Summer employee positions at the Department of Computer Science 2023

The Department of Computer Science is now 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 2023. 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 and doctoral students & postdocs at the department.

Ready to apply?

See the complete list of the available topics below (will be fully updated by January 2^nd)), choose the topic(s) (max. 5) that interest you the most, choose them from the multiple-choice questionnaire on the application form and list them in the order of preference.

Please submit your application through our recruitment system. The application form will open on January 2^nd and close on January 31^st, 2023, at 23:59 Finnish time (UTC +2).  

Link to the application form:

https://www.aalto.fi/en/open-positions/summer-employee-positions-2023-at-the-department-of-computer-science

Are you an international student or coming from abroad?

Please check the Aalto Science Institute AScI internship programme for international summer employees.
https://www.aalto.fi/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 Maaria Ilanko from HR Team. [email protected].

Summer employee topics 2023

1. Placebo effect of AI technology

Supervisor: Assistant professor Robin Welsch    
Email: [email protected]
Number of open positions: 1

Studies evaluating artificial intelligence (AI) technologies could be fundamentally flawed due to the placebo effect, i.e., user's experiencing benefits from belief in the effect of a sham treatment. Human-centered AI technologies are designed to support humans and extend human capabilities, raising high expectations for the improvement of people's lives. This research internship will run empirical studies to investigate how the presentation of an AI system can produce placebo effects, i.e., how AI can facilitate task completion in the absence of a functional system.

The main task in this internship are the design, planning, and execution of an empirical user study. This includes the creation of prototypes, the preparation of study materials but also doing research with users in the lab.

Candidates for the intern position should have the following skills:

• knowledge in quantitative user studies and/or AI
• good knowledge of quantitative data analysis
• strong interest in experimental psychology research

You will gain experience in the following areas

• Artificial intelligence and user’s mental models
• Usability & User Experience
• Conducting User studies
• Application of psychology to human-computer interaction

2. 3D visualisation of and pattern recognition from large-scale data from multi-physics simulations

Supervisor: Associate Professor Maarit Korpi-Lagg    
Email: [email protected]
Number of open positions: 2

Large-scale simulations of, for example, magnetised fluids in stellar interiors produce huge amounts of three dimensional data, where each system state can comprise hundreds of Gigabytes or even
Terabytes. Analysis, visualisation, and even storage of such data is challenging, and special tools are
required. From the visualisation perspective, we are looking for a summer intern, who could develop further our existing Python framework, with which we create 3D visualisations from the simulation data (https://owncloud.gwdg.de/index.php/s/iAq7VQ2Rb71Xfau#/files_mediaviewer/). The task of the summer intern 1 is to enhance the existing toolbox by adding parallel processing capabilities, to better handle multiple snapshots of large datasets for animation.

Prerequisites: Good knowledge in Python, and managing Jupyter notebooks. Some knowledge of supercomputing environments is a bonus.

From the analysis and storage perspective, we need to develop tools that are capable of recognising subregions of interest, and analyse and output data only from these regions, while storing the full system states will no longer be possible in the forthcoming Exa-scale computing era. The long-term aim of the project is to develop an online or offline structure-detector assistant for the large-scale simulation toolbox. The tasks of the summer intern 2 include: To continue developing an existing code based on the FasterRCNN object detection model. The code also includes a data augmentation pipeline, which is necessary for increasing the training data size and diversity; Generating training data for the neural network using idealised simulation setups; use the generated training data for deep learning network; apply the trained network to detect the predefined structures and track their evolution in time from the real simulation data.

Prerequisites: Basic knowledge on ML is required, and being familiar with toolboxes like PyTorch or Tensorflow is an extra benefit.

3. Provable Image and Audio Steganography via Cover Source Switching

Supervisor: Assistant professor Russell W.F. Lai
Email: [email protected]
Number of open positions: 1

Steganography is the study of hiding secret information in innocuous cover objects in such a way that the presence of the information itself is concealed. A common yet unsatisfactory approach is to embed secret information by minimally distorting the cover objects, which results in stego objects with heuristic but non-provable security. An alternative approach, known as cover source switching, relies on the existence of multiple cover sources, e.g., photos with different ISO values, and seeks to generate stego objects whose distribution is close to that of a cover source, e.g., adding noise encoding secret messages to ISO 100 photos to make them look like ISO 200 photos.

