Quantization for matrix multiplication and LLMs
Event details
| Date | 20.02.2026 |
| Hour | 15:15 › 16:15 |
| Speaker | Yury Polyanskiy, MIT |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Statistical science teaches that parameters can only be learned to within a certain precision due to stochasticity of the measurement operations. Similarly, modern AI research shows that LLM's parameters may be perturbed without significantly affecting the end-to-end performance. Given that the primitive operation that LLM rely on is matrix multiplication, a natural basic question is to understand the tradeoffs and algorithms required for low-precision approximations for this operation. In this talk we will discuss information-theoretic bounds, metric entropy and practical algorithms (NestQuant, WaterSIC) that emerged as by-products of the theory.
Practical information
- Informed public
- Free
Organizer
- Rajita Chandak
Contact
- Maroussia Schaffner