Muse is a text-to-image transformer model developed by Google Research that outperforms diffusion and autoregressive models in terms of image generation while using significantly less resources. The model is trained on a masked modeling task in a discrete token space, where it predicts randomly masked image tokens using text embeddings from a pre-trained large language model.

Muse's efficiency is due to its use of discrete tokens, which makes it more efficient than pixel-space diffusion models such as Imagen and DALL-E 2, as well as its parallel decoding, which makes it more efficient than autoregressive models like Parti.

If you want to learn more about Muse, you can read about it in this blog.

Categories : Computer Science . Machine Learning

Press Ask Flow below to get a link to the resource

---

Related

• 

  Museion Fellowship: Cultivating Institution-Builders

  The Museion Fellowship, launched by the consultancy Beck&Stone, is a professional program designed to cultivate the capabilities of instit..

  Computer Science . Design . Entrepreneurship . Personal Growth . Others

• 

  Early Career Research Group Leader for Machine Learning in Science

  The Cluster of Excellence "Machine Learning - New Perspectives for Science" at the University of Tübingen is seeking aspiring scientists f..

  Machine Learning . Personal Growth . Others

• 

  a16z Build: Talent Engineer Fellowship

  The Talent Engineer Fellowship by a16z Build is a specialized program designed for individuals who want to sit at the intersection of engi..

  Computer Science . Machine Learning . Entrepreneurship . Personal Growth

• 

  Google Software Engineering Intern, Summer 2027

  Google's Software Engineering Internship in India offers a unique opportunity for students to work on real-world projects that impact mill..

  Computer Science

• 

  IMPRS-TRUST PhD Program in Computer Science

  The International Max Planck Research School for Trustworthy Computing (IMPRS-TRUST) is a world-class doctoral program focused on the fund..

  Computer Science . Machine Learning

• 

  Digital Product School by UnternehmerTUM

  The Digital Product School (DPS) is Europe’s most successful training program for cross-functional teams focused on building digital produ..

  Computer Science . Machine Learning . Design . Personal Growth