Shapley Head Pruning: Identifying and Removing Interference in Multilingual Transformers
–arXiv.org Artificial Intelligence
Multilingual transformer-based models demonstrate remarkable zero and few-shot transfer across languages by learning and reusing language-agnostic features. However, as a fixed-size model acquires more languages, its performance across all languages degrades, a phenomenon termed interference. Often attributed to limited model capacity, interference is commonly addressed by adding additional parameters despite evidence that transformer-based models are overparameterized. In this work, we show that it is possible to reduce interference by instead identifying and pruning language-specific parameters. First, we use Shapley Values, a credit allocation metric from coalitional game theory, to identify attention heads that introduce interference. Then, we show that removing identified attention heads from a fixed model improves performance for a target language on both sentence classification and structural prediction, seeing gains as large as 24.7\%. Finally, we provide insights on language-agnostic and language-specific attention heads using attention visualization.
arXiv.org Artificial Intelligence
Oct-11-2022
- Country:
- Oceania > Australia
- Victoria > Melbourne (0.04)
- New South Wales > Sydney (0.04)
- North America
- Dominican Republic (0.04)
- United States
- Minnesota > Hennepin County
- Minneapolis (0.14)
- Michigan > Washtenaw County
- Ann Arbor (0.04)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- California > Santa Clara County
- Palo Alto (0.04)
- Minnesota > Hennepin County
- Canada > Quebec
- Montreal (0.04)
- Europe
- United Kingdom > England
- Cambridgeshire > Cambridge (0.04)
- Italy > Tuscany
- Florence (0.04)
- Ireland > Leinster
- County Dublin > Dublin (0.04)
- France > Provence-Alpes-Côte d'Azur
- Bouches-du-Rhône > Marseille (0.04)
- Denmark > Capital Region
- Copenhagen (0.04)
- Belgium > Brussels-Capital Region
- Brussels (0.04)
- United Kingdom > England
- Asia > China
- Hong Kong (0.04)
- Oceania > Australia
- Genre:
- Research Report (0.82)
- Industry:
- Leisure & Entertainment (0.34)
- Technology: