Research map / Architectures and efficiency / Pruning and efficient tuning
Gradient-Based Diversity Optimization with Differentiable Top-$k$ Objective
Tianyi Zhou, Sebastian Dalleiger, Ece Calikus, Aristides Gionis
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 |
|---|---|---|---|
| fine-tune NMF with DDT (direct diversity objective) | NMF · Yahoo-R2 | diversity: 0.09 -> 0.98 | Improved |
| fine-tune NMF with DDT | NMF · MovieLens | diversity: 0.62 -> 1.00 | Improved |
| fine-tune NMF with DDT | NMF · Netflix | relevance: 0.88 -> 0.97 | Improved |
| add MDR implicit reweighting | NMF · Netflix | diversity: 0.84 -> 0.98 | Improved |
| use DDT instead of post-hoc DPP reranking | NMF · Yahoo-R2 | relevance: 0.58 -> 0.85 | Improved |
| use DDT instead of DGRec | DGRec · KuaiRec | relevance: 0.18 -> 0.95 | Improved |
| use DDT instead of MDR with neural network model (ablation) | NCF · MovieLens/Netflix | diversity | Improved |
| vary static beta toward diversity (lower beta) (ablation) | DDT/MDR · Netflix, MovieLens | recall | No clear effect |
| use adaptive beta instead of static (ablation) | DDT · MovieLens (NCF) | relevance and diversity | Improved |
| fine-tune pretrained model instead of training from scratch (ablation) | NCF · MovieLens | relevance | Improved |
| apply DDT with k=10 on Yahoo-R2 (NMF) | NMF · Yahoo-R2 | in-objective diversity gain: 77.4 (8.7) | Improved |
| apply DDT to NMF on Coat for items k+1~2k | NMF · Coat | out-of-objective diversity gain: -0.5 (4.7) at k=5 | No clear effect |
Closest papers by meaning
- Welfarist Formulations for Diverse Similarity Search (ICLR 2026 · similarity 0.88)
- Tree of Preferences for Diversified Recommendation (NeurIPS 2025 · similarity 0.84)
- DivGCL: A Graph Contrastive Learning Model for Diverse Recommendation (AAAI 2025 · similarity 0.84)
- From Pairwise to Ranking: Climbing the Ladder to Ideal Collaborative Filtering with Pseudo-Ranking (AAAI 2025 · similarity 0.84)
- Sparse Meets Dense: Unified Generative Recommendations with Cascaded Sparse-Dense Representations (NeurIPS 2025 · similarity 0.83)
- No More Tuning: Prioritized Multi-Task Learning with Lagrangian Differential Multiplier Methods (AAAI 2025 · similarity 0.83)
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.
