Phasor Agents are dynamical systems where the internal state is a Phasor Graph, a weighted graph of coupled Stuart-Landau oscillators. They learn weights via three-factor local plasticity using eligibility traces gated by sparse global modulators and oscillation-timed write windows, without backpropagation. To address stability challenges, they separate wake tagging from offline consolidation, introducing deep-sleep-like gated capture and REM-like replay. Experiments demonstrate that this mechanism enhances learning stability, improves maze success rates, and generates latent learning signatures consistent with an internal model.

This paper introduces 'basic inequalities' for first-order iterative optimization algorithms, presenting a simple and versatile framework that connects implicit and explicit regularization. The framework translates the number of iterations into an effective regularization coefficient in the loss function. It is applied to analyze training dynamics and prediction risk bounds for various optimization methods, including gradient descent, mirror descent, and generalized linear models, with theoretical findings supplemented by experiments.

This study investigates the effects of T-type and L-type calcium currents on synchronized activity patterns in basal ganglia networks, relevant to Parkinson's disease (PD). Simulations using an STN model show that stronger T-type currents enhance rebound bursting and expand synchronized activity, while stronger L-type currents prolong STN bursts, suggesting a role in PD pathophysiology. The findings elucidate the interplay between intrinsic cellular properties, synaptic parameters, and external inputs in shaping pathological synchronized rhythms.