É«É«À²

The Department of Computer Science is looking for summer employees

We at the Department of Computer Science want to offer motivated students a chance to work on interesting research topics with us. We are looking for BSc or MSc degree students at Aalto or other universities to work with us during the summer 2025. If you have enjoyed your studies and want to learn more about computer science, this might be your place. We do not expect you to have previous research experience; this could be the start of your bright researcher career! You will be supported by other summer employees, doctoral students and postdocs at the department.

Ready to apply?

See the complete list of the available topics below (will be fully updated by January 2^nd)) and choose the topic(s) (max. 5) that interest you the most. Please indicate them in order of preference in the relevant section on the application form.

Please submit your application through our recruitment system. The application form will open on January 7^th and close on January 31^st, 2025, at 23:59 Finnish time (UTC +2).

Link to the application form:

Are you an international student or coming from abroad?

Please check the Aalto Science Institute AScI internship programme for international summer employees.
/en/aalto-science-institute-asci/how-to-apply-to-the-asci-international-summer-research-programme

AScI arranges activities for international summer employees who have applied through their call and helps in finding an apartment in Espoo.

More information

If you have questions regarding applying, please contact Susanna Holma from HR Team. firstname.lastname@aalto.fi.

Summer employee topics 2025

Topics are listed here and will be fully updated by January 7^th at the latest.

1. Diplomityöntekijä terveysteknologian alalle

Supervisor: Sari Kujala

Contact info: sari.kujala@aalto.fi

Number of open positions: 1

Diplomityön aiheena on ikäihmisen omaishoitajan digitaaliset ratkaisut. Vaatimuksena on HCI-alan menetelmien hallitseminen ja suomen kielen osaaminen. Pääaineena mielellään HCI tai Informaatioverkostot ja tutkimusmenetelmäkurssin suoritus katsotaan eduksi.

2. Designing for Developer Experience

Supervisor: Fabian Fagerholm

Contact info: fabian.fagerholm@aalto.fi

Number of open positions: 1-3

The Mind and Software research group is looking for skilled and motivated research assistants to contribute to our research on developer experience. Developer experience refers to the cognitive, motivational, and affective experience that software developers have while developing software. We design methods and tools for studying and assessing developer experience in various contexts in modern software development.

We are looking for people to perform different kinds of tasks, and you may be involved in one or more of them in different combinations. In research-oriented tasks, you participate in planning and conducting empirical studies with software developers, collecting data through interviews, observation, or instrumentation, analysing data using qualitative and quantitative methods, and reviewing existing scientific literature. In technically oriented tasks, you participate in developing software for developer experience measurement. Finally, in design-oriented tasks, you contribute user interface and visual designs for research materials, questionnaires, and measurement tool user interfaces.

Required skills (research-oriented):

- Familiarity with empirical studies (e.g., interviews, think-aloud, cognitive task analysis).

- Understanding of the basics of HCI and/or psychology (e.g., cognitive or social psychology).

Required skills (technically oriented):

- Familiarity with modern web development (e.g., HTML5, CSS, JavaScript, Python).

- Familiarity with mobile app development (e.g., Android, iOS, React Native).

- Familiarity with collaborative software development (e.g., Git, Continuous Integration).

Required skills (design-oriented):

- Familiarity with user interface design (particularly mobile and web).

- Ability to create visually appealing design elements following existing design guidelines and with an understanding of the research goals of the design.

- Familiarity with visual design tools (e.g., Figma, Miro).

- Understanding of the basics of HCI and/or psychology (e.g., cognitive or social psychology).

In addition, we particularly appreciate skills and/or interest in:

- Understanding of research instrument development (e.g., questionnaire design).

- Specific qualitative or quantitative research methods (e.g., thematic analysis, descriptive statistics, statistical analysis).

- Teamworking skills.

- Academic writing skills (in English).

The applicant is not required to be an expert in these areas but should display a good foundation and willingness to learn and develop their skills on their own initiative as well as in collaboration with other members of the research group.

3. Methods and models for Continuous Experimentation

Supervisor: Fabian Fagerholm

Contact info: fabian.fagerholm@aalto.fi

Number of open positions: 1-3

The Mind and Software research group is looking for skilled and motivated research assistants to contribute to our research on Continuous Experimentation (CE). CE is an approach where field experiments with real users inform software product development, for example, through A/B testing. We are investigating methods and models for various aspects of the CE process and for different kinds of organisations and products.

In this project, you would contribute to ongoing research to support development of ways to identify and specify what to test in experiments, how to produce representative experiment objects to use in the experiments, and to understand how humans make decisions in the experimentation process. You would participate in conducting empirical studies with software practitioners, which may involve interviews, questionnaires, or task-oriented studies and the analysis of data from these. The position can be combined with a Master's thesis if you are enrolled at Aalto University.

Required skills:

- An interest in software product development, both in general and using experiment-based methods in particular (e.g., A/B testing).

- Basic knowledge of experimental design.

- An understanding of research with human subjects.

- Ability to read and summarise scientific literature.

- Academic writing skills (in English).

In addition, we particularly appreciate skills and/or interest in some of the following:

- Interview methods (both individual and group) and analysis of interview data (e.g., thematic analysis, cognitive task analysis).

- Questionnaire-based methods (e.g., designing an implementing online questionnaires) and analysis of questionnaire data (e.g., descriptive statistics, hypothesis testing).

- Creating prototypes of different fidelity for experiments (e.g., ranging from wireframes to working software prototypes).

- Understanding of the basics of HCI and/or psychology in human studies (e.g., cognitive or social psychology), particularly regarding human decision-making.

The applicant is not required to be an expert in these areas but should display a good foundation and willingness to learn and develop their skills on their own initiative as well as in collaboration with other members of the research group.

4. Causal approach for design optimization problem in the presence of distribution shift by leveraging human feedback

Supervisor: Zahra Rahiminasab, Samuel Kaski

Contact info: zahra.rahiminasab@aalto.fi

Number of open positions: 1

In a design optimization process a machine learning model tries to find an optimal design based on some predefined metric. However, the machine learning model is usually trained on known source environments and must be generalized to the new target environment. In this case, the problem is formulated as a domain adaptation problem. Information regarding domain shifts in environment distribution can be provided by human feedback. Causal modeling formalizes required assumptions to establish causal relationships between treatment (covariates) and outcomes. Causal exchangeability is one of the main assumptions in a classic causal modeling setting. Unconfoundedness means there is no unobserved shared cause between treatment and outcome that can affect both. However, such an assumption is violated in practice, and an unobserved confounder can be a source of distribution shift. While there are approaches that address unobserved confounders in causal literature, they are usually based on restricting assumptions that do not hold in practice. In this internship, we empirically explore the violation of such assumptions and their impacts on causal identifiability.

Prerequisites:

Python, PyTorch,  being curious about causal modeling.

