Tag
This paper proposes a unified dynamical framework for training, learning, and inference in neural systems.
This paper presents a computational framework for steering representational geometry to improve bidirectional alignment between biological and artificial neural networks, showing a 55% relative enhancement in bidirectional predictivity.
This paper studies how an agent with limited perceptual bandwidth should allocate interoceptive precision across bodily needs in a foraging task, showing that dynamically attending to the most-needed channel improves survival under a fixed precision budget.
This paper investigates the concept of 'ontological inversion' under the free energy principle, using a convolutional variational autoencoder to explore whether a synthetic environment can permanently replace a system's default generative model. The study finds a decoupling between representational accuracy and default behavior, introducing the phenomenon of 'cognitive relapse' where the system partially reverts to its original model.
This paper proposes replacing the standard point neuron model in artificial neural networks with a more realistic cortical cell model, claiming improvements in expressivity, robustness, learning speed, and reduced memorization and data requirements.