Title : Securing brain-computer interfaces: Emerging threats and defense strategies
Abstract:
Brain-Computer Interfaces are on a cross road with BCIs quickly moving out of the research avenues to clinical use and consumer applications with more than 400 ongoing clinical studies around the world and market estimates of more than $5 billion in the next decade. However, as these systems start to reach households and the everyday lives of patients, we have a burning question that the neuroscience community has not really given much thoughtful consideration on, and that is Are BCIs safe in the real world environment?
The given talk brings up the next area of neuro-cybersecurity, which is a jaw dropping break in the modern development of BCI as less than 5% of the BCI studies published touch on the issue of security. With BCIs turning into cyber-physical systems operating wheelchairs, prosthetics, and devices to support the vulnerable groups in communication, they also acquire critical vulnerabilities that provide adversaries with access to vulnerable populations, infringe neural privacy, and endanger the integrity of research.
The talk illustrates four key categories of threats, including adversarial attacks with 80-95 percent misclassification rates with imperceptible signal perturbations, privacy violations that allow to extract PINs and medical data out of neural data, model poisoning and introducing hidden front doors in machine learning systems, and physical layer attacks that can use commodity-based hardware to corrupt signals via electromagnetic interference. All the threats are evidenced in published research, and not speculation. The talk will also introduce applications such as practical defense mechanisms such as adversarial training, differential privacy, encrypted communication and real time anomaly detection, all of which retain classification accuracy of over 90 percent with a latency overhead imposed.
There is just time before BCIs can be mass deployed. This talk will help create a fundamental change in the approach of neuroscientists towards the development of BCI by realizing that the future of neural interfaces does not lie in extending the accuracy of decoding but rather in creating secure, trusting systems that patients can expect and regulators can insist. Not whether they will be attacked, but whether we shall get them secured the first thing.