References: 

[1] Hernán MA, Robins JM (2020). Causal Inference: What If. Boca Raton: Chapman & Hall/CRC-Chapter 3

[2] Zhou, Kaiyang, et al. "Domain generalization: A survey." IEEE Transactions on Pattern Analysis and Machine Intelligence 45.4 (2022): 4396-4415.

[3] Alabdulmohsin, N. Chiou, A. D’Amour, A. Gretton, S. Koyejo, M. J. Kusner, S. R. Pfohl, O. Salaudeen, J. Schrouff, and K. Tsai. Adapting to latent subgroup shifts via concepts and proxies. In International Conference on Artificial Intelligence and Statistics, pages 9637–9661. PMLR, 2023

5. Leveraging graph neural networks to understand Finnish healthcare data

Supervisor: Jorge Loría, Samuel Kaski

Contact info: jorge.loria@aalto.fi

Number of open positions: 1

Graph neural networks (GNNs) facilitate understanding complex relationships between variables in many areas of engineering and science (Scarselli, et al. 2009). An exciting application of these models is to understand how the graphs of physicians and patients can affect healthcare outcomes. Specifically, the intern will investigate prescription patterns of doctors, while leveraging characteristics of patients and clinicians. Understanding these complex relationships with GNNs will shed light in possible over(under)prescriptions, as well as help describe relationships between patient characteristics and doctor behaviors.

Prerequisites: familiarity with GNNs, experience with pytorch/jax and deep learning. Experience with Bayesian machine learning is preferred but not required.

References

F. Scarselli, M. Gori, A. C. Tsoi, M. Hagenbuchner and G. Monfardini, "The Graph Neural Network Model," in IEEE Transactions on Neural Networks, vol. 20, no. 1, pp. 61-80, Jan. 2009, doi: 10.1109/TNN.2008.2005605.

6. Improving on Upper confidence Reinforcement Learning (UCRL) leveraging human insight

Supervisor: Mahsa Asadi, Samuel Kaski

Contact info: mahsa.asadi@aalto.fi

Number of open positions: 1

UCRL is a model-based reinforcement learning (RL) approach with infinite horizon trajectory and optimizing the average reward criterion. This is one of the important baselines providing theoretical guarantees on the algorithm regret. There are various works in the literature extending this approach in order to improve the regret bound. However, there is not much work considering human in the loop of learning and analyzing the RL agent performance. In this internship, we are going to explore some of the extensions of UCRL, come up with useful human feedback, analyze our proposed algorithm and implement the idea and show that it works in practice as well.

Prerequisites: Familiarity with python programming language and reinforcement learning, being curious about how to provide theoretical guarantees.

References: 

[1] Auer, Peter, Thomas Jaksch, and Ronald Ortner. "Near-optimal regret bounds for reinforcement learning." Advances in neural information processing systems 21 (2008).

[2] Bartlett, P. L., & Tewari, A. (2012). REGAL: A regularization based algorithm for reinforcement learning in weakly communicating MDPs. arXiv preprint arXiv:1205.2661.

[3] Fruit, R., Pirotta, M., & Lazaric, A. (2018). Near optimal exploration-exploitation in non-communicating markov decision processes. Advances in Neural Information Processing Systems, 31.

7. Deep ensemble based approaches for causal discovery

Supervisor: Nazaal Ibrahim, Samuel Kaski

Contact info: nazaal.ibrahim@aalto.fi

Number of open positions: 1

Causal models, specifically causal graphs, are essential to predict how a system behaves under interventions. This is important in applications ranging from healthcare to economics. However, learning causal models from data is a nontrivial task. Many probabilistic approaches for causal model learning exist [1,2,3], which rely on Bayesian inference methods such as Stein Variational Gradient Descent and Stochastic Gradient Markov Chain Monte Carlo. Rather than using such inference methods, this project will be about investigating the use of deep ensemble style approaches [4] to learn causal models. If time permits, there is room to focus on incorporating extra information sources such as human feedback [5], and the use of these methods in Bayesian experimental design settings.

Prerequisites: Familiarity with PyTorch and/or JAX, experience with deep learning and Bayesian machine learning.

[1] Lorch, Lars and Rothfuss, Jonas and Schölkopf, Bernhard and Krause, Andreas. DiBS: Differentiable Bayesian Structure Learning, NeurIPS 2021. https://proceedings.neurips.cc/paper/2021/hash/ca6ab34959489659f8c3776aaf1f8efd-Abstract.html

[2] Annadani, Yashas and Pawlowski, Nick and Jennings, Joel and Bauer, Stefan and Zhang, Cheng and Gong, Wenbo. BayesDAG: Gradient-Based Posterior Inference for Causal Discovery, NeurIPS 2023. https://openreview.net/forum?id=woptnU6fh1&noteId=tPDMTdnWfQ 

[3] Toth, Christian and Knoll, Christian and Pernkopf, Franz and Peharz, Robert. Effective Bayesian Causal Inference via Structural Marginalisation and Autoregressive Orders. ICML Workshop on Structured Probabilistic Inference & Generative Modeling 2024. https://openreview.net/forum?id=WrKcYKI6kL 

[4] Lakshminarayanan, Balaji and Pritzel, Alexander and Blundell, Charles. Simple and Scalable Predictive Uncertainty Estimation Using Deep Ensembles. NeurIPS 2017. https://proceedings.neurips.cc/paper/2017/hash/9ef2ed4b7fd2c810847ffa5fa85bce38-Abstract.html

[5] Ibrahim, Nazaal and John, S. T. and Guo, Zhigao and Kaski, Samuel. Targeted Causal Elicitation. NeurIPS Workshop on Causal Machine Learning for Real-World Impact 2022. 

8. Accelerating Bayesian inference with modern deep learning techniques

Supervisor: Daolang Huang, Samuel Kaski

Contact info: daolang.huang@aalto.fi

Number of open positions: 1

Pre-trained large language models have advanced rapidly in recent years, yet the intersection of traditional Bayesian inference and neural networks remains in its early stages of development. This project explores whether we can develop powerful pre-trained models to accelerate Bayesian inference tasks. The primary focus will be investigating novel amortized inference methods by leveraging cutting-edge neural network techniques to significantly speed up inference in various Bayesian tasks. These tasks include Bayesian optimization, Bayesian experimental design, and simulation-based inference. The intern will have the freedom to choose and explore topics based on their interests. This work builds on our group’s prior research [1, 2, 3]. Collaborators will have the opportunity to contribute directly to projects aimed at publishing in leading venues.

Prerequisites: experience with deep learning, familiarity with concepts of Bayesian inference, and reinforcement learning. Proficiency with Python (PyTorch).

[1] Chang, P. E., Loka, N., Huang, D., Remes, U., Kaski, S., & Acerbi, L. (2024). Amortized probabilistic conditioning for optimization, simulation and inference. arXiv preprint arXiv:2410.15320.

[2] Huang, D., Guo, Y., Acerbi, L., & Kaski, S. (2024). Amortized Bayesian Experimental Design for Decision-Making. Neurips.

[3] Zhang, X., Huang, D., Martinelli, J., & Kaski, S. (2024). PABBO: Preferential Amortized Black-Box Optimization.

