enhances physics-informed neural networks with adaptive sampling and weighting.
problem Challenges in training physics-informed neural networks on complex problems.
method Hybrid adaptive sampling and weighting method.
result Consistently improves prediction accuracy and training efficiency.
A new method for deep learning under distribution shift by iteratively refining importance weighting.
problem Handling distribution shift in deep learning models when training and test data distributions differ.
method Dynamic Importance Weighting (dynamic IW) that iterates between weight estimation and weighted classification, using a pre-trained feature extractor and stochastic optimization.
result Dynamic IW outperforms state-of-the-art methods in experiments with various types of distribution shift on multiple datasets.
Predicting neural network accuracy from weights without testing.
problem Predicting neural network performance based on weights alone.
method Used simple statistics of weights to rank neural networks' performance.
result Simple predictors can rank networks' performance with high accuracy (R2 score > 0.98).
Generative model initializes 2-layer network weights for small datasets.
problem Approximating functions with 2-layer networks using small datasets and gradient-based training.
method Initialize hidden weights with a learned proposal distribution parameterized as a deep generative model. Refine with gradient-based post-processing and regularization.
result Demonstrates effectiveness of the approach with numerical examples.
For fast and energy-efficient deployment of trained deep neural networks on resource-constrained embedded hardware, each learned weight parameter should ideally be represented and stored using a single bit. Error-rates usually increase when this requirement is imposed. Here, we report large improvements in error rates …
S-SGD adds symmetrical noise to weights to avoid sharp minima in deep learning.
problem SGD does not always converge to a flat minimum, leading to poor generalization.
method Symmetrical weight noise injection in SGD.
result S-SGD outperforms conventional SGD and weight-noise injection methods in large batch training.
RSO uses random weight perturbations to train deep networks without gradients.
problem Training deep neural networks efficiently and without gradient information.
method RSO is a gradient-free Markov Chain Monte Carlo approach that updates weights based on mini-batch loss reduction.
result RSO achieves high accuracy (99.1% on MNIST) with significantly fewer updates than traditional methods.
The paper investigates how neural network weights evolve to monitor training progress.
problem Monitoring the training progress of neural networks in a cost-effective manner.
method Investigates the evolution of neural network weights in weight space.
result DNN models evolve on unique, smooth trajectories in weight space that can be used to track training progress.
Improves transfer learning by weighting importance based on test-over-training density.
problem Distribution shift in training and test data.
method Joint and dynamic importance-predictor estimation, causal mechanism transfer.
result Enhanced transfer learning performance in complex, high-dimensional tasks.
Weibull weight-scale parameter λ evolves during AdamW training, with alignment, injection, and decay forces driving its growth and relaxation.
problem Understanding the evolution of the Weibull weight-scale parameter λ during AdamW training. method Deriving a leading-order three-force decomposition of the squared weight norm from AdamW updates.
result The alignment force dominates the rise phase, contributing 88-94% of the absolute force budget across four random seeds.
The Importance Weighted Auto Encoder (IWAE) objective has been shown to improve the training of generative models over the standard Variational Auto Encoder (VAE) objective. Here, we derive importance weighted extensions to AVB and AAE. These latent variable models use implicitly defined inference networks whose approx…
Prior knowledge can be used to improve predictive performance of learning algorithms or reduce the amount of data required for training. The same goal is pursued within the learning using privileged information paradigm which was recently introduced by Vapnik et al. and is aimed at utilizing additional information avai…
We explore the problem of selectively forgetting a particular subset of the data used for training a deep neural network. While the effects of the data to be forgotten can be hidden from the output of the network, insights may still be gleaned by probing deep into its weights. We propose a method for "scrubbing'" the w…
This paper evaluates heuristics and hyperparameters in weight-sharing NAS methods.
problem Improving the performance of weight-sharing NAS methods.
method Systematic evaluation of heuristics and hyperparameters in weight-sharing NAS algorithms.
result Some heuristics negatively impact super-net and stand-alone performance correlation.
Deep weight factorization improves neural network training through smooth optimization of sparse penalties.
problem Challenges in applying sparse regularization in neural networks due to non-differentiability of penalties.
method Introduces deep weight factorization, decomposing weights into multiple factors for smooth optimization of L1-penalized networks. result Deep weight factorization outperforms shallow factorization and pruning methods consistently across various architectures and datasets.
We propose Stochastic Weight Averaging in Parallel (SWAP), an algorithm to accelerate DNN training. Our algorithm uses large mini-batches to compute an approximate solution quickly and then refines it by averaging the weights of multiple models computed independently and in parallel. The resulting models generalize equ…
New method for training deep neural networks with regularization, converging to better generalization.
problem Improving generalization of deep neural networks through explicit regularization.
method Regularizer Mirror Descent (RMD) method, inspired by convergence properties of stochastic mirror descent (SMD).
result RMD converges to a point close to the minimizer of the cost function, leading to better generalization performance.
Neural networks trained with backpropagation, the standard algorithm of deep learning which uses weight transport, are easily fooled by existing gradient-based adversarial attacks. This class of attacks are based on certain small perturbations of the inputs to make networks misclassify them. We show that less biologica…
Current deep neural networks (DNNs) can easily overfit to biased training data with corrupted labels or class imbalance. Sample re-weighting strategy is commonly used to alleviate this issue by designing a weighting function mapping from training loss to sample weight, and then iterating between weight recalculating an…
Constraints improve deep neural network training by stabilizing and enhancing robustness.
problem Vanishing/exploding gradients and poor weight magnitudes in deep neural networks.
method Weight-constrained stochastic dynamics using Langevin dynamics framework.
result Enhanced exploration of the loss landscape and improved generalization.
Paper proves multiplicative weight updates can train neural networks without learning rate tuning.
problem Vanishing and exploding gradients in gradient descent for compositional functions.
method Proves descent lemma for compositional functions using multiplicative weight updates and derives Madam optimizer.
result Madam optimizer trains state-of-the-art neural networks without learning rate tuning.
In information retrieval (IR) and related tasks, term weighting approaches typically consider the frequency of the term in the document and in the collection in order to compute a score reflecting the importance of the term for the document. In tasks characterized by the presence of training data (such as text classifi…
Adaptive weights improve physics-informed neural networks and deep operator networks.
problem Training physics-informed neural networks and deep operator networks can be challenging, leading to unsatisfactory accuracy and efficiency.
method Proposes a pointwise adaptive weighting method that balances the residual decay rate across different training points.
result Our proposed approach of balanced residual decay rates offers advantages including bounded weights, high prediction accuracy, fast convergence rate, low training uncertainty, low computational cost, and ease of hyperparameter tuning.
FAIR method uses adversarial training to learn fair instance weights.
problem Reduces bias in machine learning predictions through fair instance weighting.
method Adversarial training to learn instance weighting function ensuring fair predictions.
result Demonstrates better trade-off between accuracy and fairness compared to other models.
Mimetic initialization improves Transformer training on small datasets.
problem Difficulty in training Transformers on small datasets.
method Initialize self-attention layers to look like pre-trained models.
result Vanilla Transformers trained with mimetic initialization achieve higher accuracy.
Proposes SWA for adversarial training to improve model robustness.
problem Overfitting in adversarial training due to insufficient sample complexity.
method Adversarial training with Stochastic Weight Average (SWA).
result Improves model robustness on CIFAR-10, CIFAR-100, and SVHN datasets.
Averaging recent model checkpoints speeds up training time.
problem Training large vision or language models is time-consuming.
method Average the weights of the k latest checkpoints.
result Speeds up training by dozens of epochs, saving up to 68 GPU hours.