PhD Ieva Daukantas has been accepted for a research stay abroad at Stanford University
Ieva Daukantas is working on her PhD in Information Security at ITU, and at the moment she is a visiting researcher at Stanford University, USA. She investigates formal methods and proofs for security guarantees in machine learning, and she is highly motivated by the fact that her field of research emerges from a high demand in the industry.
ResearchalgorithmsIT securityartificial intelligence
Written 8 June, 2023 08:06
Since the early days of Ieva Daukantas’ academic journey, she has aspired to find connecting points between diverse research areas. Ieva Daukantas holds a bachelor’s degree in psychology including core courses of economics from Vilnius University. She worked in the industry for five years, and she saw the demand for IT expertise increase severely. The Lithuanian researcher came to Denmark in 2017 to start her MSc in Software Development at ITU.
"The educational skillset I have is quite unique and combines knowledge about human brains, behavior and computer science. It helps me to approach theoretical challenges from many different angles and find solutions that solve practical problems,” says Ieva Daukantas.
Now, Ieva Daukantas is working on her PhD in Information Security at ITU and Stanford University. “To me, knowing that my research emerges because of a huge business demand and has a very broad applicability is very motivating. My PhD supervisors are especially knowledgeable and supportive, and I find the working environment at ITU very accepting and encouraging. It allows me the flexibility that I need while combining work and motherhood, and I have felt welcome from day one,” says Ieva Daukantas.
Investigating security guarantees in machine learning
In her PhD, Ieva Daukantas investigates formal methods and proofs for security guarantees in machine learning. She explores different tools, such as automated theorem provers, to evaluate if algorithms will behave as intended in security attack scenarios.
“Practically, such research is essential in the systems that make life-related decisions. For example, a self-driven car is controlled by a neural network. However, how can we make sure that it will not fail in some unforeseen circumstances? How will it behave when getting unforeseen data as an input? Will the decisions such as “turn left”, “turn right”, “accelerate”, “stop” be adequate even if the neural network gets attacked by hackers?” Ieva Daukantas says.
Ieva Daukantas points out that there are different ways of providing such guarantees. “An ambition is to have a theoretical guarantee of the model behavior in certain set ups. They can be expressed mathematically, and it is possible to calculate error bounds within a given specification on a set of constraints. A theoretical proof can have innumerous applications. However, state-of-the-art neural networks are huge in scale, hard to explain, and they evolve very fast. This is what makes my research challenging,” Ieva Daukantas says.
Ieva Daukantas emphasizes that her research has broad applicability and can be used for various purposes that need security verification and safety certification, for instance for automotive, medical, financial and many other fields where neural networks are applied.
Visiting researcher at Stanford University
As part of her PhD, Ieva Daukantas is now visiting Stanford University, USA. She sees her time abroad as a perfect opportunity to expand her knowledge, create connections, explore different research environments, and learn different approaches to her research.
“Stanford University is one of the leading educational institutions in the world. This visit is a perfect opportunity for me to expand my knowledge in the field, and to learn from my peers. It is such a privilege to participate in the state-of-the-art research,” Ieva Daukantas says.
Ditte Ørsted Johansen, Press Officer, phone +45 25 55 04 47, email ditj@itu.dk