Your Research Library

6 documents across 3 collections

Introduces the Transformer architecture based solely on attention mechanisms, dispensing recurrence and convolutions entirely.

Proposes a bidirectional transformer pre-training approach that achieves SOTA on 11 NLP tasks.

Applies transformer architecture directly to image patches for classification at scale.

Presents PPO, a family of policy gradient methods for RL that balances simplicity and performance.

Describes GPT-4, a large-scale multimodal model capable of processing image and text inputs.

Introduces DDPM, connecting diffusion models with denoising score matching for high-quality image generation.