9. Expert-driven Conformal Prediction Set Refinement Improves Robustness in Bayesian Optimization

Supervisor: Marshal Sinaga, Samuel Kaski

Contact info: marshal.sinaga@aalto.fi

Number of open positions: 1

Bayesian Optimization (BO) is widely regarded as a gold-standard method for optimizing expensive black-box functions. To enhance its robustness against model misspecification and covariate shifts, several studies have introduced conformal prediction sets as a promising approach [1, 2]. However, existing methods predominantly rely on purely data-driven techniques. This internship seeks to explore the integration of human expert knowledge to refine or reconstruct conformal prediction sets, leveraging the domain-specific insights that experts provide to improve robustness in BO. You will have the freedom to explore the experimental or theoretical aspects.

Prerequisites: 

Solid understanding of Bayesian inference.

Having a background in GP or BO will be a plus.

Familiarity with GP’s or BO’s library (e.g., GPyTorch, GPflow, BoTorch, etc.) will be a plus.

References: 

Stanton, S., Maddox, W., & Wilson, A. G. (2023, April). Bayesian optimization with conformal prediction sets. In International Conference on Artificial Intelligence and Statistics (pp. 959-986). PMLR.

Deshpande, S., Marx, C., & Kuleshov, V. (2024, April). Online Calibrated and Conformal Prediction Improves Bayesian Optimization. In International Conference on Artificial Intelligence and Statistics (pp. 1450-1458). PMLR.

10. Improving the robustness of high-dimensional data modeling with shrinkage priors and projective predictive inference

Supervisor: Ersin Yilmaz, Samuel Kaski

Contact info: ersin.yilmaz@aalto.fi

Number of open positions: 1

In high-dimensional data modeling, a critical issue is the presence of weak effects —predictors with small yet meaningful contributions to the response variable—which are often overlooked or excluded by traditional methods. These weak signals can hold valuable information, especially in complex systems where subtle patterns drive critical insights. Projection predictive inference offers a robust solution to the challenges of considering weak effects and interpretability in high-dimensional settings, particularly in "small n, large p" problems. This approach, as detailed in Piironen et al. [1], separates predictive modeling and feature selection into two stages: first, constructing a reference model using all predictors for maximum accuracy, and second, projecting this onto simpler submodels to retain predictive power while reducing complexity. Compared to well-known regularization methods like adaptive Lasso [2] or Elastic Net [3], projection predictive inference ensures optimal trade-offs between sparsity and prediction by explicitly preserving the information from the reference model that involves the weak effects. It is possible to explore extensions of this method, including its integration to multi-output regression to capture dependencies among outputs, integration with latent factor models for handling structured data [4], and the incorporation of advanced shrinkage priors such as the horseshoe+ [5] to improve sparsity and stability. These methodological developments can expand its adaptability to high-dimensional problems where the weak effects and interpretability are critical.

Prerequisites: Bayesian machine learning, regression techniques, a willingness to explore theoretical inferences, and familiarity with RStan in R and Python.

References

[1]. Piironen, J., Paasiniemi, M., & Vehtari, A. (2020). Projective inference in high-dimensional problems: Prediction and feature selection. Electronic Journal of Statistics, 14(2), 2155–2197.

[2]. Zou, H. (2006). The adaptive Lasso and its oracle properties. Journal of the American Statistical Association, 101(476), 1418–1429.

[3]. Hastie, T., Tibshirani, R., & Wainwright, M. (2015). Statistical learning with sparsity: The lasso and generalizations. Chapman & Hall/CRC.

[4]. Bhattacharya, A., & Dunson, D. B. (2011). Sparse Bayesian infinite factor models. Biometrika, 98(2), 291–306.

[5]. Bhadra, A., Datta, J., Polson, N. G., & Willard, B. (2019). Horseshoe+ estimator: Optimal properties and practical considerations. Bayesian Analysis, 14(4), 1105–1132.

11. Stackelberg games for human-AI collaborative machine learning

Supervisor: Alvar Haltia, Nazaal Ibrahim, Samuel Kaski

Contact info: alvar.haltia@aalto.fi

Number of open positions: 1

Human-AI collaboration involves interaction between both the human agent and the AI agent where each agent can receive information from different sources, and may have differing value functions. The interaction between both agents can be modelled as a turn-based game called a Stackelberg game - a canonical model for sequential games. They involve 2 components:

- A leader who first chooses an action.

- A follower(s) who then chooses the best-response based on the leader's action.

The solution concept, the Stackelberg equilibrium, is where the leader's long term reward is maximized assuming best-response from the follower.

In the context of Human-AI collaboration with similar action/response dynamics, this framework can be used to model the behaviour of the human in response to the AI’s actions. This can be used to infer a user model, or to train robust AI.

Some existing examples of using Stackelberg games in machine learning include Inverse (Reinforcement) Learning [1], modelling distribution shift under adversarial users [2], Reinforcement Learning from Human Feedback (RLHF) [3] and assistive driving [4].

The project will involve:

- A comprehensive literature review on Stackelberg games in machine learning.

- Formulating and analyzing a concrete scenario for AI assistance in a human-AI collaborative setting in the Stackelberg framework, including applying existing algorithms/developing new algorithms for it.

Prerequisites: Experience in Python and some machine learning framework like PyTorch or JAX. Familiarity with reinforcement learning will be helpful.

[1] Ward, W., Yu, Y., Levy, J., Mehr, N., Fridovich-Keil, D., & Topcu, U. (2023). Active Inverse Learning in Stackelberg Trajectory Games. arXiv. arXiv:2308.08017. https://arxiv.org/abs/2308.08017 

[2] Brückner, M., & Scheffer, T. (2011). Stackelberg games for adversarial prediction problems. In Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 547–555). New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/2020408.202049 

[3] Makar-Limanov, J., Prakash, A., Goktas, D., Greenwald, A., & Ayanian, N. (2024). STA-RLHF: Stackelberg Aligned Reinforcement Learning with Human Feedback. Coordination and Cooperation for Multi-Agent Reinforcement Learning Methods Workshop. https://openreview.net/forum?id=Jcn8pxwRa9 

[4] Zhao, Y., & Zhu, Q. (2024). Stackelberg Meta-Learning Based Shared Control for Assistive Driving. arXiv preprint arXiv:2403.10736. https://arxiv.org/abs/2403.10736

12. Maximally Autonomous AI assistant with co-adaptive user modeling

Supervisor: Yujia Guo, Samuel Kaski

Contact info: yujia.guo@aalto.fi

Number of open positions: 1

Sequential decision-making has become a cornerstone of modern AI systems, enabling remarkable advances in areas such as autonomous vehicles, healthcare, and personal assistants. However, existing AI assistants [1] often fail to account for the dynamic nature of human decision-makers' beliefs and their adaptation to AI-generated recommendations. This project focuses on building a Maximally Autonomous AI Assistant capable of providing co-adaptive assistance by inferring human beliefs about the AI’s recommendations and decision-making processes [2, 3, 4]. The assistant will dynamically adjust its autonomy level, offering help only when necessary to maximize human productivity and minimize cognitive overload. Human beliefs will be modeled probabilistically and integrated with reinforcement learning to adaptively determine the optimal degree of autonomy across different scenarios.

