Sun Jan 29 2023
Thu Jan 26 2023

MusicLM: Generating Music from Text

Music generation
Natural Language Processing
Music composition

Presents MusicLM, a model for generating high-fidelity music from text. MusicLM generates music at 24 kHz that remains consistent over several minutes.

AI-generated music can be used for background music, advertisements, and even content creation for social media channels.

DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature

Language models
Machine learning
Natural Language Processing
Academic institutions
Content creators

Detects samples from pre-trained LLMs using the local curvature of the model's log probability function.

DetectGPT can help businesses and institutions detect if a piece of text is machine-generated, which can be useful for plagiarism detection and evaluating student performance.

Text-To-4D Dynamic Scene Generation

Neural Radiance Fields
Text-to-Video
Computer Vision
Gaming and entertainment
Architecture
Film production

Presents MAV3D (Make-A-Video3D), a method for generating three-dimensional dynamic scenes from text descriptions.

MAV3D can be used to create virtual environments for gaming, architecture, and film production.

simple diffusion: End-to-end diffusion for high resolution images

Diffusion models
Computer Vision
Image Generation
High quality image generation for fashion or design industry

Achieves SotA on image generation among diffusion models without sampling modifiers on ImageNet.

This paper proposes simple yet effective techniques to improve denoising diffusion for high resolution images, achieving state-of-the-art image generation among diffusion models without sampling modifiers on ImageNet. This can be useful for businesses that require high quality image generation, such as in the fashion or design industry.

Molecular Language Model as Multi-task Generator

Deep Generative Models
Natural Language Processing
Molecular Generation
Molecular generation for drug discovery or materials design

Proposes MolGen, a pre-trained molecular language model that effectively learns and shares knowledge across multiple generation tasks and domains.

This paper proposes MolGen, a pre-trained molecular language model that efficiently generates molecules with desired properties and can learn and share knowledge across multiple generation tasks and domains. This can be useful for businesses in the pharmaceutical or chemical industries that require molecular generation for drug discovery or materials design.

Wed Jan 25 2023
Tue Jan 24 2023
Mon Jan 23 2023
Thu Jan 19 2023