Reading Group on Cognitive Abilities in Artificial Systems

ACT-R account of how practice compiles declarative and procedural steps into efficient skills, linking cognitive theory to skill-acquisition dynamics.

SOAR-based robotic control study showing how symbolic cognitive architectures can support perception-action loops in embodied agents.

SPAUN-style spiking-neuron cognitive modeling work that grounds symbolic task-solving behavior in biologically constrained neural dynamics.

Classic Adaptive Resonance Theory paper on stable yet plastic category learning under continual input.

Active predictive-coding framework for learning reference frames and compositional part-whole structure from interaction.

AlphaZero milestone showing how general self-play reinforcement learning can achieve superhuman planning performance without domain heuristics.

DORA-inspired analogy work modeling how relational structure and shape generalization can emerge in development.

Recent robotics foundation-model approach integrating vision, language, and action for broad policy generalization across robotic tasks.

Analyzes in-context learning as implicit algorithm learning in transformers, clarifying which function families can be learned from prompts alone.

Foundational associative-memory formulation showing attractor dynamics as a computational principle for content-addressable memory.

Natural-RL framing that emphasizes one-shot lifetime constraints, pushing RL toward settings closer to real-world agent experience.

Free-energy-principle tutorial and comparison paper clarifying active inference assumptions relative to mainstream learning and control frameworks.