My research interests lie at the intersection of two kinds of cognitive control: the control of memory representations and the control of choices and actions. This intersection is especially interesting in the face of uncertainty (e.g. about the probability or magnitude of rewards, or the trajectories of actions that lead to reward) and unreliability of statistical structures in the world (e.g. the probabilities of encountering certain states, sequences, or associated rewards). I use behavioral experiments, neuroimaging, and computational modeling to study the evoked and strategic interaction between working memory (WM) and episodic memory (EM) representations in prospection, planning, and decision-making. My main areas of research are as follows.
a. Remembering the future: Prospective representations of goals and upcoming tasks
The realization of planned action requires us to engage in performing ongoing tasks while holding on to representations of the future tasks we plan to do. How do our brains encode and maintain representations of planned tasks to control our future actions? Does the cognitive load during the retention period influence how the brain represents planned tasks? I use fMRI and MVPA to address these questions by studying the representations that control:
(I) future task-sets and intended time of executing planned action (the 'what' and 'when' of prospective intentions) in time-based prospective memory (Momennejad and Haynes, 2012)
(II) future task-sets under varieties of working memory load in event-based prospective memory (Momennejad and Haynes, 2013),
(III) the dynamic representation of the identity and order of planned task-sequences (Momennejad, Reverberi, Haynes, in prep),
(IV) the representation of task-reward associations (Wisniewsky et al, 2016)
(V) the role of episodic future simulation of a goal (e.g. in prospective memory) on the success of goal-directed behavior (Momennejad, Norman, Cohen, in prep).
b. Learning to control: The strategic recruitment of working and episodic memory in cognitive control
Have a plan? Your ability to successfully realize it depends on parameters outside your control. The nature of task demands, the perceptual properties of the world, and individual differences in memory capacity limits create resource allocation problems for the execution of planned action. In the face of such parameters, the brain strategically recruits episodic and working memory strategies/representations to successfully realize planned action, e.g. in prospective memory. For instance, if your working memory is limited or ongoing tasks exert high load, you could increase your episodic memory strategy. We propose mechanistic and normative accounts of WM-EM interactions in prospective memory in two modeling projects:
(I) deep neural networks (Momennejad*, Tomov*, Norman, Cohen, in prep, see poster PDF file) and
(II) normative models (Momennejad, Norman, Cohen, Lewis-Peacock, Singh, Lewis) that simulate human behavior in prospective memory tasks.
In these models we simulate behavior by optimizing memory parameters under different conditions. Future directions of these models will focus on learning algorithms with which an agent learns to optimize WM-EM representations for adaptive control of prospective behavior.
c. Changing the past: Replay and updating predictive representations guide adaptive behavior under uncertainty
(I) Using behavioral experiments and reinforcement learning models, I study predictive representations involved in adaptive behavior, e.g. the role of successor representations in varieties of goal-directed decision-making in the face of devaluation and uncertainty (Momennejad*, Russek* et al, under review, preprint available ).
(II) Using reinforcement learning models, we discuss how predictive representations offer algorithms that bridge model-free and model-based mechanisms in sequential decision problems (Russek*, Momennejad* et al, under review, preprint available).
(III) Using MVPA, reinforcement learning, and model-based fMRI analysis, I study the role of and replay and predictive representations in goal-directed behavior such as retrospective revaluation (Momennejad, Otto, Rhee, Daw, Norman, in prep).
d. Collective dynamics: Mnemonic convergence and emergent behavior in multi-agent networks
Understanding the cognitive control of memories and actions is not complete without understanding how collective phenomena interact with control at the individual level. I've worked with Alin Coman and Stacey Sinclair on innovative projects that explore the emergence and dynamics of collective memories and collective behavior. To this end, we have used multi-person behavioral experiments and graph analysis on mnemonic convergence (Coman, Momennejad et al, 2016; Momennejad, Duker, Coman, in prep), and multi-agent simulations of emergent patterns of actions outside of individual intentions, e.g. marginalization (Momennejad, Piloto, Sinclair, in prep).