Ali Teshnizi

Ali Teshnizi

PhD student in MS&E

Hello!

    I’m a PhD student in Management Science & Engineering at Stanford, working with Prof. Madeleine Udell. Previously, I completed a master’s degree in Computer Science at University of Toronto and did my undergrad in Computer Engineering at Sharif University of Technology.

My Research Interests

    I use Machine Learning to tackle Constrained Optimization problems. Currently, I’m working on methods that leverage Graph Neural Networks (GNNs) and Reinforcement Learning (RL) to solve Constrained Optimization problems on graphs.

That was a bunch of fancy (?) terms. What does it mean exactly?

    Graph-structured data is abundant around us, and we’ve been modeling and solving real-world problems with graphs for decades. We’ve already developed optimal and fast solutions to a subset of these problems. However, some others are too hard, complicated, or time-consuming to solve using explicit mathematical and algorithmic methods (NP Problems are a good example).

    Neural networks are good at extracting complicated patterns from high-dimensional data. Instead of trying to solve these intricate graph problems explicitly, we can use GNNs to embed graphical information into numerical representations, feed those to an RL model, and tune the model to generate a solution by maximizing/minimizing an objective of interest.

Graph Learning and Reinforcement Learning for Constrained Combinatorial Optimization

It sounds cool! But is it actually useful?

    This approach is scalable and can result in close-to-optimal results. It, however, is not fully interpretable and lacks optimality guarantees. These characteristics make it a perfect fit for a subgroup of graph optimization problems, but not for all of them. Proper modeling of the input information, expressive architecture design, and problem-specific constraint enforcement are some of the challenges. Making all of the modules work together can also be tricky.

    Here is a list of some domains where this approach can be useful:

  • Supply Chain Management
  • Grid Optimization
  • Social Media Marketing
  • Circuit Design
  • Network Optimization

I want to maximize/minimize something on a graph. Can we talk?

    Of course! Feel free to shoot me an email at teshnizi@stanford.edu. I’d also love to chat about anything else that you find related and interesting!