notesum.ai
Published at November 19Large Language Models for Combinatorial Optimization of Design Structure Matrix
cs.CE
cs.AI
cs.CL
Released Date: November 19, 2024
Authors: Shuo Jiang1, Min Xie1, Jianxi Luo1
Aff.: 1Department of Systems Engineering, City University of Hong Kong, 83 Tat Chee Ave, Kowloon Tong, Hong Kong
| Methods | Activity-Based DSMs | Parameter-Based DSMs | ||
|---|---|---|---|---|
| Unmanned Aerial Vehicle | Microfilm Cartridge | Heat Exchanger | Automobile Brake System | |
| Stochastic Methods111Consistent with Figure 3, all three GA settings show optimization performance with number of unique sequences explored = 10,000. | ||||
| GA (Exploration-focused setting) | 6.0±0.0 | 8.1±0.3 | 5.7±1.0 | 3.8±0.7 |
| GA (Exploitation-focused setting) | 7.4±1.5 | 9.9±1.4 | 7.6±1.4 | 6.6±1.7 |
| GA (Balanced setting) | 6.0±0.0 | 8.4±0.5 | 6.2±1.2 | 4.2±1.2 |
| Deterministic Methods222All deterministic methods can be regarded as single-trial approaches, as they produce consistent outcomes for identical inputs across repeated executions. However, certain nodes might share the same measures. For these nodes, a random order is applied, and the final statistical evaluation results are obtained from 10 runs. | ||||
| Out-In Degree [28] | 10.0±0.0 | 12.3±0.5 | 10.3±0.6 | 5.7±0.9 |
| Eigenvector [29] | 15.0±0.0 | 13.8±0.7 | 13.1±1.7 | 11.1±1.4 |
| Walk-based (Exponential) [30] | 15.0±0.0 | 12.0±0.0 | 8.0±0.0 | 11.0±0.0 |
| Walk-based (Resolvent) [31] | 9.0±0.0 | 12.0±0.0 | 8.0±0.0 | 11.0±0.0 |
| Visibility [32] | 25.6±2.2 | 8.8±0.7 | 6.0±1.3 | 3.0±0.0 |
| LLM-driven Methods (Ours) | ||||
| Single-trial with knowledge | 6.6±0.7 | 8.0±0.0 | 4.8±0.6 | 4.4±0.9 |
| Single-trial without knowledge | 11.9±3.4 | 8.1±0.3 | 5.8±1.1 | 5.9±1.4 |
| 5-trial with knowledge | 6.1±0.3 | 8.0±0.0 | 4.0±0.4 | 3.0±0.0 |
| 5-trial without knowledge | 7.5±0.9 | 8.0±0.0 | 4.9±0.5 | 4.0±1.0 |
| 20-trial with knowledge | 6.0±0.0 | 8.0±0.0 | 3.6±0.5 | 3.0±0.0 |
| 20-trial without knowledge | 6.4±0.7 | 8.0±0.0 | 4.1±0.3 | 3.4±0.7 |