Tag: Operations Research

Optimization at Neurips 2024

Optimization was one of the top five topics at NeurIPS 2024, reflecting a clear trend of increasing interest in this area compared to previous years. While optimization in machine learning often focuses on specific tasks like training neural networks (using algorithms such as gradient descent and its variants), my focus here is on optimization in a broader sense.

This includes research exploring the use of AI methods for classical optimization problems, bridging the gap between traditional optimization techniques and modern AI-driven approaches. Below is a curated list of papers that stood out in this domain.

Non-Linear Programming

  1. Dual Lagrangian Learning for Conic Optimization
    Poster | Paper
  2. BPQP: A Differentiable Convex Optimization Framework for Efficient End-to-End Learning
    Poster | Paper
  3. IPM-LSTM: A Learning-Based Interior Point Method for Solving Nonlinear Programs
    Poster | Paper

Combinatorial Optimization

  1. Controlling Continuous Relaxation for Combinatorial Optimization
    Poster | Paper
  2. Learning to Handle Complex Constraints for Vehicle Routing Problems
    Poster | Paper
  3. ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution
    Poster | Paper

Stochastic and Contextual Optimization

  1. Optimal Algorithms for Online Convex Optimization with Adversarial Constraints
    Poster | Paper
  2. Improved Algorithms for Contextual Dynamic Pricing
    Poster | Paper
  3. Conformal Inverse Optimization
    Poster | Paper
  4. Multi-Stage Predict+Optimize for (Mixed Integer) Linear Programs
    Poster | Paper
  5. There is No Silver Bullet: Benchmarking Methods in Predictive Combinatorial Optimization
    Poster | Paper
  6. Regret Minimization in Stackelberg Games with Side Information
    Poster | Paper

Linear Programming and MILP

  1. GLinSAT: The General Linear Satisfiability Neural Network Layer
    Poster | Paper
  2. SymILO: A Symmetry-Aware Learning Framework for Integer Linear Optimization
    Poster | Paper
  3. Learning Generalized Linear Programming Value Functions
    Poster | Paper
  4. Rethinking the Capacity of Graph Neural Networks for Branching Strategy
    Poster | Paper
  5. On the Power of Small-Size Graph Neural Networks for Linear Programming
    Poster | Paper

Application-Specific Optimization

  1. Randomized Sparse Matrix Compression for Large-Scale Constrained Optimization in Cancer Radiotherapy
    Poster | Paper
  2. DistrictNet: Decision-Aware Learning for Geographical Districting
    Poster | Paper
  3. Approximately Pareto-Optimal Solutions for Bi-Objective k-Clustering
    Poster | Paper
  4. Autoregressive Policy Optimization for Constrained Allocation Tasks
    Poster | Paper

New Perspectives

  1. Optimization Algorithm Design via Electric Circuits
    Poster | Paper
  2. FERERO: A Flexible Framework for Preference-Guided Multi-Objective Learning
    Poster | Paper

Optimization for Machine Learning

While there are many works related to optimization in machine learning, the following papers stood out for their relevance to OR techniques.

  1. Safe and Efficient: A Primal-Dual Method for Offline Convex CMDPs under Partial Data Coverage
    Poster | Paper
  2. Gradient-Free Methods for Nonconvex Nonsmooth Stochastic Compositional Optimization
    Poster | Paper
  3. Can Learned Optimization Make Reinforcement Learning Less Difficult?
    Poster | Paper
  4. Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization
    Poster | Paper
  5. SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization
    Poster | Paper
  6. ROIDICE: Offline Return on Investment Maximization for Efficient Decision Making
    Poster | Paper
  7. Adaptive Proximal Gradient Method for Convex Optimization
    Poster | Paper
  8. Functional Bilevel Optimization for Machine Learning
    Poster | Paper
  9. First-Order Minimax Bilevel Optimization
    Poster | Paper

Equity, Shelters and Natural Disaster

Where do you go if a natural disaster impacts your home? What if it is a huge disaster that affects your whole city? City dwellers expect city municipality, federal government crisis management agencies, and humanitarian organizations to provide services to all disaster-affected persons without considering race, gender, class, and sexual orientation. This is only achievable if we explicitly consider the equality of service in the decision-making process.


The picture below shows the map of one of the most populated districts of Istanbul. Istanbul is a disaster-prone city where experts expect a major earthquake in the next ten years. In the figure, green squares represent the potential location of shelters, and the red dots present the center of neighborhoods. Due to budgetary restrictions, humanitarian organizations can only open and maintain a few shelters in each district.

Potential Shelter locations and population concentration points in Kartal district, Istanbul

In a study, we design an intelligent algorithm that takes shelter location decisions considering all operational and budgetary restrictions while reducing the inequality of physical distance to shelters among all affected people. We observe that humanitarian organizations can make inequality-aware decisions and simultaneously increase their operations’ efficiency and reduce costs. For more detail, refer to our published paper. Also, the python implementation of our optimization algorithm is available at GitHub for free.

Online Printing Production Scheduling

It might be the case that you order to print a custom-made happy birthday cart for someone before, or at least think about it and check a few websites that offer this kind of services. Now days online shopping is the main selling channel for almost every industry. You might be supersized to find out the even Domins order their special pizza boxes to printing companies online.

Online printing orders has very tight deadlines and and very low margin profits for the printing companies. Thus, efficient use of equipment and material is crucial for printing companies. This is where Operations Research can help.

In collaborative with an online printing company, we designed a fast, flexible and reliable optimization algorithm. Our algorithm decide on the printing plate layout, number of copies necessary from each plate and the printing date of the plate. A simple implementation of the algorithm in Python can be found in the following GitHub repository (github.com/mahdims/2D-Bin-packing).