Prerequisites: Bayesian inference, reinforcement learning, experience with deep learning and Python (PyTorch).

[1] De Peuter S, Kaski S. Zero-shot assistance in sequential decision problems[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2023, 37(10): 11551-11559.

[2] Nikolaidis S, Zhu Y X, Hsu D, et al. Human-robot mutual adaptation in shared autonomy[C]//Proceedings of the 2017 ACM/IEEE International Conference on Human-Robot Interaction. 2017: 294-302.

[3] Çelikok M M, Murena P A, Kaski S. Modeling needs user modeling[J]. Frontiers in Artificial Intelligence, 2023, 6: 1097891.

[4] Bhat S, Lyons J B, Shi C, et al. Effect of Adapting to Human Preferences on Trust in Human-Robot Teaming[C]//Proceedings of the AAAI Symposium Series. 2023, 2(1): 5-10.

13. Integrate experts' knowledge to develop privacy-preserving models with improved utility

Supervisor: Yaohong Yang, Samuel Kaski

Contact info: yaohong.yang@aalto.fi

Number of open positions: 1

Differential privacy is a technique that enables the development of machine learning algorithms with strong privacy guarantees. This approach protects sensitive information while still allowing useful insights to be extracted from the data. Recent advancements have demonstrated that it is possible to achieve both strong privacy and high accuracy by leveraging a two-step process: first, pre-training machine learning models on publicly available, non-sensitive data, and then fine-tuning these models on sensitive data with differential privacy mechanisms in place. This approach minimizes the privacy risk during the fine-tuning phase while maintaining the model's overall performance. However, different datasets have different tradeoffs between privacy level and model performance. How to integrate human knowledge to develop privacy-preserving models with improved utility remains to be unexplored. This project will review the literature on existing solutions and explore potential research directions. 

Requirements: familiarity with some probabilistic machine learning

References:

[1] De, S., Berrada, L., Hayes, J., Smith, S. L., & Balle, B. (2022). Unlocking high-accuracy differentially private image classification through scale. arXiv preprint arXiv:2204.13650.

[2] Sander, T., Stock, P., & Sablayrolles, A. (2023, July). Tan without a burn: Scaling laws of dp-sgd. In International Conference on Machine Learning (pp. 29937-29949). PMLR.

[3] Ozaki, R., Ishikawa, K., Kanzaki, Y., Takeno, S., Takeuchi, I., & Karasuyama, M. (2024, March). Multi-Objective Bayesian Optimization with Active Preference Learning. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 38, No. 13, pp. 14490-14498).

14. Toward Robust and Reliable Algorithmic Recourse

Supervisor: Corinna Coupette

Contact info: corinna.coupette@aalto.fi

Number of open positions: 1–2

As part of a collaboration between the Telos Lab at Aalto University, Finland () and the CHARM Lab at the University of Waterloo, Canada (), we invite applications for several projects conducting frontier research in the area of algorithmicrecourse [1]. Algorithmic recourse provides individuals with actionable steps to reverse or contest unfavorable decisions made by AI systems. By enabling affected individuals to understand and respond to automated decisions, algorithmic recourse plays a crucial role in the development of responsible artificial intelligence. Depending on interns’ preferences and skills, topics for summer projects include:

1. Strategic behavior and robust recourse: Design recourse mechanisms that are resilient to manipulation while still offering meaningful and actionable options to users. (Desirable: Familiarity with game theory.)
2. Secure and private recourse:Develop robust algorithms that balance the need for effective recourse with the imperative to protect against security breaches and privacy violations. (Desirable: Familiarity with threat modeling and privacy-preserving ML techniques.)
3. Algorithmic recourse and law: Analyze how algorithmic recourse relates to legal requirements (e.g., the right to an explanation) and legal procedures (e.g., judicial proceedings). (Required: Legal background.)
4. Recourse in networks: Analyze the challenges of algorithmic recourse for node-level and edge-level graph-learning tasks and develop robust algorithms for algorithmic recourse in networks. (Required: Experience working with graphs and networks.)
5. Benchmarking recourse methods: Contribute to an open-source benchmark suite for algorithmic-recourse methods (cf. ; [2]). (Desirable: Experience in software-engineering.)

Programming Languages: Python, Julia, R
Libraries/Frameworks: TensorFlow, PyTorch, Scikit-learn, game-theory libraries (e.g., Nashpy), CounterfactualExplanations.jl, CausalML, CAI Algorithmic Recourse, OpenMined (privacy-preserving ML)
Potential Tasks: Literature review; algorithm development and experimental validation; theoretical analysis; benchmark design, validation, and testing; algorithm implementation; collaboration and community engagement

References:

1. Karimi, A.-H., Barthe, G., Schölkopf, B., & Valera, I. (2022). A Survey on Algorithmic Recourse: From Counterfactual Explanations to Interventions. ACM Computing Surveys (CSUR).
2. Pawelczyk, M. et al. (2021). "CARLA: A Python Library to Benchmark Algorithmic Recourse and Counterfactual Explanation Algorithms." NeurIPS Datasets and Benchmarks.

15. Machine Learning: Foundations and New Frontiers

Supervisor: Vikas Garg

Contact info: vikas.garg@aalto.fi

Number of open positions: multiple

Applications are invited for various internship positions in our group (see e.g., [1-15] below for our contributions to representation learning, graph neural networks, generative models, drug discovery, and climate prediction).  An ideal student would be eager to push the frontiers of science; have strong mathematical, theoretical, statistical, or algorithmic background; and be comfortable with programming in Deep Learning. We particularly invite students with strong Pure or Applied Math, Physics, and Biochemistry backgrounds to apply.  We also value diversity and encourage candidates from underrepresented backgrounds to apply.  Indeed, our group is very diverse – previously, we have hosted interns from around the globe: e.g., universities in US such as MIT, Berkeley, Harvard, Brown; Europe such as Oxford; and Asia such as IIT Bombay. 

Interns in our group from previous years have produced stellar research, resulting in multiple publications [1, 2, 3] at premier conferences including an Oral presentation at NeurIPS.  Here’s an article (https://shorturl.at/9ctHH) if you would like to know more about how it is working with our group.

Projects are flexible/customized to be aligned with each intern - topics of interest may include (but are not limited to):
(1) Generative Models, Neural ODEs/PDEs/SDEs 
(2) Transformers, State Space Models, Large Language Models
(3) (Temporal) Graph Neural Networks, Topological Deep Learning, Topological Data Analysis (e.g., Persistent Homology) 
(4) Differential Geometry/Information Geometry/Algebraic/Spectral Methods for Deep Learning 
(5) Learning under limited data, distributional shift, and/or uncertainty; Conformal Prediction 
(6) (Approximate) Equivariant and Invariant models
(7) Fair, interpretable, or explainable methods
(8) Reinforcement learning, multiagent systems, and AI-assisted human-guided models 
(9) Applications in drug discovery, material design, climate prediction, quantum chemistry, etc. 
(10) Quantum Machine Learning 
 

Representative publications (interns marked in bold): 
[1] Brilliantov et al. Compositional PAC-Bayes: Generalization of GNNs with persistence and beyond, NeurIPS, 2024.

