Affective State Estimator (Valence-Arousal + Appraisal + Modulation)

Map internal cognitive signals to a Valence-Arousal-Dominance space by processing reward prediction error, epistemic uncertainty, novelty, and homeostatic deviation to produce a 3D affective representation for mood-aware AI agents.

Affective state estimation is used in reinforcement learning to drive adaptive behavior by modulating learning rate, attention, exploration strategy, memory prioritization, and risk sensitivity based on computed valence, arousal, and dominance.

Integrate appraisal variables like relevance, congruence, coping potential, and norm compatibility to derive mood states and guide modulation targets, ensuring mood-congruent bias for adaptive behavior in cognitive architectures.

Yes, the Affective State Estimator requires PyTorch as its sole dependency to support both analytic and optional learned mappings for converting cognitive signals into bounded PAD affective representations.

Yes, you can use the deterministic, bounded modulation vectors outputted by the estimator to directly control exploration strategy tuning, risk sensitivity, and learning rate adjustments across various AI environments.

Provide reward_prediction_error, epistemic_uncertainty, prediction_error_magnitude, novelty_score, and homeostatic_deviation as inputs to generate valence, arousal, dominance, appraisal results, mood, and modulation vectors.