Imagen 2 y cuestionario de difusión optimizado para recompensas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

1. ¿Qué técnica generativa central utiliza Imagen 2?
   • Difusión que elimina el ruido aleatorio en una imagen.
   • Predicción de tokens autorregresivos
   • Algoritmos genéticos
   • Dibujo vectorial basado en reglas
2. In reward-tuned text-to-image research, what does a learned reward model do?
   • It removes the image diffusion process
   • It scores generated images according to patterns in human feedback
   • It converts images into shipping labels
   • It replaces the text prompt with random noise
3. How is human-preference tuning for text-to-image models often compared with language-model alignment?
   • It is the same process as WordPiece tokenization
   • It uses only hand-written image rules
   • It is conceptually similar to reinforcement learning from human feedback (RLHF)
   • It removes all human judgments from training
4. In the cited reward-weighted text-to-image method, what objective guides fine-tuning?
   • Minimizing image file size regardless of the prompt
   • Predicting the next text token without using image data
   • Maximizing the number of colors in every output
   • Maximizing likelihood weighted by a reward function trained on human feedback
5. In supported Imagen 2 workflows, what does SynthID do to generated images?
   • Embedding an invisible watermark to help identify AI-generated images
   • Resolución mejorada
   • Traduciendo el mensaje
   • Comprimiendo el archivo
6. Which Imagen 2 improvement did Google specifically report in its 2024 announcement?
   • Native generation of spoken audio
   • More realistic hands and faces and fewer distracting visual artifacts
   • A guarantee that every object count is exact
   • Replacing diffusion with autoregressive text prediction
7. What tradeoff does classifier-free guidance control in diffusion generation?
   • Image width versus token vocabulary size
   • Watermark visibility versus font size
   • Prompt adherence versus sample diversity
   • Training data volume versus ship speed
8. What did Google document for Imagen 2.0 on Vertex AI in 2024?
   • Real-time speech transcription
   • Automatic conversion of images into 3D CAD files
   • A guarantee of exact text rendering in every image
   • Inpainting and outpainting for photo editing