[2] Pham and Garg. What do Graph Neural Networks learn? Insights from Tropical Geometry, NeurIPS, 2024.
[3] Immonen(*), Souza (*), and Garg. Going beyond persistent homology using persistent homology. NeurIPS, 2023 (Oral).  
[4] Kogkalidis, Bernardy, and Garg. Algebraic Positional Encodings, NeurIPS, 2024 (Spotlight). 
[5] Mercatali(*), Verma(*), et al. Diffusion Twigs with Loop Guidance for Conditional Graph Generation, NeurIPS, 2024.
[6] Karczewski, Souza, and Garg. On the Generalization of Equivariant Graph Neural Networks, ICML, 2024.
[7] Verma et al. ClimODE: Climate and Weather Forecasting with Physics-informed Neural ODEs, ICLR, 2024 (Oral). 
[8] Verma, Heinonen, and Garg. AbODE: Ab initio antibody design using conjoined ODEs, ICML, 2023.
[9] Garipov et al. Compositional Sculpting of Iterative Generative Processes, NeurIPS, 2023.
[10] Alvarez-Melis(*), Garg(*), and Kalai(*). Are GANs overkill for NLP?, NeurIPS, 2022 (Spotlight)

[11] Souza, Mesquita, Kaski, and Garg. Provably expressive temporal graph networks, NeurIPS, 2022.

[12] Mercatali, Freitas, and Garg. Symmetry-induced disentanglement on graphs, NeurIPS, 2022. 
[13] Garg, Jegelka, and Jaakkola. Generalization and Representational Limits of Graph Neural Networks, ICML, 2020.
[14] Ingraham, Garg, Barzilay, and Jaakkola. Generative Models for Protein Design, NeurIPS, 2019. 
[15] Garg and Jaakkola. Solving graph compression via Optimal Transport. NeurIPS, 2019. 

16. Distances in geometric (intersection) graphs

Supervisor: Sándor Kisfaludi-Bak

Contact info: sandor.kisfaludi-bak@aalto.fi

Number of open positions: 1

Geometric graphs are graphs where vertices are associated with points in some metric space (typically Euclidean space) and edges are weighted according to the distance of those points in the ambient space. In geometric intersection graphs the vertices are associated with objects (e.g., disks, squares) and edges are unweighted and they correspond to pairwise intersecting objects. In this project we will investigate the metric properties of such graphs: depending on the specific setting, we will be interested in distance computation in the graph, the possibility of finding so-called spanners (subgraphs that represent distances), or low-distortion embeddings of such graphs into other important graph metrics.  

This is a theory project where you will be expected to write short technical report in LaTeX by the end of the internship. Please only apply if you have a strong background in algorithm theory as well as in discrete mathematics or geometry. You need to be highly skilled in writing formal mathematical proofs. It will be an advantage if you have already had some exposure to computational geometry.

17. Legal Network Science

Supervisor: Corinna Coupette

Contact info: corinna.coupette@aalto.fi

Number of open positions: 1–2

As part of a collaboration between the Telos Lab at Aalto University, Finland () and the Center for Digital Law at Singapore Management University (), we invite applications for several projects conducting frontier research in the area of legal network science. Starting from the premise that legal systems are complex systems, legal network science develops and applies network methods to study the structure, function, and dynamics of legal systems [1]. To this end, we work with large amounts of legal data (from treaties, legislation, regulation, and court decisions to patents, company disclosures, administrative permits, and metadata on legal systems), which we often collect and clean ourselves using web-scraping techniques and NLP tools (see, e.g., [2]). Depending on interns’ preferences and skills, topics for summer projects include:

1. Models and methods for multilayer legal networks. (Desirable: Familiarity with multilayer networks.)
2. Models and methods for temporal legal hypergraphs (see [3]).  
   (Desirable: Familiarity with hypergraphs and DAG representations of temporal networks.)
3. Accounting for institutions or procedures in legal networks. 
   (Required: Familiarity with constitutional law or procedural law in at least one jurisdiction.)
4. Integrating legislative, regulatory, and judicial networks. (Desirable: Data-engineering experience.)
5. Building an open research infrastructure for computational legal studies. 
   (Desirable: Data-engineering or software-engineering experience.)

All projects require comfort with computational data analysis and (1) basic knowledge of networks (e.g., from a course in network analysis or complex networks) or (2) basic knowledge of at least one legal system (e.g., from a degree program or a minor); experience in developing models, measures, or methods for data analysis is a plus.

Programming Languages: Python, Julia, R
Libraries/Frameworks: Handling dataframes, networks, databases, visualization, NLP tasks, ML tasks, scraping tasks
Potential Tasks: Literature review; data collection, data preprocessing, and data engineering; network modeling; theoretical analysis; method development and experimental validation; data analysis and data visualization; software development, validation, and testing; collaboration and community engagement

References:

1. Ruhl, JB, Katz, DM, and Bommarito, MJ (2017). Harnessing legal complexity. Science. 
2. Coupette, C., Beckedorf, J., Hartung, D., Bommarito, MJ, and Katz, DM (2021). Measuring law over time. Frontiers in Physics. 
3. Coupette, C., Katz, DM, and Hartung, D. (2024). Legal Hypergraphs. Philosophical Transactions of the Royal Society A. 

18. Computational Perspectives on Democracy

Supervisor: Corinna Coupette

Contact info: corinna.coupette@aalto.fi

Number of open positions: 1–2

We invite applications for several projects developing computational perspectives on democracy. While modern societies and technologies have changed dramatically over the past decades, the structure and procedures of our democratic systems have remained largely unaltered. In this project, we analyze existing decision-making protocols in deliberative assemblies (such as parliaments), investigate their shortcomings, and explore potential strategies to improve democratic processes. Depending on interns’ preferences and skills, topics for summer projects include:

1. Data science for democracy. Analyze political and legal data, e.g., to characterize power dynamics or quantify fairness and transparency in existing policymaking systems [1], and develop methods to study domain-specific text data. (Desirable: Experience handling complex data, esp. from law or political science)
2. Agent-based modeling of policymaking systems. Simulate current policymaking systems and state-of-the-art alternatives to evaluate and compare their performance. 
   (Desirable: Experience with computational modeling and simulation design)
3. Theoretical approaches to democratic processes. Analyze the mathematical and computational properties of existing procedures for collective decision-making [2], especially with regard to their power dynamics, fairness, and transparency, and investigate the properties of promising alternatives. (Desirable: Experience with mathematical modeling and interest in societal applications of theoretical computer science)
4. Application development for experiments on collective decision-making. Contribute to the development of an application to facilitate (gamified) experiments on collective decision-making under different policymaking systems. (Desirable: Experience with application development, HCI, or game design)

Programming Languages: Python (for 1., 2., and 4.), frontend-development software stack (for 4.)
Libraries/Frameworks: Handling dataframes, databases, modeling, simulation, data analysis, data visualization, NLP, data mining, ML, scraping, application development (topic-dependent)
Potential Tasks: Literature review (all); data collection, data preprocessing, and data engineering (1.); computational modeling and simulation (2.); problem formalization and theoretical analysis (3.); requirements engineering and application design (4.); method development and validation (all); data analysis and visualization (1., 2.); software development, validation, and testing (1., 2., 4.); collaboration and community engagement (all)

References:

1. Emerson, P. ed. (2007). Designing an all-inclusive democracy: Consensual voting procedures for use in parliaments, councils and committees. 

