Software
- Balans — Online Learning Meta-Solver for Decision Making
Best paper nominee at IJCAI. Contributed to the COIN-OR Open-Source Foundation. - Forge — Foundational Optimization Representations
The first foundational pre-trained model for mixed-integer programs. Generalizes across problems, sizes, and tasks. - Mab2Rec — Bandit-based Recommender Systems
Downloaded more than 3M+ times in the AI community. Deployed by enterprises worldwide. - Text2Model — LLM Modeling Copilots
Streamline LLM copilot development, evaluation, and deployment.
All Software
LLMs, Agents & Co-Pilots
- Text2Model: LLM Modeling Copilots
- Learn2Zinc: Fine-Tuning Small Language Models
- Gala: Global LLM Agents
- Text2Zinc: A Cross-Domain Dataset for Modeling Optimization and Satisfaction Problems in MiniZinc
- iCBS: iterative Combinatorial Brain Surgeon
Responsible AI
- BoolXAI: Explainable AI using expressive Boolean formulas
- Jurity: Fairness and bias mitigation, and recommender system evaluation techniques
Personalization & Recommender Systems
- Mab2Rec End-to-end framework for building bandits-based recommenders from raw interaction data to feature generation, model evaluation and deployment
- MABWiser Parallelizable, contextual multi-armed bandits library with simulation capabilities
Natural Language Processing
- Ner4Opt: Named-entity extraction for optimization
- TextWiser Text featurization NLP library that unifies several text embedding techniques
Pattern Mining
- Seq2Pat Sequence-to-Pattern generation library to find frequent patterns from large sequence databases. Joint work with CMU
Feature Selection
- Selective A white-box feature selection library that supports unsupervised and supervised selection methods for classification and regression tasks
- FSiHD Feature Selection in High-Dimensional Classification. The bronze winner at the Interspeech Speaker Trait Challenge in 2012. Computer Speech & Language’15, Interspeech’12
Combinatorial Optimization & Satisfaction
- Balans: Bandit-based Adaptive Large-Neighborhood Search Meta solver for mixed-integer programming problems using online learning
- Explorateur: State-Space-Search
- PathFinder: CP Decomposition for Capacitated Vehicle Routing with Time-Windows (CVRP-TW) This solver was a finalist of the 2018 AIMMS/MOPTA Modelling Competition Constraint programming based decomposition approach to solve a realistic variant of the vehicle routing problem.
- 3S: Satisfiability Solver Selector. This solver won 2 gold and 7 medals in total at the International SAT Solver Competition in 2011. CP’11, SAT’11
- Hegel: Dialectic Search Solver for Set Cover Problem Dialectic search based local search approach for solving Set Covering Problems. CP’09
Algorithm Configuration & Portfolios
- ISAC: Instance-Specific Algorithm Configuration. Algorithm Portfolio based on the Instance-Specific Algorithm Configuration (ISAC) approach. ECAI’10
- GGA: Gender-based Genetic Algorithm for Parameter Tuning. A state-of-the-art parameter tuning and algorithm configuration tool based on genetic algorithms. It is the backbone of many competition-winning constraint solvers and algorithm portfolios. Re-implemented and now maintained by Optano GmbH. CP’09
- DASH: Dynamic Adaptive Search Heuristics & Non-Model Based Search Guidance Set Partitioning Problem (SPP) Benchmark Instance Generator. EJOR’16, AAAI’12
- ASLib: Algorithm Selection Library. This library consists of scenarios and instances for benchmarking from different domains, such as SAT, CSP, and ASP.