Vision Transformer Training With Decoupled Weight Decay

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Vision Transformer (ViT) neural networks face challenges in achieving state-of-the-art performance on downstream tasks while maintaining computational efficiency, particularly when scaled up with increased parameters.

Innovation Solution

Modifications to the ViT architecture and training process, including decoupling weight decay values, generating classification outputs from all output elements without extra positional inputs, and storing momentum values with reduced precision, allow for effective training of large-scale ViT networks that improve performance on downstream tasks without excessive resource consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If ViT model size is increased to achieve better downstream task performance, then classification accuracy improves, but computational resource consumption and memory usage increase excessively

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent changes key training parameters including decoupling weight decay values for different parameter groups, using reduced precision (e.g., bfloat16) for momentum values in optimizers, and adjusting learning rate schedules. These parameter changes enable larger model sizes to be trained efficiently without proportionally increasing computational resource consumption, thus improving classification accuracy while controlling energy usage.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If ViT model size is increased to achieve better downstream task performance, then classification accuracy improves, but memory usage increases excessively

Engineering Contradiction:
Improveclassification accuracyVSAvoidmemory usage
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent employs reduced precision storage for momentum values in optimizer states, using formats like bfloat16 instead of full precision floating point. This parameter change reduces the memory footprint of training state variables, enabling larger model sizes that improve classification accuracy without proportionally increasing memory usage.

Inventive Principle:
Principle #35Parameter changes

3Stability of the object's composition

If weight decay is applied uniformly to all parameters, then training stability is maintained, but performance on downstream tasks deteriorates

Engineering Contradiction:
Improvetraining stabilityVSAvoiddownstream task performance
Core Design Contradiction:
Stability of the object's compositionVSMeasurement precision

Solution Approach 1:

The patent applies different weight decay values to different parameter groups within the model. Specifically, it decouples weight decay for transformer block parameters versus other parameters, allowing each group to have optimized regularization strength. This local differentiation maintains training stability for core transformer operations while improving downstream task performance by adjusting regularization appropriately for different parameter types.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4095758B1Training large-scale vision transformer neural networks
Publication Date: 2026.05.20 GOOGLE LLC
  • EP4095758B1 patent drawingFigure 1
  • EP4095758B1 patent drawingFigure 2
  • EP4095758B1 patent drawingFigure 3A

AI summary

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training Vision Transformer (ViT) neural networks.