Brandt, F. et al. (eds.) (2016). Handbook of Computational Social Choice. 

19. Positions in Probabilistic Machine Learning

Supervisor: Arno Solin (for more info Mohammad Vali, mohammad.vali@aalto.fi)

Contact info: arno.solin@aalto.fi

Number of open positions: 2

We are seeking motivated and talented interns to join our current research projects focused on probabilistic machine learning with positions in tractable modelling, uncertainty quantification in deep learning, and multi-modal (computer vision + language) modelling. This project is part of our broader initiative aiming to advance the frontiers of understanding and develop novel methods in the field of machine learning. More specifically, the research interests are in uncertainty quantification in large-scale machine learning models, as well as combining semantic understanding with scene reconstruction (Gaussian splatting / NeRF models).

Interns will have the opportunity to work on cutting-edge research problems, including uncertainty quantification in neural networks and the development of innovative inference methods. Our team values creativity, analytical skills, and a collaborative spirit. A successful candidate is expected to have knowledge of probabilistic modelling and approximate inference, and general machine learning methods as well as experience with programming in Python (e.g., TensorFlow, JAX, PyTorch, etc.).

This internship presents a unique opportunity to contribute to significant research in a dynamic and supportive environment. We encourage students who are enthusiastic about probabilistic modelling and have a keen interest in language models and computer vision to apply. Highlight your specific skills and interests in your application to align with our team's needs.

See the supervisor's home page for representative publications: https://arno.solin.fi

20. Looking for a master thesis worker interested in human-centered security for older adults

Supervisor: Verena Distler         

Contact info: verena.distler@aalto.fi

Number of open positions: 1

We are looking for an excellent student to start their master thesis in March 2025. This study will be conducted in collaboration with a Finnish organization. You will investigate older adults’ impressions of the security of a new mobile application and various authentication methods. You will learn about qualitative research methods, usability tests and human-centered security. 

Tasks: 

You will conduct a qualitative study (involving interviews, user tests) with older adults. Your tasks include participant recruitment, organizational tasks relating to the study setup, conducting the study with participants, transcribing and analyzing the data, writing the master thesis. 

Requirements:

•      Study will be conducted in Finnish. In addition, excellent English is required.

•      Prior experience with conducting qualitative studies and writing skills are an advantage. 

•      You should be an empathic and patient person who can put themselves in the shoes of their research participants. 

•      Start date (fixed): March 2025

21. Exascale physics simulations with the help of machine learning

Supervisor: Maarit Korpi-Lagg

Contact info: maarit.korpi-lagg@aalto.fi 

Number of open positions: 1

Our research group [1] aims at unveiling the mysteries of the magnetized universe using numerical simulations designed to efficiently run in world’s largest supercomputers. Grid-based numerical solutions to model physical and engineering problems are computationally expensive, requiring large parallel calculations and integrating for many hours or even months. Particularly in astrophysics the problems under investigation can require very high resolution to resolve the turbulent dynamics which are ubiquitous in space, while also requiring massive domains to contain the large structures of interest, such as stars or galaxies, and the whole universe! We are looking for highly motivated individuals who want to combine state-of-the-art computer science skills with exciting application domain science. 

Machine learning (ML) can help to mitigate these challenges: it can be applied to high-fidelity turbulence data to learn models. These can then be applied at coarser resolution to reproduce the small-scale effects below the resolution of the grid, hence called sub-grid-scale (SGS) models. When a successful ML model is found, it would reduce the load on the computation for modelling these large objects, while retaining the essential realistic feedback of the subgrid processes.

We have developed, together with NVAITC professionals [2], a proof-of-concept simulation-in-the-loop high performance computing pipeline to train and validate simple ML approaches to form the SGS model. While the HPC pipeline is now ready to run with top tier supercomputers such as LUMI-G [3] and the forthcoming Roihu [4], the ML techniques still require more work. The goal of the internship is specifically to improve the ML approach. This work is of uttermost importance to enable physics simulations of this kind at exascale. 

References:

[1] Group homepage: /en/department-of-computer-science/astroinformatics

[2] NVAITC homepage: https://fcai.fi/nvaitc

[3] LUMI documentation: https://docs.lumi-supercomputer.eu

[4] Roihu, the forthcoming Finnish supercomputer: 

22. Ohjelmoinnin peruskurssi Y1: harjoitustehtävien suunnittelu

Supervisor: Kerttu Pollari-Malmi 

Contact info: kerttu.pollari-malmi@aalto.fi

Number of open positions: 2

Tehtävänä on keksiä uusia harjoitustehtäväideoita kurssille CS-A1111 Ohjelmoinnin peruskurssi Y1, kirjoittaa tehtävänannot suomeksi ja englanniksi sekä laatia tehtäville automaattiset tarkistimet A+-järjestelmään. Työssä vaaditaan ideointikykyä, hyvää ohjelmointitaitoa sekä erinomaista suomen ja hyvää englannin kielen taitoa. 

23. Massively Parallel Algorithms for Graph Problems

Supervisor: Jara Uitto

Contact info: jara.uitto@aalto.fi

Number of open positions: 1

Parallel processing of data and distributed computing are gaining attention and becoming more and more vital as the data sets and networks we want to process are overgrowing the capacity of single processors. To understand the potential of modern parallel computing platforms, many mathematical models have emerged to study the theoretical foundations of parallel and distributed computing. In this project, we study algorithm design in these models with a particular focus on the Massively Parallel Computing (MPC) and Local Computation Algorithms (LCA) models.

The problems we study are often in (but not limited to) the domain of graphs, that serve as a very flexible representation of data. We are interested in, for example, the computational complexities of classic problems such as finding large independent sets, matchings, flows, clustering problems, etc.

The applicant is assumed to have a solid knowledge of mathematics, knowledge on the basics of graph theory, and a good command of English. No prior knowledge in distributed computing is required, although it might be helpful.
 

24. Interactive slides for CS education

Supervisor: Jaakko Lehtinen

Contact info: jaakko.lehtinen@aalto.fi

Number of open positions: 2

The prevalent paradigms for slide preparation (e.g. PowerPoint) do not readily support embedding interactive content on slides. Yet, interactivity and animation would significantly help conveying concepts over static diagrams and plots — think about aliasing in sampling and reconstruction, order of operations in a sorting algorithm, routing in a network... 

