Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Preference-Driven Multi-Objective Combinatorial Optimization with Conditional Computation
Authors: Mingfeng Fan, Jianan Zhou, Yifeng Zhang, Yaoxin Wu, Jinbiao Chen, Guillaume Sartoretti
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the efficacy and versatility of POCCO by applying it to two state-of-the-art neural methods for MOCOPs. Experimental results across four classic MOCOP benchmarks demonstrate its significant superiority and strong generalization. |
| Researcher Affiliation | Academia | Mingfeng Fan National University of Singapore EMAIL Jianan Zhou Nanyang Technological University EMAIL Yifeng Zhang National University of Singapore EMAIL Yaoxin Wu Eindhoven University of Technology EMAIL Jinbiao Chen Sun Yat-sen University EMAIL Guillaume Adrien Sartoretti National University of Singapore EMAIL |
| Pseudocode | Yes | Algorithm 1 Preference-driven MOCO Input: Instance distribution G, weight vector distribution λ, number of training steps E, batch size B, number of tours K per subproblem; Output: The trained policy network θ; 1: Initialize policy network θ. 2: for e = 1 to E do 3: λb SAMPLEWEIGHTVECTOR( λ); Gb SAMPLEINSTANCE ( G), b {1, , B} 4: πj,b SAMPLESOLUTIONS(pθ( |Gb, λb)), j {1, , K} , b {1, , B} 5: yb j,p PAIRWISEPREFERENCE(1[πj,b πp,b]), j, p {1, K}, b {1, , B} 6: Calculate gradient θL(θ) according to Eq. (5) 7: θ ADAM(θ, θL(θ)) 8: end for |
| Open Source Code | Yes | The code and dataset are publicly released for reproducibility.2 2https://github.com/mingfan321/POCCO |
| Open Datasets | Yes | We conduct extensive experiments to evaluate the effectiveness of the proposed POCCO across various MOCOPs, including multi-objective traveling salesman problem (MOTSP) [39], multi-objective capacitated vehicle routing problem (MOCVRP) [64], and multi-objective knapsack problem (MOKP) [23]. |
| Dataset Splits | No | We conduct extensive experiments to evaluate the effectiveness of the proposed POCCO across various MOCOPs, including multi-objective traveling salesman problem (MOTSP) [39], multi-objective capacitated vehicle routing problem (MOCVRP) [64], and multi-objective knapsack problem (MOKP) [23]. In this work, we consider three commonly used problem sizes: n = 20/50/100 for MOTSP and MOCVRP, and n = 50/100/200 for MOKP. ... Both models are trained for 200 epochs, with each epoch processing 100,000 randomly sampled instances and a batch size of B = 64. |
| Hardware Specification | Yes | All experiments are implemented in Python and conducted on a machine with NVIDIA Ampere A100-80GB GPUs and an AMD EPYC 7742 CPU. |
| Software Dependencies | No | All experiments are implemented in Python and conducted on a machine with NVIDIA Ampere A100-80GB GPUs and an AMD EPYC 7742 CPU. The code and dataset are publicly released for reproducibility.2 ... We use the Adam optimizer [28] with a learning rate of 3 × 10^−4 and a weight decay of 10^−6. |
| Experiment Setup | Yes | Both models are trained for 200 epochs, with each epoch processing 100,000 randomly sampled instances and a batch size of B = 64. We use the Adam optimizer [28] with a learning rate of 3 × 10^−4 and a weight decay of 10^−6. The N weight vectors used for decomposition are generated following [10], with N = 101 for κ = 2 and N = 105 for κ = 3. |