A new scaling law predicts optimal batch size for training models.
problem Finding the optimal batch size for training models efficiently.
method Proposed a three-term scaling law that considers model size, training data, training steps, and batch size.
result The three-term law accurately recovers the optimal batch size and can be robustly fit with fewer training runs.
SNGM improves large-batch training accuracy.
problem Improving generalization in large-batch training.
method Stochastic Normalized Gradient Descent with Momentum.
result SNGM achieves better test accuracy than MSGD and other large-batch methods.
Training deep neural networks with Stochastic Gradient Descent, or its variants, requires careful choice of both learning rate and batch size. While smaller batch sizes generally converge in fewer training epochs, larger batch sizes offer more parallelism and hence better computational efficiency. We have developed a n…
Empirical study on SGD hyperparameters and adversarial robustness.
problem Effect of SGD hyperparameters on adversarial robustness and generalization.
method Empirical observation of learning rate, batch size, and momentum effects on adversarial robustness and generalization.
result Constant learning rate to batch size ratio leads to good generalization and almost constant adversarial robustness.
Distributed implementations of mini-batch stochastic gradient descent (SGD) suffer from communication overheads, attributed to the high frequency of gradient updates inherent in small-batch training. Training with large batches can reduce these overheads; however, large batches can affect the convergence properties and…
Adaptive batch size schedules improve language model training efficiency and generalization.
problem Dilemma of choosing batch sizes in large-scale model training.
method General-purpose adaptive batch size schedules compatible with data and model parallelism.
result Adaptive batch size schedules outperform constant batch sizes and heuristic warmup schedules.
Large batch training improves deep learning performance without needing warmup.
problem Slow convergence at early epochs in large batch training.
method Proposes CLARS algorithm and analyzes convergence rate.
result Proposed algorithm outperforms gradual warmup and state-of-the-art large-batch optimizers.
Proposes a method to train deep models with one-element batches.
problem Training deep models with small batches (one element) is challenging.
method Splits the batch into historical and current elements for training.
result Allows training on higher resolution images with one-element batches.
AdaScale SGD adapts learning rates for large-batch training efficiently.
problem Adapting learning rates for large-batch training to balance speed-ups and model quality.
method Adaptive learning rate adaptation based on gradient variance.
result AdaScale achieves reliable speed-ups for a wide range of batch sizes without degrading model quality.
New research shows many batch selection methods for training work just as well as full batch training.
problem Finding optimal batch selection methods for training.
method Analysis of mini-batch Gradient Descent (GD) and Stochastic GD (SGD) with various batch selection rules.
result All mini-batch schedules, including deterministic ones, generalize optimally for smooth Lipschitz-convex/nonconvex/strongly-convex loss functions.
SWAP uses large mini-batches to train DNNs faster with good generalization.
problem Training deep neural networks with small mini-batches is time-consuming.
method SWAP computes an approximate solution with large mini-batches and refines it by averaging weights of multiple parallel models.
result SWAP trains models as well as small-batch training but in significantly less time.
AdAdaGrad optimizes batch sizes for deep learning models, reducing the generalization gap.
problem The generalization gap between large-batch and small-batch training in deep learning.
method AdAdaGrad introduces adaptive batch size strategies derived from adaptive sampling methods.
result AdAdaGradNorm converges to a first-order stationary point with a rate of O(1/K) in K iterations.
Large batch training with DP-SGD reduces model performance due to implicit bias.
problem Large batch training with DP-SGD reduces model performance.
method The study analyzes the phenomenon of implicit bias in Noisy-SGD (DP-SGD without clipping) and its theoretical solutions for linear models.
result The implicit bias in large batch training with DP-SGD is amplified by additional noise, similar to SGD.
Seesaw optimizes training by balancing learning rate and batch size, accelerating model pretraining.
problem Optimizing training efficiency for large language models with adaptive optimizers.
method Develops a principled framework for batch-size scheduling, introducing Seesaw which multiplies learning rate by 1/√2 and doubles batch size.
result Empirically, Seesaw reduces wall-clock time by approximately 36% compared to cosine decay, matching theoretical limits.
This paper explores the limits of large batch sizes in deep learning.
problem Understanding the optimal batch size for deep learning models.
method Detailed numerical optimization and experimental analysis of large batch sizes.
result Improvement of 18% top-1 test accuracy with an optimized batch size recipe.
