CAIC 9.6 Flashcards
What does the diffusion model offer in the generation process?
Flexibility and controllability
Users can control the trade-off between sample quality and diversity by adjusting diffusion steps.
What are the domains where diffusion models have shown great promise?
- Computer vision
- Natural language processing
- Audio synthesis
Diffusion models are capable of generating high-quality data with fine-grained details.
What are two popular models based on the diffusion approach?
- DALL-E 2
- Stable Diffusion
These models have been developed by the open-source community and private companies.
What is DALL-E 2?
A text-to-image model developed by OpenAI
It was first released in January 2022 and can generate and manipulate images from text descriptions.
What are some applications of DALL-E 2?
- Inpainting
- Outpainting
- Image-to-image translation
The images generated can be used for creating art and generating marketing materials.
What are the two key steps in DALL-E 2 training?
- CLIP training
- GLIDE training
CLIP learns the semantic linking of text and images, while GLIDE learns to reverse the image from visual embeddings.
What does the CLIP model do?
Outputs a text-conditioned visual encoding
It is trained with hundreds of millions of images and their associated descriptions.
What is the purpose of the GLIDE model in DALL-E 2?
To reverse the image from the visual embeddings generated by CLIP
GLIDE is based on the diffusion model.
What is Stable Diffusion?
An algorithm developed by Compvis and sponsored by Stability AI
It is a text-to-image model effective at generating high-quality images from text descriptions.
What architecture does Stable Diffusion employ?
- CLIP encoder
- UNET as the denoising neural network
It is an open-source model with code and model weights released to the public.
What concerns are associated with diffusion models?
- Copyright infringement
- Creation of harmful images
These concerns arise despite the powerful capabilities of the models.
What is the business problem faced by the retail bank?
High customer churn rate
The bank needs to identify potential churners to offer incentives and prevent them from leaving.
Why is it more expensive for the bank to acquire a new customer?
It is far more expensive than offering incentives to keep an existing customer
Preventive measures are essential to reduce potential churn.
What environment will be set up for the ML experiments?
A Jupyter environment on the local machine
This setup is necessary as there is no ML tooling available.
Where can the dataset for bank customers’ churn be accessed?
Kaggle site
The dataset contains features such as credit score, gender, and balance.
What is the target variable in the bank customers’ churn dataset?
Exited
This variable indicates whether a customer churned or not.
Fill in the blank: The dataset contains 14 columns for features such as credit score, gender, and balance, and a target variable column, _______.
Exited
