Many games (such as Minesweeper, Sudoku, Nonograms, and Rubik’s Cube) require players to engage in logical and combinatorial reasoning. While solving these games from a modern AI perspective is often straightforward, generating interesting levels/puzzles can be surprisingly challenging. As a result, logic and programming games provide an interesting domain for experimenting with AI algorithms. Sample topics: creating level generators for Minesweeper that avoid guesswork; designing a Nonogram generator for given images; implementing an algorithm to solve programming puzzles.

This research-oriented work aims to verify whether Reinforcement Learning methods can be used in place of classical discrete optimization methods in a given optimization problem.

The aim of the work is to implement a system that generates logic programs based on a given objective function and specification of available predicates. The system should use an existing implementation of genetic programming algorithms.

In operations research, discrete optimization methods are often used, i.e. algorithms that find a solution that maximizes/minimizes a given objective function. In many situations, these methods can be difficult to apply, e.g. in environments where we have no control over the involved agents (when planning a street grid, we have no control over drivers). Alternative methods come from game theory and explicitly take into account the agents’ preferences, i.e. they find solutions in which none of the agents would behave differently than expected by the algorithm — such solutions are called stable. The aim of the work would be to solve an operations research problem that involves interactions between agents using both approaches and compare the results.

The aim of the work is to implement an algorithm that will find the minimal formal grammar (e.g. regular language) that describes a given set of sentences. Alternatively, this problem can be understood as an attempt to uncover the structure of a finite automaton based on the results it outputs.

One of the most difficult elements of a programmer’s work is requirements analysis - i.e. understanding the client’s needs. The aim of this work is to create a tool that will create a constraint programming model based on known solutions to a given problem.

Routing problems are an important area of operations research, varying in many constraints such as: time windows, vehicle capacity constraints, contractual requirements, mandatory work breaks, maximum time for storing goods in the vehicle, etc. The aim of the work would be to create an application that would allow “clicking” on the appropriate requirements and generating code/model that would solve the given variant of the routing problem.

This research-oriented work aims to reproduce the local search algorithm for integer optimization problems proposed in 2023.

The aim of the work is to implement a tool that solves a chosen discrete optimization problem. The work can be enriched with a comparison of several methods and/or the creation of a graphical user interface. Sample problems: arranging a class schedule with elective subjects, scheduling nurses’ work, planning a street grid.

MiniZinc is a popular language for modeling discrete optimization problems and is often used in the prototyping phase of solutions. The problem is its deployment in a production environment and often requires manual translation into the API of a specific solver. The aim of the work would be to automate this process, i.e. to create a tool that compiles the MiniZinc model into a program written using an existing popular solver. The work can be divided into two stages: frontend - parsing the MiniZinc model and backend - code generation.