Large-batch SGD is important for scaling training of deep neural networks. However, without fine-tuning hyperparameter schedules, the generalization of the model may be hampered. We propose to use batch augmentation: replicating instances of samples within the same batch with different data augmentations. Batch augment…
Modern deep neural network training is typically based on mini-batch stochastic gradient optimization. While the use of large mini-batches increases the available computational parallelism, small batch training has been shown to provide improved generalization performance and allows a significantly smaller memory footp…
New batch selection strategy improves deep learning model performance.
problem Training deep neural networks efficiently.
method Submodular function maximization for mini-batch selection.
result Deep models trained with proposed batch selection outperform SGD and baseline.
Large-batch training approaches have enabled researchers to utilize large-scale distributed processing and greatly accelerate deep-neural net (DNN) training. For example, by scaling the batch size from 256 to 32K, researchers have been able to reduce the training time of ResNet50 on ImageNet from 29 hours to 2.2 minute…
Adaptive batch sizes improve local gradient methods in distributed training.
problem Communication bottlenecks in distributed deep learning.
method Adaptive batch size strategies for local gradient methods.
result Adaptive batch sizes reduce minibatch gradient variance and improve training efficiency.
The most straightforward method to accelerate Stochastic Gradient Descent (SGD) computation is to distribute the randomly selected batch of inputs over multiple processors. To keep the distributed processors fully utilized requires commensurately growing the batch size. However, large batch training often leads to poor…
As an indispensable component, Batch Normalization (BN) has successfully improved the training of deep neural networks (DNNs) with mini-batches, by normalizing the distribution of the internal representation for each hidden layer. However, the effectiveness of BN would diminish with scenario of micro-batch (e.g., less …
Batch normalization makes deep residual networks train faster.
problem Training deep residual networks with large depths.
method Downscaling the residual branch by a normalizing factor early in training.
result Normalized residual blocks compute functions close to the identity function early in training.
This paper benchmarks batch RL algorithms on Atari, finding DQN and partially-trained policies perform best.
problem Deep RL algorithms fail in batch setting.
method Benchmarked batch RL algorithms on Atari using a single partially-trained policy.
result Many batch RL algorithms underperform DQN and partially-trained policies.
In an increasing number of domains it has been demonstrated that deep learning models can be trained using relatively large batch sizes without sacrificing data efficiency. However the limits of this massive data parallelism seem to differ from domain to domain, ranging from batches of tens of thousands in ImageNet to …
Small-GAN speeds up GAN training by using coresets.
problem Slowness and high memory usage in GAN training with large batches.
method Draw a large batch of samples from the prior, compress using Coreset-selection, and use cached Inception activations for random projection.
result Significantly reduces training time and memory usage for modern GAN variants.
SGD's performance improves with critical batch size, minimizing SFO complexity.
problem Optimizing SGD's performance with batch size and learning rate.
method Analysis of SGD using constant and decaying learning rates, focusing on batch size effects.
result SGD with critical batch size minimizes SFO complexity.
Pipelined Backpropagation trains large models without batches efficiently.
problem Training large models efficiently on hardware with limited batch sizes.
method Fine-grained Pipelined Backpropagation with Spike Compensation and Linear Weight Prediction.
result Fine-grained Pipelined Backpropagation with a batch size of one matches the accuracy of SGD for multiple networks.
New analysis reveals batch size effects on stochastic conditional gradient methods.
problem Understanding the role of batch size in stochastic conditional gradient methods.
method Deriving a new analysis focusing on momentum-based stochastic conditional gradient algorithms (e.g., Scion).
result Increasing batch size initially improves optimization accuracy but can degrade performance beyond a critical threshold.
In stochastic optimization, using large batch sizes during training can leverage parallel resources to produce faster wall-clock training times per training epoch. However, for both training loss and testing error, recent results analyzing large batch Stochastic Gradient Descent (SGD) have found sharp diminishing retur…
ABS dynamically adjusts batch size based on policy stability, improving RL performance.
problem Diminishing returns with large batch sizes in RL due to non-stationary data.
method Adaptive Batch Scaling (ABS) with Behavioral Divergence metric.
result Larger batch sizes can improve RL performance, contrary to conventional wisdom.
