Voluntary behavior of humans appears to be composed small, elementary building blocks, or behavioral primitives. While this modular organization seems crucial for the learning complex motor skills and flexible adaption new circumstances, problem meaningful, compositional abstractions from sensorimotor experiences remains an open challenge. Here, we introduce a computational architecture, termed as surprise-based modularization into event-predictive structures (SUBMODES) that explores...

Abstract From about 7 months of age onward, infants start to reliably fixate the goal an observed action, such as a grasp, before action is complete. The available research has identified variety factors that influence goal‐anticipatory gaze shifts, including experience with shown events and familiarity agents. However, underlying cognitive processes are still heavily debated. We propose our minds (i) tend structure sensorimotor dynamics into probabilistic, generative event‐predictive, event...

Flexible, goal-directed behavior is a fundamental aspect of human life. Based on the free energy minimization principle, theory active inference formalizes generation such from computational neuroscience perspective. theory, we introduce an output-probabilistic, temporally predictive, modular artificial neural network architecture, which processes sensorimotor information, infers behavior-relevant aspects its world, and invokes highly flexible, behavior. We show that our trained end-to-end...

Humans can make predictions on various time scales and hierarchical levels. Thereby, the learning of event encodings seems to play a crucial role. In this work we model development via autonomously learned latent codes. We present recurrent neural network architecture, whose inductive biases foster sparsely changing state that compress sensorimotor sequences. A higher level learns predict situations in which states tend change. Using simulated robotic manipulator, demonstrate system (i)...

During the observation of goal-directed actions, infants usually predict goal at an earlier age when agent is familiar (e.g., human hand) compared to unfamiliar mechanical claw). These findings implicate a crucial role developing agentive self for infants’ processing others’ action goals. Recent theoretical accounts suggest that predictive gaze behavior relies on interplay between experience (top-down processes) and perceptual information about action-event (bottom-up information; e.g.,...

Mental representations in infants are sparse and grow richer over development. Anticipatory eye fixation studies show that at around 7 months start to predict the goal of an observed action, e.g., object targeted by a reaching hand. Interestingly, goal-predictive gaze shifts occur earlier age when (animate) hand subsequently manipulates later action is performed inanimate actor, mechanical claw. We introduce CAPRI² (Cognitive Action PRediction Inference Infants), computational model explains...

Mental representations of the environment in infants are sparse and grow richer during their development. Anticipatory eye fixation studies show that aged around 7 months start to predict goal an observed action, e.g., object targeted by a reaching hand. Interestingly, goal-predictive gaze shifts occur at earlier age when hand subsequently manipulates later action is performed inanimate actor, mechanical claw. We introduce CAPRI2 (Cognitive Action PRediction Inference Infants), computational...

A common approach to prediction and planning in partially observable domains is use recurrent neural networks (RNNs), which ideally develop maintain a latent memory about hidden, task-relevant factors. We hypothesize that many of these hidden factors the physical world are constant over time, changing only sparsely. To study this hypothesis, we propose Gated $L_0$ Regularized Dynamics (GateL0RD), novel architecture incorporates inductive bias stable, sparsely states. The implemented by means...

From about 7 months of age onwards, infants start to reliably fixate the goal an observed action, such as a grasp, before action is complete. The available research has identified variety factors that influence goal-anticipatory gaze shifts, including experience with shown events and familiarity agents. However, underlying cognitive processes are still heavily debated. We propose our minds (i) tend structure sensorimotor dynamics into probabilistic, generative event-predictive...

A critical challenge for any intelligent system is to infer structure from continuous data streams. Theories of event-predictive cognition suggest that the brain segments sensorimotor information into compact event encodings, which are used anticipate and interpret environmental dynamics. Here, we introduce a SUrprise-GAted Recurrent neural network (SUGAR) using novel form counterfactual regularization. We test model on hierarchical sequence prediction task, where sequences generated by...

Voluntary behavior of humans appears to be composed small, elementary building blocks or behavioral primitives. While this modular organization seems crucial for the learning complex motor skills and flexible adaption new circumstances, problem meaningful, compositional abstractions from sensorimotor experiences remains an open challenge. Here, we introduce a computational architecture, termed surprise-based modularization into event-predictive structures (SUBMODES), that explores identifies...

Humans can make predictions on various time scales and hierarchical levels. Thereby, the learning of event encodings seems to play a crucial role. In this work we model development via autonomously learned latent codes. We present recurrent neural network architecture, whose inductive biases foster sparsely changing state that compress sensorimotor sequences. A higher level learns predict situations in which states tend change. Using simulated robotic manipulator, demonstrate system (i)...

During the observation of goal-directed actions, infants usually predict goal when action and agent are familiar, but they do not as easily or unfamiliar. Recent theoretical accounts suggest that predictive gaze behavior relies on a complex interplay between bottom-up- (e.g., agency cues) top- down information prior experience with action), depending an observer’ knowledge about unfolding event. Based these accounts, we hypothesized during grasping actions performed by mechanical claw,...