We seek to further explore the possibility of provable steganography via the technique of cover source switching using images and/or audio as mediums. Depending on the student's interests, potential tasks include surveying the literature, identifying new cover sources suitable for steganographic purposes, conducting experiments to determine or verify noise models of cover sources, and implementing stego-encoders. For the latter tasks, the student likely needs to write tools to parse and manipulate digital image and/or audio files at bit level.
Strong programming skills and a good command of English are required. No prior knowledge in steganography is needed. Helpful but not necessary are knowledge on digital signal processing.

4. Lattice-based Cryptography

Supervisor: Assistant professor Russell W.F. Lai
Email: [email protected]
Number of open positions: 1

Lattice-based cryptography has emerged as one of the main techniques for public-key cryptography. Besides potential security against both classical and quantum attackers and connections to well-studied computational lattice problems, lattices also provide rich algebraic structures enabling advanced functionalities such as fully homomorphic encryption (FHE), attribute-based encryption (ABE) for circuits, and succinct proof systems (SNARK).

We seek to advance lattice-based cryptography on many fronts. The concrete topic varies depending on the student's interests. Some examples are identifying and analyzing new and existing cryptographic assumptions, constructing cryptographic primitives with new functionalities, and implementing recently proposed constructions.
The applicant is assumed to have mathematical maturity, i.e., feeling comfortable to read and write formal mathematical statements and proofs, and a good command of English. Applicants with prior knowledge in cryptography are prioritised. For implementation projects, strong programming skills are required. Helpful but not necessary are knowledge on lattice theory, number theory, and algebraic number theory.

5. Asynchronous Optimization Methods for Federated Learning

Supervisor: Assistant professor Alex Jung
Email: [email protected]
Number of open positions: 1

One widely used design principle for federated learning methods is total variation (TV) minimization. TV minimization is an instance of regularized empirical risk minimization using total variation as the penalty term. This project studies the behavior of asynchronous optimization methods to solve TV minimization.

Prerequisites
- Broad knowledge of convex optimization and distributed algorithms
- Basic Python coding skills (use of scikit-learn is a plus)

Background Reading:
Dimitri P. Bertsekas and John N. Tsitsiklis,  Parallel and Distributed Computation Numerical Methods, Athena Scientific, 2015

6. Clustered Federated Learning via Total Variation Minimization

Supervisor: Assistant professor Alex Jung
Email: [email protected]
Number of open positions: 1

One widely used design principle for federated learning methods is total variation (TV) minimization. TV minimization is an instance of regularized empirical risk minimization using total variation as the penalty term. This project studies the cluster structure of the resulting local models and their dependence on the heterogeneity of local datasets.

Prerequisites
- Good grip on convex optimization
- Basics of graph theory (e.g., spectral properties of Laplacian matrix)
- Basic Python coding skills (use of scikit-learn is a plus)

Background Reading:
Networked Federated Learning, SarcheshmehPour, Yasmin ; Tian, Yu ; Zhang, Linli ; Jung, Alexander, https://arxiv.org/abs/2105.12769

7. Trustworthy Empirical Risk Minimization

Supervisor: Assistant professor Alex Jung
Email: [email protected]
Number of open positions: 1

A widely used design paradigm for machine learning methods is empirical risk minimization (ERM). ERM involves design choices for the data (representation), model (hypothesis space) and loss function (metric) used by the resulting ML method. From a technical perspective, these design choices are guided by computational and statistical aspects.

However, a recent focus has been on a third aspect which is the trustworthiness of the resulting ML method. This project studies how to ensure trustworthy ML by suitable design choices for data, model and loss.

