Table 1 Comparison of Transformer-based Fusion Approaches.

From: Multi-source heterogeneous data fusion and intelligent prediction modeling for chemical engineering construction projects based on improved transformer architecture

Approach

Data Types

Attention Mechanism

Domain Adaptation

Industrial Validation

Standard Transformer9

Homogeneous sequences

Single-scale self-attention

General purpose

Limited

Vision Transformer10

Image + text

Patch-based attention

Computer vision

None

TimesFormer11

Video sequences

Spatial–temporal attention

Video analysis

None

Industrial BERT12

Text + numerical

Pre-trained embeddings

Manufacturing

Simulated data

Proposed Method

Multi-modal heterogeneous

Multi-scale adaptive

Chemical engineering

Real-world deployment

Back to article page