Shared Embedding

Shared embedding techniques aim to improve efficiency and interpretability in machine learning models by representing different data points (e.g., users, items, variables) as vectors in a common space. Current research focuses on applying shared embeddings in multi-task learning scenarios, particularly for recommendation systems, and addressing challenges like negative transfer and cold-start problems through methods such as task-specific gating networks and reinforcement learning-based embedding assignment. This approach offers benefits in terms of improved model performance, reduced memory consumption, and enhanced interpretability, impacting various applications from recommendation systems to federated learning environments where privacy is paramount.

Papers

May 10, 2024

Interpretable Multi-task Learning with Shared Variable Embeddings
Maciej Żelaszczyk, Jacek Mańdziuk
Inherent Interpretability Sparse Attention Input Embeddings Shared Embedding

August 16, 2023

STEM: Unleashing the Power of Embeddings for Multi-task Recommendation
Liangcai Su, Junwei Pan, Ximei Wang, Xi Xiao, Shijie Quan, Xihua Chen, Jie Jiang
Multi Task Learning Multi Task Jina Embeddings Real Power STEM Education Task Specific Embeddings Task Preference Shared Embedding

August 14, 2023

AutoAssign+: Automatic Shared Embedding Assignment in Streaming Recommendation
Ziru Liu, Kecheng Chen, Fengyi Song, Bo Chen, Xiangyu Zhao, Huifeng Guo, Ruiming Tang
Recommender System Recommendation Performance Identity Hider Shared Embedding

October 3, 2022

Federated Graph-based Networks with Shared Embedding
Tianyi Yu, Pei Lai, Fei Teng
Graph Based Centralized Learning Based Network Shared Embedding

Shared Embedding

Papers

Interpretable Multi-task Learning with Shared Variable Embeddings

STEM: Unleashing the Power of Embeddings for Multi-task Recommendation

AutoAssign+: Automatic Shared Embedding Assignment in Streaming Recommendation

Federated Graph-based Networks with Shared Embedding