ARIA Research Lab at NJIT

The ARIA Research Lab in the Department of Computer Science at the Ying Wu College of Computing, New Jersey Institute of Technology (NJIT), led by Dr. Shivvrat Arya, develops methods for trustworthy, structured, and efficient artificial intelligence, integrating learning, reasoning, and optimization to build AI systems that are interpretable, reliable, and scalable. Our research focuses on foundational advances in neuro-symbolic reasoning and probabilistic inference and neural combinatorial optimization, alongside applications in structured and multimodal intelligence and AI for scientific discovery.

Research Directions

• Neuro-Symbolic Reasoning and Probabilistic Inference We combine neural networks with symbolic structure, classical algorithms, and mathematical optimization to develop fast, reliable, and scalable methods for probabilistic inference and reasoning under uncertainty.

• Neural Combinatorial Optimization We design learning-based solvers—integrating deep reinforcement learning, representation learning, and classical optimization—to tackle large-scale discrete and graph-structured decision-making problems under complex constraints.

• Structured and Multimodal Intelligence We integrate explicit temporal, procedural, and relational structure with human feedback to build reliable, interpretable models for procedural video understanding and vision-language reasoning.

• AI for Scientific Discovery We develop machine learning methods that incorporate domain knowledge, biological interactions, and scientific constraints into structured generative models, with an emphasis on computational biology and single-cell genomics.

Joining the lab

We are always looking for curious, rigorous, and collaborative students who are excited about building the next generation of structured, explainable, and reliable AI systems.

• Ph.D. students: Opportunities to work on core problems in neuro-symbolic reasoning, probabilistic inference, neural combinatorial optimization, and structured deep learning.
• Undergraduate students: Opportunities to join active projects, learn how to conduct research, and build toward increasingly independent work through options such as CS 488 and CS 489.
• M.S. students: Opportunities for deeper research through faculty-mentored projects or the CS 700B/CS 701B thesis pathway, subject to advisor availability.

If you are interested in joining the ARIA Research Lab, please review the student research opportunities and follow the application instructions there. Briefly describe your background, relevant coursework or projects, and which of the lab’s research directions you are most excited about.

Selected projects & highlights

• NeuPI – Neural Probabilistic Inference
  A neural engine for probabilistic graphical models and circuits that distills complex queries into neural network approximators, accelerating query answering from minutes to microseconds for real-time decision-making.

• Neural Dual Bounds & Learning to Condition – Neural-Augmented Classical Solvers
  Learned heuristics and valid-by-construction dual bounds integrated directly into classical inference algorithms, accelerating solver convergence and shrinking search spaces while preserving theoretical guarantees.

• RELINK – Neural Combinatorial Optimization
  Deep reinforcement learning policies for sequential discrete decisions over complex networks, tackling large-scale network optimization and influence maximization under privacy constraints.

• CaptainCook4D – Procedural Activity Understanding
  A large-scale egocentric 4D dataset and benchmark for procedural task understanding, studying how AI systems recognize multi-step workflows, detect procedural errors, and assist users in complex environments.

• CoLa-VAE – AI for Scientific Discovery
  A cell-cell communication-aware variational autoencoder incorporating dynamic graph Laplacian constraints, advancing structured representation learning in single-cell genomics and computational biology.