BRIEF: Bi-level Coreset Selection for Efficient Instruction Tuning in LLMs
Summary: BRIEF introduces bi-level coreset selection for LLM instruction tuning, decomposing SFT loss into knowledge vs. instruction-following contributions. A submodular composite-gradient objective yields bounded-approximation subset selection, cutting tuning cost ~3x while improving downstream accuracy. (summarized by gpt-5.4-mini on Apr 12 2026)
Incoming Non-self Citations Over Time
No non-self incoming citations found for this paper in this database.
Authors
- 1. Chaoyuan Shen (Beijing Institute of Technology)
- 2. Chi Zhang (Beijing Institute of Technology)
- 3. Chengliang Chai (Beijing Institute of Technology)
- 4. Jiacheng Wang (Beijing Institute of Technology)
- 5. Jia Yuan (University of Arizona)
- 6. Yuping Wang (Beijing Institute of Technology)
- 7. Ye Yuan (Beijing Institute of Technology)
- 8. Guoren Wang (Beijing Institute of Technology)
- 9. Lei Cao (Massachusetts Institute of Technology)
BibTeX Citation
@article{shen_vldb26,
title = {{BRIEF: Bi-level Coreset Selection for Efficient Instruction Tuning in LLMs}},
author = {Shen, Chaoyuan and Zhang, Chi and Chai, Chengliang and Wang, Jiacheng and Yuan, Jia and Wang, Yuping and Yuan, Ye and Wang, Guoren and Cao, Lei},
journal = {PVLDB},
series = {{VLDB} '26},
volume = {19},
number = {6},
pages = {1264--1277},
doi = {10.14778/3797919.3797933},
url = {https://doi.org/10.14778/3797919.3797933},
year = {2026}
}Incoming Citations (Sorted by Pagerank)
Showing 0 of 0 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 2 of 2 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 3,284 | Camel: Managing Data for Efficient Stream Learning | 2022 | SIGMOD | 7.4560271e-05 |
| 3,941 | GoodCore: Data-effective and Data-efficient Machine Learning through Coreset Selection over Incomplete Data | 2023 | SIGMOD | 6.9138042e-05 |
Previous
Page 1 / 1
Next
Semantically Similar Papers
| # | Overall Rank | Paper | Year | Venue |
|---|---|---|---|---|
| 1 | 10,488 | OctoSelector: Efficient and Effective Batch-Aware Model Selection for Large Language Models | 2026 | SIGMOD |
| 2 | 10,776 | Efficient Cooperation-Aware Key and Value Management for LLM Inference | 2026 | VLDB |
| 3 | 6,860 | SpareLLM: Automatically Selecting Task-Specific Minimum-Cost Large Language Models under Equivalence Constraint | 2025 | SIGMOD |
| 4 | 10,838 | Unified Static–Dynamic Pruning for Efficient LLM Inference | 2026 | VLDB |
| 5 | 11,767 | Self-Training for Label-Efficient Information Extraction from Semi-Structured Web-Pages | 2023 | VLDB |
| 6 | 8,316 | ThriftLLM: On Cost-Effective Selection of Large Language Models for Classification Queries | 2025 | VLDB |
| 7 | 10,821 | Data-efficient Online Training for Direct Alignment in LLMs | 2026 | VLDB |
| 8 | 8,997 | Cut Costs, Not Accuracy: LLM-Powered Data Processing with Guarantees | 2026 | SIGMOD |
| 9 | 8,888 | Optimized Batch Prompting for Cost-effective LLMs | 2025 | VLDB |
| 10 | 7,090 | LEAD: Iterative Data Selection for Efficient LLM Instruction Tuning | 2026 | VLDB |