MM-TS: Multi-Modal Temperature and Margin Schedules for Contrastive Learning with Long-Tail Data

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• MM-TS extends uni-modal temperature scheduling to multi-modal contrastive learning by dynamically adjusting temperature during training.method 95%
• The method adapts temperature based on the local distribution of each training sample to handle long-tail data.method 90%
• Temperature scheduling can be effectively integrated within a max-margin framework.technical 90%
• This integration unifies the InfoNCE loss and max-margin objective approaches in multi-modal contrastive learning.technical 85%
• The proposed dynamic temperature and margin schedules improve performance on evaluated datasets.result 90%
• The method leads to new state-of-the-art results in the field.result 90%
• Standard multi-modal datasets often follow imbalanced, long-tail distributions.technical 80%
• In uni-modal contrastive learning, the temperature parameter controls the strength of attraction and repulsion forces.technical 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card