Prerequisites
- Completed Introductory course on machine learning, such as CS-C3240 Machine Learning or CS-EJ3211 Machine Learning with Python
- Basic Python coding skills (use of scikit-learn is a plus)

Background Reading:
 A. Jung, "Machine Learning: The Basics," Springer, Singapore, 2022
 A. Jung, "Trustworthy ML", lecture notes for the Unite! summer school on Human-Centred Machine Learning, 2022, https://github.com/alexjungaalto/Teaching/blob/master/HCML2022/HCML2022_TrustworthyML.pdf
 High-Level Expert Group on Artificial Intelligence, "ETHICS GUIDELINES FOR TRUSTWORTHY AI", 2019, https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=60419

8. Harjoitustehtävien ja niiden automaattitarkistimien laatiminen kurssille CS-A1111 Ohjelmoinnin peruskurssi Y1

Supervisor: Senior University Lecturer Kerttu Pollari-Malmi
Email: [email protected]
Number of open positions: 2

These positions require a fluent command of written and spoken Finnish.

Tehtävänä on laatia yhdessä toisen kesätyöntekijän kanssa uusia harjoitustehtäviä syksyn CS-A1111 Ohjelmoinnin peruskurssi Y1 -kurssia varten sekä automaattisia tarkistimia näille tehtäville. Lisäksi kesätyöntekijä toimii lisäassistenttina osassa Y1-kesäkurssin harjoitusryhmissä. Tehtävä vaatii ohjelmointitaitoa Pythonilla ja hyvää ideointikykyä. Koska kurssin kieli on suomi, on myös hyvä suomen kielen taito välttämätön.

9. Sampo-portals and linked data services on the Semantic Web

Supervisor: Professor Eero Hyvönen https://seco.cs.aalto.fi/u/eahyvone/
Email: [email protected]
Number of open positions: 1-2

Semantic Computing Research Group (SeCo) (https://seco.cs.aalto.fi) is a leading research group in the field of applications of Linked Data in Digital Humanities. We do research at the Department of Computer Science in Aalto as well as at the University of Helsinki, HELDIG Centre for Digital Humanities. We solicit new student members in our research group, interested in semantic web technologies, full stack web development, and data analyses in the field of Cultural Heritage. The open positions are related to developing new Sampo portals and web infrastructures as described here: https://seco.cs.aalto.fi/applications/sampo/

In particular, we will be focusing on data about the Finnish opera and music performances during the 19th century (“Oopperasampo”, in collaboration with the Sibelius Academy), and on the conferment traditions 1640-2023 of the Academy of Turku and University of Helsinki (“Promootiosampo”, in collaboration with the University of Helsinki).

It is possible earn study credits as part of the work, and also extend the work later into a master’s thesis.

10. Data-efficient learning for Model-based Reinforcement learning

Supervisor: Professor Samuel Kaski
Academic contact person: Shibei Zhu [email protected]
Number of open positions: 1

Reinforcement learning has shown its success in many domains, such as robotics, healthcare, NLP, games, etc. Model-free RL approaches learn policies to solve complex tasks via trial-and-error without prior knowledge. However, this often requires a large amount of experience data sampled from the environment, especially when the state-space is large. In this latter case, it can converge to a suboptimal solution space due to insufficient explorations in the environment.

Model-based reinforcement learning differs from its model-free counterpart by estimating the model of the environments (i.e., transition dynamic) while learning the task. It holds the promise of providing a more data-efficient solution by reducing the learning time using its estimated model.
In this project, you will be using/combining different techniques, such as Imitation Learning, to build the decision-making model and dynamic model for model-based RL to speed up the learning process.

References:
https://arxiv.org/abs/1903.00374
https://www.deepmind.com/blog/agents-that-imagine-and-plan  
https://arxiv.org/abs/1707.06170
https://arxiv.org/abs/1707.06203

11. Benchmark of active learning strategies for human-in-the-loop assisted molecular design

Supervisor: Professor Samuel Kaski
Academic contact person: Yasmine Nahal [email protected]
Number of open positions: 1

Not all knowledge is explicit and currently usable for machine learning modeling in molecular generation and drug design. A new and emerging area in machine learning is knowledge elicitation from human experts to improve the prediction accuracy of machine learning models. Drug discovery projects usually start with a small set of active chemical compounds, thus limiting the real-world usage of standard machine learning techniques resulting in low-quality models. These models are still needed for molecular generation and multi-objective optimization to find better new active compounds. Since predicted active compounds always fall in the hands of medicinal chemists for validation, we are currently investigating how we can benefit from an additional information source as domain experts during the learning process of bioactivity models so we can improve their predictive accuracy at a minimal cost.

