LIMIP: Lifelong Learning to Solve Mixed Integer Programs

Authors: Sahil Manchanda, Sayan Ranu

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate LIMIP on a series of NP-hard problems and establish that in comparison to existing baselines, LIMIP is up to 50% better when confronted with lifelong learning.
Researcher Affiliation Academia Sahil Manchanda and Sayan Ranu Department of Computer Science and Engineering Indian Institute of Technology, Delhi {sahil.manchanda,sayanranu}@cse.iitd.ac.in
Pseudocode Yes The detailed steps of training a sequence of tasks in lifelong fashion through LIMIP are described in Algorithm 1 in Appendix A.4.
Open Source Code Yes The codebase can be found on https://github.com/ideaiitd/Li MIP .
Open Datasets No The paper describes generating its own datasets based on known problems like Set Cover and Independent Set, but does not provide concrete access information (link, DOI, specific citation with authors/year) for publicly available versions of the exact datasets used for training. For example, it cites Balas and Ho (1980) for Set Cover, which defines the problem, not a specific dataset instance used for training.
Dataset Splits Yes For each task, we generate 150,000 branching samples extracted using 10,000 generated instances for training and 30000 validation/test samples generated using 2000 instances.
Hardware Specification Yes We use a system running on Intel Xeon 6248 processor with 96 cores and 1 NVIDIA A100 GPU with 40GB memory for our experiments.
Software Dependencies Yes We use SCIP (Gamrath et al. 2020) as the backend solver... Gamrath, G.; Anderson, D.; Bestuzheva, K.; Chen, W.-K.; Eifler, L.; Gasse, M.; Gemander, P.; Gleixner, A.; Gottwald, L.; and Halbig, K. 2020. The SCIP Optimization Suite 7.0. ZIB-Report.
Experiment Setup Yes 4.2 Experimental Setup and Parameters ... We use attention mechanism with 2 heads. We set the default buffer size to 500. For details of all parameters and system settings, we refer to App A.6.