Wed Feb 08 2023
Mon Feb 06 2023

Actually, gradient descent can be seen as attention that applies beyond the model's context length!

Transformer attention
Large pretrained language models
Natural Language Processing
Machine Learning
NLP
Language modeling
Model designing

Explains the working mechanism of In-Context Learning in large pretrained language models as a kind of implicit finetuning through Transformer attention that has a dual form of gradient descent based optimization. Offers insights for future model designing.

Can help in understanding how In-Context Learning works and provide insights for future model designing.

Zero-shot Image-to-Image Translation

Image-to-image translation
Pre-trained text-to-image generative models
Computer Vision
Machine Learning
Image editing
Computer Vision

Proposes pix2pix-zero, an image-to-image translation method that can preserve the content of the original image without manual prompting. Outperforms existing and concurrent works for both real and synthetic image editing.

Can improve image editing processes without the need for manual prompting and can outperform existing methods.

Languages are Rewards: Hindsight Finetuning using Human Feedback

Finetuning
Language comprehension capabilities of language models
Natural Language Processing
Machine Learning
Natural Language Processing
Language Modeling

Conditions the model on outputs paired with hindsight feedback and finetunes the model to predict the most preferred output, resulting in better performance on summarization and dialogue tasks.

Can improve language models by learning from any form of feedback, regardless of its polarity, and aligning with human preferences.

Sun Feb 05 2023
Thu Feb 02 2023
Wed Feb 01 2023
Tue Jan 31 2023