This project explores the use of modern web technologies (e.g. WebGPU, D3.js, reveal.js, ...)  for creating interactive visualizations on talk slides. In particular, we seek to map out (1) approaches that allow inclusion of arbitrary code on presentation slides and enable seamless integration with a presentation framework (such as reveal.js) and the interactive content, (2) good design practices and reusable code components that allow both high productivity in creating the visualizations and support long-term maintenance, (3) options for at least partial WYSIWYG editing of slide layouts. We will work mostly in the context of graphics education, but with an eye for other types of content too.

Prerequisites: excellent programming skills in multiple languages, familiarity with modern web technologies, solid understanding of the foundations of computer graphics (evidenced, for instance, by high performance in CS-C3100 and/or CS-E5520), tendency for independent, critical, and proactive thinking. 

25. Bayesian workflow

Supervisor: Aki Vehtari

Contact info: aki.vehtari@aalto.fi

Number of open positions: 1

You will take part in developing computational diagnostic tools for different parts of Bayesian workflow (see, e.g. https://arxiv.org/abs/2011.01808). Possible more specific topics include model checking diagnostics, cross-validation, better priors, inference diagnostics. Prerequisites: Bayesian inference and MCMC.
 

26. Y1-kurssin tentit ja Y2-kurssin tehtävät

Supervisor: Sanna Suoranta 

Contact info: sanna.suoranta@aalto.fi 

Number of open positions: 2

These positions require a fluent command of written Finnish. Behärskar svenska sprÃ¥ket är en fördel. However, English is also needed.

Tässä tehtävässä tehdään tenttitehtäviä ja niiden automaattitarkistimia kurssille CS-A1111 Ohjelmoinnin peruskurssi Y1 sekä harjoitustehtäviä kurssille CS-A1121 Ohjelmoinnin peruskurssi Y2. Tehtävänantojen miettiminen vaatii luovuutta ja automaattitarkistimien tekeminen vaatii tarkkuutta ja kykyä ajatella monenlaisia ratkaisuja. Tehtävä vaatii ohjelmointitaitoa Python-kielellä. Aiempi kokekemus A+-järjestelmän automaattisten tarkistimien teosta on hyödyllinen. Kurssit ovat tarjolla suomeksi ja englanniksi (ja tentti myös ruotsiksi), joten vähintään suomea ja englantia tulee hallita erinomaisesti. 

27. Security and deceptive design

Supervisor: Sanna Suoranta 

Contact info: sanna.suoranta@aalto.fi 

Number of open positions: 1

Many webservices lure users to choose options that are best interest of the service, not the user. Such user interface techniques are called dark patterns or deceptive design. They may, for example, lead users to give permission to information they do not intended. We are making experiments and reviews how the dark patterns affect information security decisions made by common users. The work requires knowledge of user interface and user experience design, information security, and programming skills (mobile devices and web software).
 

28. AI for Public Services: Co-design & Reducing Legal Uncertainty

Supervisor: Ana Paula Gonzalez Torres, Karolina Drobotowicz, Petri Vuorimaa

Contact info: anapaula.gonzaleztorres@aalto.fi, drobotowicz.karolina@aalto.fi

Number of open positions: 1

With the emerging adoption of algorithmic decision-making and generative AI across societal sectors, there is a need to reduce legal uncertainty in policies and regulations (e.g., the European Union’s Artificial Intelligence Act) and to involve civil society in design and deliberations of such. We want to address these needs in the Civic Agency in AI (CAAI) project, conducted in collaboration with providers of AI-based public services in Finland (read more here: https://crai-cis.aalto.fi/civic-agency-in-ai/). 

We are seeking to hire a summer researcher to join the CAAI team. The concrete research tasks are up for negotiation with the applicants and will include aspects of Human-Computer Interaction (HCI) and regulation/policy. Example tasks we would like to involve a summer researcher in are:

•              Assistance in organizing and facilitation of workshops

•              Literature review 

•              Assistance in conducting interviews 

•              Qualitative and/or quantitative data analysis 

We value high levels of creativity, independent work capabilities and curiosity. We are looking for applicants with a mix of technical and qualitative skills and/or interest in one or more aspects: HCI, AI policy, societal impact of AI, data analytics. 

29. Deep generative modeling and probabilistic machine learning, with applications to biomedicine and dynamical systems

Supervisor: Harri Lähdesmäki

Contact info: harri.lahdesmaki@aalto.fi

Number of open positions: 3

There is an abundance of high-dimensional datasets and dynamical systems around us, but many of these are too complex to be modeled explicitly using classical statistical and computational methods. Examples include large-scale electronic health records, single-cell datasets, protein structure prediction, as well as real-world dynamical systems. Recent machine learning methods that build on latent variable models and deep generative models provide an efficient approach to tackle such complicated modeling problems. We are looking for summer interns to join our research group and collaborate with other group members to develop novel deep generative models and probabilistic machine learning methods for these important applications. 

Summer internship positions offered by our research group cover a variety of machine learning methods, such as 1) variational autoencoders, 2) latent neural ODE and PDE models, 3) high-dimensional Bayesian optimization, 4) Gaussian processes, 5) longitudinal models, among others. Applications include i) large-scale electronic health records, ii) treatment response prediction and monitoring using single-cell datasets, iii) protein interaction predictions, as well as iv) real-world dynamical (physical) systems. Applicant can choose the topic based on his/her own preference to specific machine learning methods and/or applications. Tasks for summer internship can be adapted to fit student's skills and interest. Work can be continued after the summer. 

We expect that applicant has knowledge of deep learning, deep generative models and probabilistic machine learning (e.g. based on Aalto's courses "Deep learning" and "Probabilistic machine learning"). Basic knowledge of single-cell technologies and differential equations is beneficial if a student wants to work on those topics.
 

30. Theory of distributed and parallel computing

Supervisor: Jukka Suomela

Contact info: jukka.suomela@aalto.fi

Number of positions: 1

Our research group "Distributed Algorithms" is looking for a summer intern to help us with our research related to the theoretical foundations of distributed and parallel computing. We expect a good understanding of mathematics (especially in discrete math and graph theory) and algorithms and theoretical computer science. We also often try to outsource our work to computers, so if you have good programming skills and/or some knowledge of e.g. SAT solvers or proof assistants, it is a plus. We have also exciting opportunities for those who are interested in quantum computation in the distributed setting. For more information, see

31. Software developers for our course infrastructure

Supervisor: Jukka Suomela

Contact info: jukka.suomela@aalto.fi

Number of open positions: 1–3

We are hiring summer interns to help us with the development of the computer systems that keep our courses "Computer as a Tool", "Programming Parallel Computers", "Competitive Programming", and "Distributed Algorithms" up and running. We are looking for summer interns who have got strong programming skills. Some prior experience with developing course automation systems is a plus. For more information on these courses, see: https://lapio.cs.aalto.fi https://ppc.cs.aalto.fi https://plus.cs.aalto.fi/cs-e4595/ https://jukkasuomela.fi/da2020/

32. Measuring Shared User Experiences Using Psychological and Psychophysiological Methods

Supervisor: Prof. Robin Welsch

 Contact info: robin.welsch@aalto.fi

Number of open positions: 1

The Engineering Psychology Group is seeking motivated students to support our research on human interaction with technology, focusing on areas such as human-AI interaction, human augmentation, and user experiences in virtual and extended reality (VR/XR) environments. One of our projects explores shared experiences in XR by combining psychological methods with physiological sensors such as electrodermal activity (EDA), electrocardiograms (ECG), and respiration tracking. 

