Silun Zhang - newly appointed Assistant Professor at the Department of Mathematics
Silun Zhang, a recently appointed assistant professor specializing in the mathematical foundations of artificial intelligence, shares insights about his academic journey, interests, and future aspirations.
After completing his Ph.D. from the Division of Optimization and Systems Theory at KTH in 2019, Silun Zhang worked as a Postdoctoral Fellow and Associate at MIT, where he was also a course developer and instructor in the MicroMaster Program in Statistics and Data Science (SDS). Now, he is returning to his academic hometown - KTH.
"I am thrilled to join KTH again as an Assistant Professor in the Department of Mathematics, reuniting with old friends and colleagues while embracing new challenges in the acquainted and pleasant environment."
Academic freedom paves the way for changing the world
The most striking aspect that attracts Silun to work in academia is the unique freedom and autonomy during the research process, allowing researchers to influence the world in a more efficient and effective way.
"As faculty members, we have the privilege of selecting our research directions with minimal constraints. We are free to delve into any subject that aligns with our interests and expertise, encouraging innovation and creative exploration. Moreover, we retain the autonomy to set our own research schedules, a luxury often not found in industry settings. In addition, in contrast to others, academia offers us the remarkable ability to collaborate with experts both within and beyond the academic realm, with minimum technical and patent barriers."
From discoveries to startups
The above conditions uniquely expedite solving the urgent needs of society and make research outcomes and state-of-the-art discoveries easy to transform into startups and spinoffs.
"One example is the Private Automated Contact Tracing (PACT) algorithm launched jointly by Apple and Google during COVID-19 to trace the contact history of smartphone users in a confidential and privacy-preserving way. Prof. Ronald Rivest, the developer of the PACT algorithm, initially regarded the problem as a course project, and collected the major ideas from a mere 60-minute discussion with us in a class. The impetus behind this discussion was purely our shared passion and curiosity, but the impact and potential of the resulting innovation are huge."
"Another illustrative example is our work in the field of data exchange, which has been transferred to the startup entitled OwnData in 2021. This endeavor secured seed funding from MIT Sandbox and is dedicated to establishing a platform that allows internet users to trade the use rights of their datasets in a consensual and privacy-preserving manner. All these attempts to enrich lives through scientific and technological innovations are made possible by the unique asset of faculty in research freedom."
Developing intelligent systems for better informational security
Silun studies how to exchange information in networked systems while still protecting individual privacy. By interconnecting many individuals, networked systems provide each node with access to more timely and relevant information, which can improve system performance, robustness, and versatility.
"However, the exchange of information between nodes in a network can come at the cost of individual privacy, even when local information is sensitive or disastrous to disclose. To address this issue, we have developed privacy-preserving approaches for network coordination that draw inspiration from Secret Sharing in cryptography. This approach allows all participants in a network to coordinate their behaviors and knowledge without exposing individual states to neighbors, and moreover, it is designed to be resistant to node collusion and resilient to node failures. The secret-sharing-based approach can also protect the private dataset of each participant in distributed optimization and federated learning."
Another research topic is moment-based modeling and interventions of large-scale connected systems.
"These systems are ubiquitous around us, such as animal communities, smart grids, economy systems, and human societies. Traditional methods based on individual agents quickly become intractable when the system scale becomes large and are often computationally prohibitive. To solve this problem, we propose a moment-based approach to modeling the dynamics of large-scale networked systems. By characterizing the evolution of the systems using a sequence of generalized moments, we can accurately capture key statistics and behaviors of the system in a low-dimensional moment space. With a set of proper kernels, the moments carry enough information such that the state distributions and/or some specific collective behaviors of the original systems can be reconstructed with a reasonable granularity."
This method is applicable to various applications of large-scale decentralized systems, including sensor networks, transportation systems, infrastructure networks, human-robot teaming, and many problems in collective behaviors occurring in biology, economy, and social sciences.
"In the future, I will continue my research endeavors to construct networked and intelligent systems with more computational efficiency, higher-level autonomy, and better informational security."
Collaborations with the industry
"The industrial impact of good research may arrive late but never absent. This assertion is particularly true within the field of Systems Theory, a discipline born and nurtured in solving engineering challenges involving multiple interactive and complex objects."
"Therefore, my research endeavors focus on both advancing understanding, methods, and theory in Intelligent Physical Systems, and providing innovative solutions to the emerging challenges encountered in industry. At present, some research collaborations with our industrial partners, including SAAB and Zenseact, have been brewing in the realm of advanced autonomy. "
Text: Danai Deligeorgaki, Marta Marko-Tisch