Research map / Optimization / Implicit bias and generalization
Stochastic Shared Embeddings: Data-driven Regularization of Embedding Layers
Liwei Wu, Shuqing Li, Cho-Jui Hsieh, James L Sharpnack
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SSE-Graph instead of graph Laplacian regularization | MF on Movielens1m · Movielens1m | RMSE: 1.0464 -> 1.0145 | Improved |
| Use SSE-Graph instead of graph Laplacian regularization | MF on Movielens10m · Movielens10m | RMSE: 1.9755 -> 1.9019 | Improved |
| Add SSE-SE to MF | MF · Douban; Movielens10m; Netflix | RMSE: 0.7339 -> 0.7201; 0.8851 -> 0.8715; 0.8941 -> 0.8842 | Improved |
| Add SSE-SE alongside dropout | MF · Douban; Movielens10m; Netflix | RMSE: 0.7296 -> 0.7185; 0.8813 -> 0.8678; 0.8897 -> 0.8790 | Improved |
| Replace dropout with SSE-SE | BPR · Movielens1m; Yahoo Music; Foursquare | P@5: 0.6548 -> 0.6813; 0.3315 -> 0.3498; 0.0184 -> 0.0262 | Mixed |
| Add SSE-SE to SASRec input embeddings only (ablation) | SASRec (2 blocks) · Movielens1m | NDCG@10: 0.5941 -> 0.6092 | Improved |
| Add SSE-SE to SASRec output embeddings only (ablation) | SASRec (2 blocks) · Movielens1m | NDCG@10: 0.5941 -> 0.6085 | Improved |
| Add SSE-SE to both SASRec input and output embeddings (ablation) | SASRec (2 blocks) · Movielens1m | NDCG@10: 0.5941 -> 0.6200 | Improved |
| Increase SASRec depth from 2 to 6 blocks while using SSE-SE on both embedding types (ablation) | SASRec · Movielens1m | NDCG@10: 0.6200 -> 0.6265 | Improved |
| Add SSE-SE to Transformer encoder and decoder embeddings | Transformer · WMT English-to-German newstest, 2008–2018 | BLEU: 21.0 -> 21.4; 20.7 -> 21.1; 22.7 -> 23.0; 20.6 -> 21.0; 20.6 -> 20.8; 25.3 -> 25.2; 26.2 -> 27.2; 28.4 -> 29.2; 32.1 -> | Mixed |
| Add SSE-SE during BERT pre-training | BERT-Base · IMDB Test Set | Accuracy: 0.9518 -> 0.9542 | Improved |
| Add SSE-SE during BERT pre-training | BERT-Base · SST-2 Test Set | Accuracy (%): 93.8 -> 94.3 | Improved |
Closest papers by meaning
- Asymptotics of $\ell_2$ Regularized Network Embeddings (NeurIPS 2022 · similarity 0.87)
- On the Theoretical Limitations of Embedding-based Link Prediction (ICML 2026 · similarity 0.86)
- Asymptotics of Network Embeddings Learned via Subsampling (JMLR 2023 · similarity 0.86)
- DGE: Deep Generative Network Embedding Based on Commonality and Individuality (AAAI 2020 · similarity 0.86)
- Graph Representation Learning via Ladder Gamma Variational Autoencoders (AAAI 2020 · similarity 0.85)
- Feature Learning in $L_2$-regularized DNNs: Attraction/Repulsion and Sparsity (NeurIPS 2022 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