In practice, we will use active learning to query the most informative feedback from the expert sequentially then apply probabilistic inference to update the model parameters after observing the feedback. Your task would be to benchmark different active learning and data acquisition strategies to find out the best match with human expertise.

Required for a successful internship:
- A good general understanding of active learning in the Bayesian probabilistic
framework.
- Solid Python programming skills.
- Stan or any other probabilistic programming language is a plus, although not mandatory.

The major outcome of this research project is the development of a general human-in-the-loop methodology that would enable human experts in chemistry to participate more actively in the learning process of machine learning models designed for molecular generation. If you are interested in adding your special contribution to this project, we would be glad to have you onboard!

12. Implementing cutting edge privacy-preserving methods

Supervisor: Professor Samuel Kaski
Academic contact person: Lukas Prediger [email protected]
Number of open positions: 1

Research in privacy-preserving methods for machine learning is rapidly advancing. Recent years have seen many publications of improvements to the originally differentially-private (DP) SGD algorithm that forms the foundation for many relevant applications. We are looking for a student interested in learning about technical details of privacy-preserving machine learning and probabilistic methods to help us keep our software package for privacy-preserving probabilistic programming (https://github.com/DPBayes/d3p) up-to-date and increase its fidelity and security.

Your main task will be to perform a small benchmarking study of the mentioned advances from the recent literature for the DP-SGD algorithm of d3p to determine the most suitable improvements. This will require you to review the recent literature, implement the techniques described there and define a suitable framework for measuring and comparing their performance.

We require you to have solid Python programming skills and basic understanding of probabilistic machine learning. Hands-on experience with the JAX and NumPyro libraries, automated testing / continuous integration as well as familiarity with variational inference is a plus. You will have the opportunity to learn about the state of the art in privacy-preserving machine learning and work closely with our Doctoral Candidates active in that area.

13. Simulation-based inference for computationally expensive models

Supervisor: Professor Samuel Kaski
Academic contact person: Ayush Bharti [email protected]
Number of open positions: 1

Simulation-based inference (SBI) methods are used to fit complex, simulator-based models with intractable likelihood function to data. Such models appear in a various field of science and engineering such as population genetics, radio propagation, and cosmology. However, inference becomes challenging when the simulator is computationally expensive. The sample efficient SBI methods (e.g.  Papamakarios et al. 2019 and Greenberg et al. 2019) assume that evaluating the model anywhere in the parameter space incurs the same cost, and they try to minimize the number of model evaluations. This assumption is quite often not true in practice, with certain regions of the parameter space being computationally more expensive than others.

In this project, you will develop methods that take into account the run-time of the model during inference and apply them to computationally expensive simulators in your preferred applied area. Students with strong background in mathematics and statistics are especially encouraged to apply.

References:
[1] Greenberg, D., Nonnenmacher, M., and Macke, J. (2019). Automatic posterior transformation for likelihood-free inference. In Chaudhuri, K. and Salakhutdinov, R., editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 2404–2414. PMLR
[2] Papamakarios, G., Sterratt, D., and Murray, I. (2019). Sequential neural likelihood: Fast likelihood-free inference with autoregressive flows. In Chaudhuri, K. and Sugiyama, M., editors, Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 837–848. PMLR.

