During the last few weeks of class, students will present (in groups of 2-3) recent research papers related to the course topic. Presentations should give an overview of the problem setting, context within the related literature, model, proofs (or proof sketches) of the main results, and experimental results. Presentations can use slides and the whiteboard.

Each group will be assigned one class period to present the paper, and students not presenting are expected to attend class prepared to engage and ask questions. Groups should schedule a meeting with the course instructor at least one week before their assigned presentation slot with a draft of their presentation to obtain feedback. The presentation grade will be determined as follows:

• 25%: First draft of the presentation, presented to the course instructor during a meeting at least one week before the assigned presentation slot.
• 75%: In-class presentation.

Potential topics for paper presentations are listed below. You’re welcome to present a paper not on this list if it’s related to the course topic; just email the instructor to check.

• Different prediction models for algorithms with predictions
  • Online Algorithms with Multiple Predictions
  • Algorithms with Prediction Portfolios
  • Augmenting Online Algorithms with ε-Accurate Predictions
  • MAC Advice for Facility Location Mechanism Design
  • Binary Search with Distributional Predictions
  • Parsimonious Learning-Augmented Caching
• Learning-augmented algorithms in different problem settings
  • Mechanism Design with Predictions
  • Clock Auctions Augmented with Unreliable Advice
  • Learning-Based Frequency Estimation Algorithms
  • Learning-Augmented Streaming Algorithms for Approximating MAX-CUT
  • The Case for Learned Index Structures
  • A Model for Learned Bloom Filters, and Optimizing by Sandwiching
  • Learning Augmented Binary Search Trees
  • Learning-Augmented Search Data Structures
• Learning for optimization
  • Learning to Branch
  • Learning to Warm-Start Fixed-Point Optimization Algorithms
  • Learning-augmented smooth integer programs with PAC-learnable oracles
  • Unrolled Neural Networks for Constrained Optimization
  • Optimizing Solution-Samplers for Combinatorial Problems: The Landscape of Policy-Gradient Methods

Check back soon for a more detailed grading rubric and additional paper suggestions.