Postdoctoral Research Associate, Adaptive Learning Agents and Computational Cognitive Science

The NAIADEL Lab in the University of Arizona Department of Psychology seeks a Postdoctoral Research Associate with expertise in developing and analyzing adaptive learning agents to investigate the computational mechanisms of human intelligence and agency. The lab builds a neurocomputational theory of adaptive intelligence and agency. We study how people and artificial agents infer the structure of their environment, judge what they can control, revise those judgments after setbacks, and change their strategies when conditions change.

Computational cognitive models and modern deep learning are our main instruments for this work. We treat deep reinforcement learning agents and sequence models as scientific models of human behavior. We train them, fit them to human data, interpret what they have learned, and compare them with process-oriented cognitive models such as sequential sampling models and reinforcement learning models. These approaches offer complementary accounts of adaptation. Meta-learning agents can model how learning strategies change with experience. Learned world models can represent environmental dynamics and support planning. Sequential sampling models characterize the dynamics of individual decisions. We ask when these accounts explain the same behavior, where they diverge, and how they can be integrated into a theory of adaptive intelligence.

The candidate will lead the development of new adaptive agents for the lab's game-based environment, in which task structure and controllability change across game missions and over time. They will train these agents on the same games that people play, compare agent and human behavior on data from more than 1,000 participants across online, in-person, and clinically characterized samples, and use the results to test and refine the lab's theory of agency. The work will generate first-author papers and an agent-based research line the candidate can carry into an independent program.

Recorded human data and a Gymnasium-compatible runtime for the lab's games allow modeling to begin in the first weeks. Work with engineers and game designers allows new agents and new environments to be tested quickly. Resources include HPC clusters, GPU/CPU nodes, cloud compute, EEG, fMRI, and eye-tracking.

The lab is building a shared computational core of models, environments, and training pipelines on which its research runs, and the postdoc will substantially shape its design. They will join an active community of computational cognitive modelers and computer scientists across campus.

International collaborations connect the candidate to machine learning, neuroscience, and clinical partners and to a global scientific network. The position offers close mentorship, funded international conference travel, dedicated compute allocations, fellowship support, and planning for independence, authorship, and project portability.

The initial appointment is for one year, renewable based on satisfactory progress. The start date is negotiable; applications are reviewed on a rolling basis.

Outstanding U of A benefits include health, dental, vision, and life insurance; paid vacation, sick leave, and holidays; U of A/ASU/NAU tuition reduction for the employee and qualified family members; access to U of A recreation and cultural activities; and more!

The University of Arizona has been recognized for our innovative work-life programs. For more information about working at the University of Arizona and relocation services, please click here.

• Design, implement, and train new reinforcement learning agents for game-based environments in which task structure and controllability change. Relevant approaches include meta-reinforcement learning, continual learning, model-based reinforcement learning and world models, and transformer-based sequence models for decision-making.
• Treat trained agents as scientific models of behavior. Fit and compare them with human data from game-based paradigms examining meta-control, cognitive flexibility, and social learning, interpret the representations and strategies they acquire, and relate them to process-oriented cognitive models such as reinforcement learning and sequential sampling models.
• Develop methods that connect agent behavior with human behavior at the level of mechanism. Use simulation, recovery analyses, and out-of-sample prediction to determine which mechanisms are identifiable and supported by the data.
• Formulate decisive comparisons among competing accounts of adaptation and help design or refine games and environments within the lab's game-based adaptive platform.
• Lead first-author manuscripts aimed at machine learning venues and at cognitive science journals, and contribute to collaborative papers with partner sites.
• Take scientific responsibility for the lab's agent research line, including theoretical framing, modeling decisions, timelines, code quality, and publication.
• Work with the lab's engineering team, clinical partners, and neuroscience collaborators to connect agents and cognitive models to adaptive tasks and clinical measures, and to extend this work to EEG and eye tracking alongside collaborators with expertise in these methods.
• Build and maintain reproducible, documented training and analysis pipelines under version control, and release code with publications.
• Contribute to grant applications and prepare independent fellowship or career-development proposals with support from the Principal Investigator.
• Co-mentor graduate and undergraduate researchers on aligned sub-projects, present findings at major national and international conferences, and contribute to shaping the intellectual direction of the lab.

Knowledge, Skills, and Abilities

• Ability to design, implement, and interrogate learning agents, and to reason about why an agent behaves as it does.
• Knowledge of deep reinforcement learning, meta-learning or in-context learning, model-based reinforcement learning, or sequence models for decision-making, with the ability to apply these methods to understanding human adaptive behavior.
• Ability to connect computational approaches to questions about human cognition and adaptive behavior.
• Ability to reason carefully about what a trained model has learned, about alternative explanations, and about what a dataset can and cannot establish.
• Strong programming skills in Python and a modern deep learning framework such as PyTorch or JAX, including the ability to develop and maintain original research code.
• Knowledge of model validation methods and reproducible research practices.
• Ability to set priorities and take ownership of a project from initial formulation through publication.
• Skill in scientific writing and the ability to effectively communicate rigorous research findings.
• Ability to collaborate across computational, experimental, engineering, and clinical perspectives, and to give and receive direct feedback.
• Ability to explain technical choices to colleagues from other backgrounds, and willingness to build shared tools and documentation that the lab can continue to use.
• Ability to demonstrate intellectual curiosity, reliability, and a collaborative approach to shared research, with a commitment to developing an independent research identity.

• PhD in computer science, machine learning, computational neuroscience, cognitive science, psychology, statistics, applied mathematics, or a closely related field. Candidates who will defend before the start date may apply.
• Demonstrated experience designing, implementing, training, and evaluating learning agents or neural-network models for sequential decision-making, supported by publications, preprints, or research code.
• Demonstrated programming skills in Python and at least one modern deep learning framework.
• A record of peer-reviewed publication appropriate to career stage.

• Experience developing new learning algorithms or substantially extending existing ones.
• Experience with meta-reinforcement learning, in-context learning, or continual reinforcement learning in non-stationary environments.
• Experience with POMDPs, belief-state inference, model-based reinforcement learning, or learned world models for decision-making under partial observability.
• Experience with transformers or other sequence models for sequential decision-making.
• Experience evaluating LLM-based agents as subjects in behavioral tasks, or developing LLM-based agents with memory, planning, or tool-use capabilities.
• Experience using neural networks as models of human or animal behavior, or interpreting the representations and strategies of trained agents.
• Experience fitting reinforcement learning or sequential sampling models to human behavior. Computational cognitive scientists who want to extend their work to developing adaptive agents are encouraged to apply.
• Experience modeling controllability, latent-state inference, exploration under uncertainty, or adaptive control.
• Experience analyzing human or animal behavioral data, or comparing model predictions with empirical data.
• Experience designing environments, tasks, or benchmarks for learning agents.
• Experience with collaborative software development, version control, and reproducible research workflows.

Please submit your CV and Cover Letter, along with one Additional Document containing the following materials:

1. Statement of Research Interest: Identify one question about adaptive intelligence or agency that you would like to pursue in this lab, and describe how you would approach it with the methods you bring.
2. Supporting Materials: One or two representative publications, preprints, or code repositories, including at least one that shows your own algorithmic work.
3. References: Names and contact details of three referees.

Please combine the Statement of Research Interest, Supporting Materials, and References into one file and upload it in the Additional Documents section of your application.