Get Started
Home
Topics
Search
Library
Research questionHow can contrastive distillation efficiently transfer representations to smaller students without memory banks or fixed temperatures?Contrastive distillation can improve student representations, but negative sampling may require external memory banks and fixed temperatures may not suit different training stages or teacher–student pairs.
AI
Computer Vision
Machine Learning
Research Paper
Latest papersRecent research connected to this question, newest first.Discriminative and Consistent Representation DistillationThe source addresses representation distillation for image classification, object detection, and cross-dataset transfer. It proposes in-batch sampling, cross-model similarity consistency, and learnable scale and bias parameters, with experiments on CIFAR-100, ImageNet, MS-COCO, STL-10, and Tiny ImageNet.research paper · Sep 2, 2026
Related questions
How can model distillation block hidden teacher-trait transfer through clean data without degrading the target task?How can we distill capable small language models with fewer training tokens without losing teacher behavior?How can one deployable LLM learn from multiple specialized teachers when the most reliable teacher varies by sample?How can weight-inheritance distillation train the full deployed student matrix rather than a restricted slice?