Gaussian Mixture Compression
- Research topic/area
- Artificial Intelligence
- Type of thesis
- Bachelor / Master
- Start time
- -
- Application deadline
- 30.06.2028
- Duration of the thesis
- 4 months(BSc) - 6 months(MSc)
Description
Model compression for Gaussian mixture is compelling for several reasons. First, expectation maximization is non convex, often requiring multiple random restarts; compressing a well converged model preserves its hard won optimum and avoids repeated runs. Second, compression without retraining is a major advantage, delivering smaller footprints and faster inference while keeping the learned distribution intact. Third, maintaining multiple storage and compute tiers of the same model—full, medium, and ultra-light—mirrors the ChatGPT-4 and 4-mini pattern: a unified capability surface scaled for latency and cost. This process enables adaptive deployment, edge compatibility, and efficient A/B testing without duplicating training pipelines and simplifies fleet managementRequirement
- Requirements for students
-
- There are no hard constraints but the more programming and math you know the more you can have fun while doing the project.
- Faculty departments
-
- Engineering sciences
Informatics
- Engineering sciences
Supervision
- Title, first name, last name
- Ali Darijani
- Organizational unit
- Computer Science(IAR/IES)
- Email address
- ali.darijani@iosb.fraunhofer.de
- Link to personal homepage/personal page
- Website
Application via email
- Application documents
-
E-Mail Address for application
Senden Sie die oben genannten Bewerbungsunterlagen bitte per Mail an ali.darijani@iosb.fraunhofer.de
Back