CAMBranch: Contrastive Learning with Augmented MILPs for Branching

Jiacheng Lin,Meng Xu,Zhihua Xiong,Huangang Wang

2024-02-06

Abstract:Recent advancements have introduced machine learning frameworks to enhance the Branch and Bound (B\&B) branching policies for solving Mixed Integer Linear Programming (MILP). These methods, primarily relying on imitation learning of Strong Branching, have shown superior performance. However, collecting expert samples for imitation learning, particularly for Strong Branching, is a time-consuming endeavor. To address this challenge, we propose \textbf{C}ontrastive Learning with \textbf{A}ugmented \textbf{M}ILPs for \textbf{Branch}ing (CAMBranch), a framework that generates Augmented MILPs (AMILPs) by applying variable shifting to limited expert data from their original MILPs. This approach enables the acquisition of a considerable number of labeled expert samples. CAMBranch leverages both MILPs and AMILPs for imitation learning and employs contrastive learning to enhance the model's ability to capture MILP features, thereby improving the quality of branching decisions. Experimental results demonstrate that CAMBranch, trained with only 10\% of the complete dataset, exhibits superior performance. Ablation studies further validate the effectiveness of our method.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

### The Problem the Paper Attempts to Solve This paper aims to address the optimization problem of branching strategies in Mixed-Integer Linear Programming (MILP) solving. Specifically: 1. **Background and Challenges**: - **Traditional Methods**: Traditional Branch and Bound (B&B) algorithms rely on manually designed heuristic rules for variable selection (i.e., branching decisions), which require extensive domain knowledge. - **Machine Learning Methods**: In recent years, significant progress has been made by enhancing B&B algorithms' branching strategies through machine learning frameworks, especially imitation learning. However, collecting expert samples for imitation learning (particularly under strong branching strategies) is very time-consuming and computationally complex. 2. **Proposed Method**: - **Data Augmentation**: The paper proposes a new framework called "Contrastive Learning Augmented MILP Branching (CAMBranch)," which generates augmented MILP instances (AMILP) through variable shifting techniques, thereby producing a large number of labeled samples based on a limited set of expert data. - **Contrastive Learning**: By leveraging the same branching decisions between MILP and its corresponding AMILP, the model is trained using contrastive learning methods to improve its ability to capture MILP features, thereby enhancing the quality of branching decisions. 3. **Experimental Results**: - Experimental results show that CAMBranch exhibits superior performance even when using only 10% of the original dataset. Particularly in handling complex problems, it outperforms other methods in terms of solving time, the number of B&B tree nodes, and the number of wins. Through these methods, the paper effectively addresses the time-consuming issue of collecting expert samples and demonstrates a new approach to achieving efficient branching decisions with limited data.

CAMBranch: Contrastive Learning with Augmented MILPs for Branching

Towards Imitation Learning to Branch for MIP: A Hybrid Reinforcement Learning Based Sample Augmentation Approach

Learning to Branch in Mixed Integer Programming

Machine Learning Augmented Branch and Bound for Mixed Integer Linear Programming

A Machine Learning-Based Approximation of Strong Branching

Parameterizing Branch-and-Bound Search Trees to Learn Branching Policies

Learning to Branch: Accelerating Resource Allocation in Wireless Networks

An Enhanced Branch-and-bound Algorithm for Bilevel Integer Linear Programming.

Learning Backdoors for Mixed Integer Linear Programs with Contrastive Learning

Reinforcement Learning for Branch-and-Bound Optimisation using Retrospective Trajectories

Learning to Branch with Tree-aware Branching Transformers

Learning in Spatial Branching: Limitations of Strong Branching Imitation

Multivariable Branching: A 0-1 Knapsack Problem Case Study

Branch and Bound in Mixed Integer Linear Programming Problems: A Survey of Techniques and Trends

Active Learning with Controllable Augmentation Induced Acquisition

Yordle: An Efficient Imitation Learning for Branch and Bound

LIMIP: Lifelong Learning to Solve Mixed Integer Programs

Branch Ranking for Efficient Mixed-Integer Programming Via Offline Ranking-Based Policy Learning.

Branch-Train-MiX: Mixing Expert LLMs into a Mixture-of-Experts LLM

A Multi-Agent Learning Framework for Mixed-Integer Linear Programming

Breaking the mold: The challenge of large scale MARL specialization