Your tasks will include scheduling and conducting participant studies, documenting research processes, setting up VR experiments, troubleshooting technical issues during testing, and managing data. 

Applicants should have good organisational and communication skills and be fluent in English. While prior experience with physiological sensors or VR setups is not required, an interest in psychology, technology, and learning new tools is essential. This role is particularly suited for bachelor’s students nearing the end of their studies or master’s students, with opportunities to develop skills relevant for thesis work. 

Joining our team offers practical experience in psychological research, advanced methods for studying human behavior, and shared experiences in the field of human-computer interaction. If you’re excited about exploring these topics, we’d love to hear from you!

33. The DNAforge design tool for nucleic acid nanostructures

Supervisor: Pekka Orponen

Contact info: pekka.orponen@aalto.fi

Number of open positions: 1-2

DNAforge (https://dnaforge.org) is a fully automated, user-friendly online design tool for DNA and RNA wireframe nanostructures [1, 2, 3]. To create a nanostructure design, the user simply uploads a 2D or 3D mesh model of the targeted structure and chooses the desired design approach, together with some parameters such as the preferred nanometer scale of the structure. The tool then performs the complex task of creating a system of DNA or RNA strands which, when synthesised, will fold to the target structure in nanoscale.

The tool was launched in Spring 2024 [4] and is attracting increasing attention in the DNA nanotechnology community. The main task in this summer project is to augment the tool with support for some currently topical research directions, such as the design of RNA:DNA hybrid nanostructures [5] and co-transcriptionally folding RNA nanostructures.

The project requires familiarity with basic algorithm design techniques, facility with combinatorial thinking, and good programming skills. Knowledge of biomolecules is not necessary, but familiarity with Javascript (or willingness to learn) is a prerequisite. For further information about our work, please see the research group webpage at https://research.cs.aalto.fi/nc/.

[1] https://en.wikipedia.org/wiki/DNA_nanotechnology

[2] https://doi.org/10.1007/s11047-017-9647-9

[3] https://doi.org/10.1021/acsnano.2c06035

[4] https://doi.org/10.1093/nar/gkae367

[5]

34. Formally Explaining Machine Learning

Supervisor: Jussi Rintanen

Contact info: Jussi.Rintanen@aalto.fi

Number of open positions: 1

As machine learning is increasingly used in the real-world, including in sensitive contexts like healthcare or law enforcement, it is essential that a user is able to inspect the reasons for a certain classification. 

Indeed, the field of explainable AI is receiving increased attention.

Methods based on formal logic (especially boolean satisfiability, SAT) have in recent years been used to offer sound and non-redundant explanations for machine learning models. These methods avoid shortcomings of other explanation methods, which may for example "explain" a decision with the same set of information as it "explains" 

the opposite decision, or offer explanations that are needlessly complicated and thus less understandable (even to the point of uselessness).

In this project, we advance the current state of the art in formally explained machine learning by implementing encodings of machine learning models into target languages (such as SAT). Challenges include e.g. scalability (efficient encodings and techniques are needed), covering different machine learning models, and covering different notions of explanations.

Necessary skills include good programming skills (e.g. python or C++) and some experience with declarative problem solving. Having completed the course CS-E3220 - Declarative Programming with a good grade is considered an advantage.

35. Formal methods for intelligent software systems

Supervisor: Jussi Rintanen

Contact info: Jussi.Rintanen@aalto.fi

Number of open positions: 1

Many AI methods provide solutions to core problems in software engineering. The goal of this project is to develop methods for analysis of software systems and automation of software development tasks. The project looks at software specification languages that view software systems as transition system models, formalized as discrete transitions from one state to the next, expressed as condition-effect rules.

The goal of the project is to develop a tool for reasoning about such transition system models. Some of the main challenges are related to the fact that real-world software system involve complex data, which cannot be directly expressed in low-level languages that support Boolean and numeric types only (such as input languages for SAT solvers, SMT solvers, Integer Prorgamming solvers, and Mixed Integer-Linear Programming solvers.) As a prerequisite for the project is knowledge of basic AI and formal methods technologies, for example as covered in CS-E3220 Declarative Programming and CS-E4800 Artificial Intelligence. Additionally , good programming skills are required.

36. Constraint Solving for Procedural Content Generation

Supervisor: Jussi Rintanen

Contact info: Jussi.Rintanen@aalto.fi

Number of open positions: 1

Procedural content generation (PCG) refers to automatically generating digital objects with some desirable characteristics, often for use in games.

Among various approaches to PCG, using constraint solving (i.e. constraint programming, Boolean satisfiability, answer set programming etc) has the advantage of offering a declarative workflow, i.e. one where the programmer can describe the desired characteristics without specifying how exactly to generate them; the computation is left to the constraint solver.

Declarative PCG has been somewhat explored, but there are challenges, such as scalability, since constraint solving is in general computationally hard.

In this project, we explore the possibilities of declarative PCG, advancing the state of the art along (some of) the following directions: 1) improving the performance of constraint-based PCG by leveraging and/or developing advances in solver technology, 2) combining constraint-based PCG with inexact methods to improve the modelling capabilities of the PCG system while keeping the benefits of declarativeness, 3) applying (declarative) PCG methods in non-traditional domains, such as designing physical objects like areas of a city.

37. Summer internship positions in computer vision and machine learning

Supervisor: Juho Kannala

Contact info: juho.kannala@aalto.fi

Number of open positions: 2

Computer vision is a rapidly developing field that is at the forefront of recent advances in artificial intelligence. Our group has broad research interests within computer vision. We are pursuing problems both in geometric computer vision (including topics such as visual SLAM, visual-inertial odometry, optical flow, image-based 3D modeling and Gaussian splatting) and in semantic computer vision (including topics such as object detection and recognition, and deep learning). We are looking for students interested in both basic research and applications of computer vision. Students with good programming skills and strong background in mathematics are especially encouraged to apply. The precise topics of the research will be chosen together with the students to match their personal interests.

Examples of our recent papers include: https://aaltovision.github.io/PIVO/, https://aaltovision.github.io/pioneer/, https://aaltoml.github.io/GP-MVS/, https://github.com/AaltoVision/DGC-Net, https://github.com/AaltoVision/hscnet. For more papers and further information visit: