---

Neural Combinatorial and Constrained Optimization

Project Overview

We invite motivated undergraduate, MS, and Ph.D. researchers to join a project on Neural Combinatorial and Constrained Optimization, focusing on learning-based approaches for discrete and NP-hard problems.

The project aims to build neural solvers that learn search strategies, constraint handling, and decision heuristics, complementing or surpassing classical optimization techniques in scalability and generalization.

---

Research Focus Areas

1. Neural Architectures for Structured Optimization

• Developing scalable Graph Neural Networks (GNNs) and attention mechanisms for combinatorial reasoning.
• Designing architectures that generalize to large-scale instances and varying graph topologies.
• Benchmarking neural solvers against classical heuristics on routing, scheduling, and coverage tasks.

2. Constrained Learning & Hybrid Solvers

• Integrating hard constraints via differentiable optimization layers and implicit differentiation.
• Creating hybrid algorithms that combine deep learning with classical methods (e.g., Branch-and-Bound).
• Learning to satisfy complex logic and discrete constraints while optimizing objectives.

---

What You’ll Work On

• Design neural architectures for optimization problems
• Develop training algorithms that respect problem constraints
• Benchmark against classical solvers and analyze trade-offs
• Apply methods to real-world optimization problems (scheduling, routing, planning)
• Build scalable implementations that handle large problem instances

---

What We’re Looking For

Essential:

• Strong background in algorithms and optimization
• Proficiency in Python and deep learning frameworks (PyTorch, Jax)
• Understanding of graph algorithms, dynamic programming, and complexity theory
• Experience with reproducible research practices
• Experience with classical optimization (integer programming, constraint programming)

Nice-to-have:

• Familiarity with graph neural networks or attention mechanisms
• Background in operations research or algorithmic game theory
• Publications in ML/AI or optimization venues

---

What You’ll Gain

• Deep expertise in neural optimization algorithms
• Experience with state-of-the-art deep learning and optimization techniques
• Understanding of both theoretical and practical aspects of optimization
• Opportunities to publish at top-tier venues (NeurIPS, ICML, ICLR, AAAI, etc.)

---

How To Apply

Please submit your details using the Google Form.

Note: Choose “Neural Combinatorial and Constrained Optimization” for the question “Which of the following match your research interests?” in the google form.

Selected students may be invited for a brief meeting to discuss fit and potential directions.

---