Multimodal AI Systems: Architectures, Training, and Applications (Transformer Principles Series)

By Wei Sun

The Transformer Principles Series is a three-volume graduate-level treatise that builds a complete mathematical and engineering understanding of modern AI systems, from the foundational attention mechanism to large language models and multimodal architectures.

Volume III - Multimodal AI Systems: Architectures, Training, and Applications extends the Transformer paradigm beyond text into vision, audio, and video. It covers modality-specific encoders and tokenizers, cross-modal fusion and contrastive alignment (CLIP, SigLIP), diffusion and flow-matching generative models, vision-language architectures (ViT, LLaVA, Q-Former), text-to-image and text-to-video generation, speech and audio processing, efficient inference for multimodal models, long-context scaling, and reasoning agents that perceive and act across modalities.

• Format: Paperback
• Language: English
• ISBN: B0H6NW8LG1
• ISBN13: 9798184326054
• Release Date: June 2026
• Publisher: Independently published
• Length: 478 Pages
• Weight: 2.43 lbs.
• Dimensions: 11.0" x 1.1" x 8.5"