Training state-of-the-art, deep neural networks is computationally expensive. One way to reduce the training time is to normalize the activities of the neurons. A recently introduced technique called batch normalization uses the distribution of the summed input to a neuron over a mini-batch of training cases to compute…
Recent hardware developments have dramatically increased the scale of data parallelism available for neural network training. Among the simplest ways to harness next-generation hardware is to increase the batch size in standard mini-batch neural network training algorithms. In this work, we aim to experimentally charac…
SP-NGD improves deep learning models' generalization with large mini-batch sizes.
problem Worse generalization performance with large mini-batch sizes in deep learning.
method SP-NGD, a natural gradient descent approach for large-scale deep learning.
result SP-NGD achieves similar generalization performance to first-order methods with accelerated convergence and negligible overhead.
Study shows how algorithmic choices affect optimal batch sizes in neural networks.
problem Understanding how batch size impacts neural network training efficiency.
method Experiments and analysis of a simple quadratic model to study algorithmic choices.
result Preconditioned optimizers like Adam and K-FAC allow larger batch sizes before diminishing returns.
A new method for graph neural networks speeds up inference and training.
problem Challenges in constructing mini-batches for large graphs in graph neural networks.
method Theoretical model of batch construction via maximizing influence score of nodes on outputs.
result Accelerates inference by up to 130x compared to previous methods.
Adam optimizer's bias is influenced by mini-batch size and momentum hyperparameters.
problem Understanding how Adam's implicit bias is affected by mini-batch size and momentum parameters.
method Theoretical framework to analyze mini-batch noise's impact on Adam's memory and bias.
result The magnitude of anti-regularization by memory depends on batch size and momentum hyperparameters.
Zero-shot anomaly detection method using batch normalization.
problem Adapting anomaly detectors to new normal data distributions without training data.
method Adaptive Centered Representations (ACR) with batch normalization.
result First zero-shot AD results for tabular data and image data.
This work improves SGD convergence by adaptively adjusting batch sizes.
problem High variance in gradient estimates with small batch sizes.
method Adaptive batch size adjustment based on model training loss.
result Adaptive batch size method requires fewer model updates with same computation.
BN refines local partition geometry in piecewise-affine networks during training.
problem Understanding the effect of BN on the function realized during training in piecewise-affine networks.
method Analyzing the geometry of switching hyperplanes and affine-region partition conditioned on a mini-batch.
result BN increases expected local partition refinement in ReLU and piecewise-affine networks.
Large batch size training of Neural Networks has been shown to incur accuracy loss when trained with the current methods. The exact underlying reasons for this are still not completely understood. Here, we study large batch size training through the lens of the Hessian operator and robust optimization. In particular, w…
Recurrent Neural Networks (RNNs) are powerful models for sequential data that have the potential to learn long-term dependencies. However, they are computationally expensive to train and difficult to parallelize. Recent work has shown that normalizing intermediate representations of neural networks can significantly im…
In this work we compare different batch construction methods for mini-batch training of recurrent neural networks. While popular implementations like TensorFlow and MXNet suggest a bucketing approach to improve the parallelization capabilities of the recurrent training process, we propose a simple ordering strategy tha…
Recency Bias selects recent uncertain samples for faster, more accurate deep learning.
problem Improving the accuracy of deep neural networks through better mini-batch selection.
method Uses historical label predictions to evaluate predictive uncertainty and selects samples proportionally.
result Reduces test error by up to 20.97% compared to existing methods in the same training time.
Optimal batch size minimizes training time for neural networks.
problem Minimizing training time for two-layer neural networks with SGD.
method Characterized optimal batch size as a function of target hardness (information exponents). Used Correlation loss SGD to overcome limitations.
result Optimal batch size minimizes training time without changing total sample complexity.
Batch normalization (batch norm) is often used in an attempt to stabilize and accelerate training in deep neural networks. In many cases it indeed decreases the number of parameter updates required to achieve low training error. However, it also reduces robustness to small adversarial input perturbations and noise by d…
New framework optimizes deep learning training by deferring large batch sizes to late stages.
problem Optimizing batch size scheduling for deep learning training efficiency.
method Introduced the functional scaling law (FSL) framework to analyze and optimize batch size scheduling.
result Large batch sizes can be deferred to late training stages without sacrificing performance.
New findings show mini-batch SGD operates in a 'Edge of Stochastic Stability' regime.
problem Understanding the stability and convergence of mini-batch SGD.
method Analyzing the mini-batch Hessian and its directional curvature.
result Mini-batch SGD operates in a different stability regime (Edge of Stochastic Stability) compared to full-batch GD.