INSAIT Research Areas
Machine learning
The abundance of data and the increasing accessibility of computational power in the 21st century create exciting opportunities for data-driven algorithms. Machine learning, the area of computer science that focuses on extracting knowledge from large datasets and using this knowledge for making predictions and decisions, now provides state-of-the-art methods for many applications of computer science.
Given the increasing practical relevance of machine learning, numerous important questions emerge. How can we ensure that learning-based algorithms are sufficiently robust, fair, and reliable to be used in practice? How can we leverage data in an optimal way for solving a certain task? At INSAIT, we seek to provide answers to such questions, striving for a thorough theoretical and practical understanding of data-driven algorithms.
Faculty & mentors involved in this research area:
- Prof. Ce Zhang (ETH Zurich, Switzerland)
- Prof. Andreas Krause (ETH Zurich, Switzerland)
- Prof. Russ Salakhutdinov (Carnegie Mellon University, USA)
- Prof. Martin Jaggi (EPFL, Switzerland)
- Prof. Martin Vechev (ETH Zurich, Switzerland)
- Dr. Nikola Konstantinov (INSAIT, Bulgaria)
Computer vision
The visual system is one of the most important tools through which humans understand and interact with the surrounding world. To a computer, however, a picture or a video is by default nothing more than a bunch of numbers, corresponding to sequences of RGB values. Computer vision is an interdisciplinary field that studies algorithms through which computers can get a high-level understanding of images. With its numerous applications to areas such as autonomous driving, medical image analysis, perceptual robotics, embodied AI, object recognition, gesture analysis, tracking and scene understanding, among others, computer vision is a crucial component of modern AI systems and a central part of the research at INSAIT.
Faculty & Mentors involved in this research area:
- Prof. Luc van Gool (INSAIT, Bulgaria)
- Prof. Otmar Hilliges (Max Plank Fellow, ETH Zurich)
- Dr. Danda Paudel (INSAIT, Bulgaria)
Quantum computing
The impressive increase in the capabilities and availability of conventional computers over the last 50 years has brought enormous benefits to our society. Quantum computing, a type of computation that can leverage properties of quantum states, has the potential to further revolutionise computer science by allowing computations that are impractical on modern computers to be performed in a matter of seconds on such novel hardware. Postulating the development of quantum computers, computer scientists need to be ready to both harvest the potential of these new types of machines and address the numerous challenges that will arise in cyber security. At INSAIT, we focus on the software aspects of quantum computation, including programming languages, optimization, verification, and algorithms.
Faculty & mentors involved in this research area:
- Prof. Martin Vechev (ETH Zurich, Switzerland)
Cyber security
Over the last few decades, the increasing adoption of wireless technologies and the Internet have transformed the way in which we exchange information, benefiting society by making it more connected. Unfortunately, this comes with various security risks, as computer systems and networks are vulnerable to insider and outsider attacks, leading to potential information disclosure or theft. INSAIT’s research in computer security aims to protect modern computer systems from malicious interventions by both studying fundamental questions in cryptography, new methods which provide formal security guarantees, and developing practical and secure communication protocols and systems.
Automated reasoning
Reasoning, the ability of humans to consciously draw conclusions based on applying logical arguments, is perhaps the most unique feature of human intelligence. Therefore, the study of automated reasoning is a cornerstone of modern artificial intelligence research, with far-reaching applications, for example in theorem proving, proof checking, and software verification. At INSAIT, we work on fundamental questions in mathematical logic and decision-making under uncertainty, as well as on applications and advancements of existing automated reasoning techniques, e.g. those used in SAT solvers and for automated proof checking.
Formal methods
Modern computer systems play a central role in our society and are often deployed in security- and privacy-critical domains, such as aviation and internet banking. In such cases, it is crucial to be able to guarantee the correctness of the used algorithms so that the resulting system is provably secure. The development of formal methods, that is, mathematically rigorous techniques used to verify the correctness of programs, is therefore vital for increasing the trust in software systems among the general public. At INSAIT, we conduct fundamental research on formal methods and also study their applications, with the goal of developing provably robust, safe, and interpretable software systems.
Computer systems and networks
Modern computers rarely perform in isolation but are rather incorporated into sophisticated computer systems and networks. This allows for leveraging the varying functionalities of different types of devices and for harnessing the benefits of horizontal scaling. Ensuring the correctness and scalability of algorithms used in such systems poses various theoretical and practical challenges for computer scientists. At INSAIT, we work on a wide range of system fields, including operating systems, networks, systems security, databases, high-performance computing, and distributed systems.
Natural language processing
Speech perception, the ability of humans to interpret and understand language, is a central piece of human intelligence. Natural language processing, the study of computational techniques for automated processing and analysis of text and speech, is therefore a cornerstone of contemporary artificial intelligence. At INSAIT, we aim to develop principled algorithms for processing massive amounts of text and speech data, studying both approaches based on deep learning and methods for incorporating insights from linguistics.
Faculty & mentors involved in this research area:
- Kristina Toutanova (Google)
Programming languages
The area of programming languages is focused on developing new methods for creating more secure, reliable, and efficient software, including new programming abstractions (e.g., Rust), program synthesis techniques (inducing programs from sketches or examples), program analysis (connected to automated reasoning), optimising compilers, type systems, symbolic reasoning, and many more. The area is especially suited for interdisciplinary research and connects well with topics in machine learning, neuro-symbolic reasoning, systems and networks, and others.
Faculty & mentors involved in this research area:
- Martin Vechev (INSAIT/ETH Zurich)
- Dimitar Dimitrov (INSAIT, Bulgaria)
Algorithms and theory
The development of modern computers has enabled the scalable application of algorithms, that is, finite sequences of rigorous instructions that solve a specific task. However, the concept of an algorithm is in fact much older, dating back to antiquity, where ancient mathematicians and philosophers concerned themselves with problems such as efficient finding of prime numbers and the greatest common divisor of two numbers. Nowadays, the area of theoretical computer science studies many advanced topics concerning the optimal design of algorithms and the computational hardness of certain tasks. By working on such challenging problems, many of which have been open for centuries, scientists at INSAIT aim at improving our understanding of what tasks are solvable on modern computers and how one can design software systems as efficiently as possible.
Data management
The increasing adoption of computers, smart devices and the Internet in the 21st century has led to a massive increase in the total volume of data that is being created and processed every year. With data being the key ingredient for many modern decision-making systems, including those based on machine learning, it is increasingly important that we store and maintain data in a scalable and secure way. At INSAIT, we study many fundamental questions regarding data management, including topics related to computational and memory efficiency, privacy, data governance, data erasure and many others.