Cooperate
RU/EN
Compression Team
Team delivers the methods of Deep Learning models compression
BUSINESS CASES
  • RAM, Energy, CPU/GPU lower consumption
  • On-device models transfer
  • Models usage on low-bit CPU
  • Speeding up calculations
TASKS CLASSES
  • Low-bit and Post-Training Quantization for complex architectures (Img-to-img, Transformer)
  • Knowledge Distillation & Pruning
  • Neural Architecture Search & SuperNets
CURRENT CHALLENGES
  • Methods for complex architectures such as Transformers and img2img
  • Unification of methods usability, transition to end-to-end optimization and on-device transfer
ACHIEVEMENTS
START COOPERATION
Contact Compression Team