As machine learning (ML) moves from purely theoretical models to practical applications, the issue of safety and fairness for ML begins to become increasingly relevant. ML has shown increasing proficiency in neuroscience, even performing better than humans in certain tasks. Together with the cornucopia of neuroscience data becoming available in the big data era, ML is destined to achieve even more superior efficacy in many tasks by leveraging it.

However, accompanying the capabilities, the inherent safety and fairness issues associated with ML have profound implications. ML models are algorithms trained on existing data, and as such, they often carry the biases of prior practice. Without vigilance and surveillance methods, even the most well-intentioned ML model can propagate the biases present in the training data. Moreover, vulnerabilities in ML models can be easily exploited by attackers advertently or inadvertently to achieve desired pernicious results or exacerbate power imbalances. Neuroscience explores how the brain executes various perceptual, cognitive, and motor tasks. ML-based artificial intelligence techniques allow big data to be processed with intelligent computers, which provides new possibilities for neuroscience such as how thousands of neurons and nodes communicate to handle massive information and how the brain generates behaviors and controls them. Researchers often overlook the risks of data-fying neuroscience because of the safety and fairness issue of ML models. Attention has been drawn to the much more profound and more fundamental problem that the safety and fairness issue in ML models exacerbates imbalances and risk, and the consequences are especially severe for neuroscience. Thus, ensuring safety and fairness in ML models for neuroscience becomes a crucial component in the advancement of ML in neuroscience.

This Special Issue welcomes original research and review articles discussing the fairness and safety implications of the use of ML in real-world neuroscience systems, proposing methods to detect, prevent, and/or alleviate undesired fairness and safety issues that ML-based systems might exhibit, analyzing the vulnerability of neuroscience ML systems to adversarial attacks and the possible defense mechanisms, and, more generally, any paper that stimulates progress on topics related to fair and safe ML in neuroscience. Hence, we hope to make sure that with the application of new technologies such as ML, across the continuum of care, technology aids humanity and not the reverse.

Potential topics include but are not limited to the following:

• Fairness and bias of ML in neuroscience
• Application of transparency to safety and fairness of ML in neuroscience
• Measurement & mismeasurement of safety and fairness
• Understanding disparities in the predicted outcome
• Construction of unbiased ML models for neuroscience
• Bias removal of ML use cases for neuroscience
• Interpretability of ML in neuroscience
• Recourse and contestability of biased ML results
• Emphasis on learning underrepresented groups
• Safe reinforcement learning in neuroscience
• Safe neuroscience robotic control
• Ethical and legal consequences of using ML in real-world systems
• Safety of deploying reinforcement learning in neuroscience
• Any novelties for enabling the safety and fairness of ML in neuroscience