Online decoding of rat self-paced locomotion speed from EEG using recurrent neural networks

In recent years, the boundaries between artificial intelligence (AI) and neuroscience have become increasingly blurred. Researchers are now leveraging insights from both fields to crack the cognitive code, making significant strides in understanding brain function and behavior. A series of groundbreaking studies has demonstrated the power of modularity, neural networks, and biomimicry in deciphering the complexities of the human mind.

One such study, published on arXiv, has successfully decoded rat self-paced locomotion speed from EEG recordings using recurrent neural networks (RNNs) (Source 1). This achievement holds promise for advancements in rehabilitation, prosthetic control, and our understanding of neural correlates of action. The researchers' use of non-invasive, cortex-wide EEG recordings and RNNs marks a significant departure from traditional methods, which often rely on intracranial implants and motorized treadmills.

The importance of modularity in both natural and artificial intelligence is another area of focus. A separate study highlights the critical role of modularity in supporting efficient learning and strong generalization abilities, characteristics consistently exhibited by humans (Source 2). The authors argue that modularity aligns with the No Free Lunch Theorem, which emphasizes the need for problem-specific inductive biases and motivates architectures composed of specialized components that solve subproblems.

A Ginzburg-Landau theory of cognitive dynamics has also been proposed, which models cognition as a coarse-grained neural activity field governed by a variational free energy (Source 3). This framework predicts a universal algebraic divergence of the susceptibility, providing a theoretical rationale for the observed avalanche size exponent in cortical dynamics. The study identifies adult cognition as a metabolically pinned non-equilibrium steady state maintained near the critical point.

Furthermore, a functionalist approach to emotion has been explored through a biomimetic reinforcement learning framework (Source 4). The researchers mathematically construct a theoretical framework grounded in functionalist principles, examining how the resulting utility function aligns with emotional valence in biological systems. This framework is applied to psychological phenomena such as humor, psychopathy, and advertising, demonstrating its breadth of explanatory power.

Lastly, a novel approach to discovering neural mechanisms of cognitive errors has been developed, which automates the discovery of viable RNN mechanisms by explicitly training RNNs to reproduce behavior, including characteristic errors and suboptimalities (Source 5). This method uses a non-parametric generative model of behavioral responses to produce surrogate data for training RNNs, capturing all relevant statistical aspects of the data.

As these studies demonstrate, the convergence of AI and neuroscience is yielding significant breakthroughs in our understanding of cognitive dynamics, emotional modeling, and neural mechanisms of behavior. By harnessing the power of modularity, neural networks, and biomimicry, researchers are making rapid progress in cracking the cognitive code, with far-reaching implications for fields such as rehabilitation, prosthetic control, and artificial intelligence.

References:

• Source 1: Online decoding of rat self-paced locomotion speed from EEG using recurrent neural networks
• Source 2: Modularity is the Bedrock of Natural and Artificial Intelligence
• Source 3: Critical Scaling and Metabolic Regulation in a Ginzburg-Landau Theory of Cognitive Dynamics
• Source 4: Functional Emotion Modeling in Biomimetic Reinforcement Learning
• Source 5: Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors