Experience the Swiss mountain air, discover Zurich’s Old Town, and work on cutting-edge research projects. The ETH Student Summer Research Fellowship Programme (ETH SSRF), run by the Department of Computer Science, offers this opportunity to around 20 students from all over the world each year. Over a period of two months, programme participants gain initial experience within a research group and enjoy an exciting range of additional activities.  From mountain hikes and trips to the chocolate museum to visits to the Swiss National Supercomputing Centre (CSCS) in Lugano, the experiences on offer from the beginning of July to the end of August are diverse.

However, the programme has a strong research focus. Students can apply for one of nine specialisms. The options cover the entire spectrum of computer science, from machine learning and programming languages to robotics and bioinformatics.

Ambition as proof

One of this year’s ETH SSRF 2025 participants is Clijo Jose. Hailing from Delhi in India, he is currently pursuing a Master’s degree in Applied Mathematics and Statistics at the Institut Polytechnique de Paris. Initially, he was unsure about applying for the SSRF programme. “I had only dealt with computer science to a limited extent during my studies,” he explains. However, during his first year of the master’s programme, he encountered machine learning and quickly developed a passion for it.

“You don’t have to know everything about your research idea already in advance. The programme offers enough time to formulate and develop the idea afterwards.”

Clijo Jose, SSRF student from India

During his time in Zurich, Clijo was a member of the Natural Language Processing and Educational Technologies Lab research group led by Professor Mrinmaya Sachan. There, he compared the natural and formal language reasoning of LLMs. More specifically, he compared the mathematical reasoning abilities of LLMs like ChatGPT in “Lean” and in English.

Lean is an open-source proof assistant and programming language that assists computer scientists and mathematicians in creating and verifying proofs interactively. “It makes it possible to formally verify a mathematical proof step by step,” Clijo explains. However, “Lean” does not understand natural languages such as English, but only a precise, structured input in the Lean language that combines logic, mathematics and programming constructs.

On the other hand, LLMs such as ChatGPT typically attempt to solve mathematical problems in natural language, but can also be used to synthesise proofs in “Lean” just like code generation. Clijo is working on a dataset and evaluation rubric to compare the performance of LLMs in English and in Lean.

According to Clijo, you should have a rough idea of your research project before the SSRF programme begins. “That doesn’t mean you have to know everything about it already. The programme offers enough time to formulate and develop the idea afterwards,” he explains. “It’s important to show ambition at times like these.”

Clijo particularly enjoyed hiking in Switzerland. “The country has so many wonderful places to visit. At the weekends, when I had time, I wanted to visit as many of them as possible with my friends and marvel at the beautiful lakes and landscapes,” he says.

Bursting the bubble

Matthijs Jansen Op de Haar has had big ambitions for a long time. He once wanted to become a doctor and was passionate about biology. “One day, I tried my hand at programming, which sparked a deep and profound interest in computer science,” he says. Originally from the Netherlands, Matthijs is currently studying for two separate bachelor’s degrees at the University of Twente: technical computer science and business information technology. “During my studies, I realised how diverse computer science is. It can be used across disciplines, whether in finance, biology or history,” he explains.

“Applying for the SSRF programme was my way of bursting my own bubble. It allowed me to broaden my horizons, meet new people and tread closer to my limits.”

Matthijs Jansen Op de Haar, SSRF student from the Netherlands

He works in Professor Mennatallah El-Assady’s Interaction, Visualisation and Intelligence Augmentation research group. There, he conducts research on so-called “mixed-initiative visual analytics systems”. In this context, “mixed initiative” means that a human and a software agent (e.g. a large language model) can both take initiative in carrying out a shared task or objective. And “visual analytics” refers to the domain which studies the visual knowledge representation of data.

Matthijs and his supervisor are investigating the domains in which researchers have already implemented these systems. “We are trying to determine the areas in which such systems – in the future – could prove particularly promising,” he explains.

As soon as he arrived in Zurich, a day before starting research, Matthijs spent an entire day exploring the city with his new friends from the SSRF programme. “We got an overview of the city and visited the Kunsthaus and the Landesmuseum, among other places, and tried lots of different foods,” says Matthijs.

He particularly appreciated the opportunity to exchange ideas with motivated and talented people. “It’s generally easy to get caught up in a bubble and be complacent with your current way of life, the people around you and your own world view. Applying for the SSRF programme was my way of bursting that bubble, allowing me to broaden my horizons, meet new people and tread closer to my limits,” he says.

An efficient algorithm for fighting tumours

Urša Zevnik from Slovenia greatly enjoyed the social aspect of the SSRF programme, too. “I had many great encounters and spontaneous experiences,” she says. For example, she was out and about in Zurich with others, intending only to buy groceries, but the day ended with them spending over two hours marvelling at the fireworks for Swiss National Day.

Urša also greatly appreciated the other SSRF students. “We did all these great things together: we tried fondue, visited different Swiss cities and participated in a swim race in the Limmat River in Zurich. Thanks to this whole experience, I made friends for life,” she says.

“The SRRF programme gave me the opportunity to challenge myself in computational problem-solving, which felt almost like playing a strategy game.”

Urša Zevnik, SSRF student from Slovenia

Urša completed her Bachelor’s degree in Biochemistry. “After that, I wanted to broaden my horizons,” she says. During her gap year, she gained research experience in the pharmaceutical industry. She then began a master’s degree in Stockholm combining biomedicine and bioinformatics. “I think it will become increasingly important in the future to apply computer approaches to other sciences,” she says.

In the Biomedical Informatics Group, led by Professor Gunnar Rätsch, Urša is working on a tool that identifies new targets, known as “epitopes”, for cancer immunotherapy. Epitopes are short protein fragments presented on cell surfaces and recognised by the immune system. The algorithm that Urša is helping to develop is designed to identify epitopes that are specific to cancer cells and can therefore be used to activate the immune system to attack only the tumour. “The danger lies in the fact that if a chosen epitope is also present in healthy cells, the therapy could damage healthy organs,” Urša explains.

Urša’s lack of formal computer science training is not an obstacle for her. “The SRRF programme gave me the opportunity to challenge myself in computational problem-solving, which felt almost like playing a strategy game,” Urša says. “It was engaging and fun, and it solidified my decision to pursue a PhD that applies computational approaches to medical research,” she says.

A single letter becomes a hard problem

Violeta Kastreva also gained valuable experience during her time in Zurich. “I learned more in those two months than I usually do in a whole year,” she says.

Originally from Bulgaria, she is currently studying for a bachelor’s degree at Sofia University St. Kliment Ohridski. As a child, she wanted to become an artist. However, she developed a keen interest in mathematics at an early age, and by the time she was in ninth grade, she had become passionate about computer science, especially theoretical computer science. “I like this kind of theoretical research. It’s not about celebrating small successes day after day, but rather a lengthy process that one day leads to a brilliant idea,” says Violeta.

“The programme feels very challenging, especially at the beginning. You might even think that you’re not good enough for the programme. But if you’ve been selected, you’ve earned your place.”

Violeta Kastreva, SSRF student from Bulgaria

Together with her team in the Algorithms and Didactics Group, led by Professor Dennis Komm, Violeta worked on a project to prove that unary tokenisation is an NP-hard problem.

To oversimplify, NP-hard means it belongs to a class of computational problems for which no efficient solution algorithm has been found. Tokenisation, on the other hand, involves breaking down a sequence of symbols into units called tokens. For instance, if the letters “ABC” frequently appear together, they can be combined into a single token. In unary tokenisation, the “text” uses only one character (for example, “AAAA”). Even in this simple setting, choosing the best split into tokens is NP-hard. She succeeded in establishing this result during her stay in Zurich.

Afterwards, Violeta and her team continued to explore related problems, extending the scope of the work. This contribution is relevant not only to theoretical computer science but also to machine learning, where understanding computational limits helps guide the design of practical algorithms.

Violeta advises future participants in the SSRF programme not to put too much pressure on themselves with regard to their research. “Everything feels very challenging, especially at the beginning. You might even think that you’re not good enough for the programme. But if you’ve been selected, you’ve earned your place,” she says.

During her time at ETH Zurich, she learned how challenging the work of a researcher can be. “It’s not like homework, where you know a solution exists,” she says. “Instead, you often work on things that you can’t be sure actually exist.” However, according to her, you can always bear in mind that your research may contribute to the further development of computer science. Even if you don’t have any groundbreaking results, that’s not a problem. “The lessons you’ve learned along the way are already very valuable,” she says.

This article written by Olivier Nüesch was originally published by Department of Computer Science at ETH Zurich.