Memory Dynamics and Decision Making

Humans and animals can produce various behavioural outcomes, even under the same conditions. Such variability may be due to small perturbations in the environment or the brain’s own activities. Neuroscience data has shown that some regions of the brain may show chaotic activities. Thus, deterministic chaos could be a source of variability sufficient to extract various behaviours in associative recall without any external noise. Here, we study this idea in a minimal, modular architecture in which a chaotic neuron (CN) generates probe signals that are mapped into the input space of a recurrent associative memory (RAM), which iterates until convergence to an attractor is achieved. Using correlated binary patterns, we compared CN-driven probes to a matched independent and identically distributed (i.i.d.) random baseline while varying the control parameters. Under controlled RAM initialization, CN-driven probing yields occupancy distributions that are indistinguishable from those obtained with matched random probes. In addition, outcome distributions can be shaped by memory-side constraints and become stronger when RAM weights are initialized randomly near zero, thus representing inter-subject variability. Overall, these results clarify when deterministic chaotic probing behaves like matched random probing and show that the observed retrieval distribution is dominated by the RAM attractor landscape and its initialization conditions.

This is an in-person presentation on July 18, 2026 (10:40 ~ 11:00 EDT).

People reason causally in two directions: predicting effects from causes and diagnosing causes from effects. Causal direction leads to asymmetric response patterns in plausibility judgements, response times, and accuracy. Yet little research has examined how these asymmetries emerge from basic cognitive processes during memory retrieval. We propose that directional asymmetries arise from divergent search strategies: diagnostic reasoning follows depth-first search, elaborating single causal chains while avoiding alternative causes, whereas predictive reasoning follows breadth-first search, readily exploring alternative effects. We present BLISA (Bidirectional causal reasoning using Lateral Inhibition in a Spreading Activation process), a computational cognitive model implementing these search strategies within the CLARION cognitive architecture. BLISA simulates individual-level choices and response times across causal structures varying in branching patterns. Simulations (N=24,000) reveal that diagnostic inferences take longer when multiple causes compete, while predictive inferences slow when multiple effects branch. Retrieval failures occur only during diagnoses with branching causes, consistent with human findings of reduced diagnostic certainty. BLISA builds on well-established memory processes that have successfully modelled predictive and diagnostic reasoning each in isolation, and unifies them under a single spreading-activation mechanism. These results demonstrate how asymmetrical causal reasoning emerges from basic retrieval mechanisms operating through distinct search patterns in spreading activation. We also present an experimental paradigm designed for human participants, and plan to validate our novel explanation for directional asymmetries in causal reasoning with behavioural data in future work.

This is an in-person presentation on July 18, 2026 (11:00 ~ 11:20 EDT).

Multi-alternative Decision Field Theory (MDFT) models preference construction as stochastic evidence accumulation with inter-option inhibition driven by similarity. Historically, MDFT has been applied to simple choice tasks where options are described by two attributes. Our goal is to extend MDFT to complex, real-world decisions where options are defined by higher-dimensional, context-rich cues. We present an MDFT extension that replaces the classical distance term with a multi-attribute distance D used to populate the similarity/inhibition matrix S while preserving its k×k form and the standard inhibition mechanism. The distance construction better reflects how higher-dimensional attributes shape option similarity under comparison, shifting predicted choice distributions and learning gradients. Building on neural learning approaches for MDFT (Venable; Rahgooy; Busemeyer), we implement the model as MDFT‑NN, which learns participant-specific attention weights and attribute preferences. A second contribution is selective attribute learning via column-wise gradient masking on the preference matrix M, enabling targeted learning of latent or emergent preference dimensions while holding other dimensions fixed to values derived from modeled experimental signals. We evaluate MDFT‑NN on a trial-level insider attack game dataset where participants repeatedly choose between Attack and Withdraw across trials with experimentally manipulated cues. We train participant-specific models across variants (standard learning, restricted attribute sets, selective preference learning) and validate learned parameters via classical MDFT forward simulation to predict trialwise Attack/Withdraw probabilities, comparing predictions to observed distributions across behavioral archetypes. Together, these advances enable preference modeling in rich contexts, separating overt context from latent subjective signals to personalize learned predictions.

This is an in-person presentation on July 18, 2026 (11:20 ~ 11:40 EDT).