14. Normalizing Flows with probabilistic transformations

Supervisor: Professor Samuel Kaski
Academic contact person: Manuel Haussmann [email protected]
Number of open positions: 1

Normalizing Flows (NF)[1] are a powerful approach for probabilistic modeling. Bayesian Neural Networks (BNNs) extend deterministic neural networks with Bayesian principles. In this project, we aim to extend prior work by Bellagente et al. (2022)[2] who relied on BNNs as the neural mappings used within an NF model. Whereas their work was motivated by and targeted at a specific application within particle physics we will evaluate the applicability of the approach to more generic tasks within generative modeling as well as extend it to a wider class of NF approaches (such as conditional NFs etc.).

Required prior skills are (i) a good understanding of probabilistic machine learning and deep learning, and (ii) hands-on experience in at least one deep learning library (pytorch, jax,...).

_____
[1] Papamakarios et al. (2021): Normalizing Flows for Probabilistic Modeling and Inference
[2] Bellagente et al. (2022): Understanding Event-Generation Networks via Uncertainties

15. Interpretability with probabilistic Generalized Additive Models

Supervisor: Professor Samuel Kaski
Academic contact person: Ti John [email protected]
Number of open positions: 1

Generalized Additive Models (GAMs) are a popular choice for building "glass-box" models that are interpretable, which is crucial in high-stakes decisions such as healthcare [1]. This is an active topic of research, with recent comparisons of accuracy vs trustworthiness of different approaches to building GAMs [2] as well as visualisation tools that allow domain experts to "fix" the model [3].

There has been a recent surge in research around neural additive models [4,5,6,7]. Yet much of this line of work does not actually consider the uncertainty in our predictions. Gaussian processes (GPs) are a natural choice for modelling uncertainty over functional relationships, and recent work [8] proposed an "ANOVA"-decomposition as an additive GP model, though they do not assess their uncertainty estimates. There is no comparison between neural and probabilistic models, and no consideration of model behaviour when data has gaps or correlations between input features.

In this project we will build on these strands of work:
- compare neural and probabilistic additive models in terms of accuracy and uncertainty,
- investigate how they actually perform in realistic cases where there may be gaps in the data and correlations between input features,
- work on improved models that handle such cases gracefully and interpretably.

Necessary skills/prerequisites:
- Hands-on experience with Python and at least one of JAX/PyTorch/TensorFlow (or similar frameworks in other languages such as Julia).
- Good understanding of neural networks and probabilistic machine learning, including Gaussian processes.

[1] Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead (Rudin 2019) https://www.nature.com/articles/s42256-019-0048-x
[2] How Interpretable and Trustworthy are GAMs? (Chang et al. 2021) https://dl.acm.org/doi/abs/10.1145/3447548.3467453
[3] GAM Changer: Editing Generalized Additive Models with Interactive Visualization (Wang et al. 2021) https://arxiv.org/abs/2112.03245
[4] NODE-GAM: Neural Generalized Additive Model for Interpretable Deep Learning (Chang et al., ICLR 2022) https://openreview.net/pdf?id=g8NJR6fCCl8
[5] Neural Basis Models for Interpretability (Radenovic et al., NeurIPS 2022) https://openreview.net/pdf?id=fpfDusqKZF
[6] Scalable Interpretability via Polynomials (Dubey et al., NeurIPS 2022) https://openreview.net/pdf?id=TwuColwZAVj
[7] Sparse Interaction Additive Networks via Feature Interaction Detection and Sparse Selection (Enouen et al., NeurIPS 2022) https://openreview.net/pdf?id=Q6DJ12oQjrp
[8] Additive Gaussian Processes Revisited (Lu et al., ICML 2022) https://arxiv.org/abs/2206.09861

16. Theoretical Computer Science

Contact person: Professor Petteri Kaski
Email: [email protected]
Number of open positions: 7-14
Student level: Bachelor’s or master’s

The Theoretical Computer Science group at Aalto University consists of more than 10 faculty members and their teams pursuing leading-edge research across a broad range of topics in theoretical computer science, including:
  - Algebraic Algorithms, Parameterized Algorithms (Petteri Kaski)
  - Approximation Algorithms, Combinatorial Optimization (Parinya Chalermsook)
  - Computational Geometry (Sándor Kisfaludi-Bak)
  - Cryptography, Security & Complexity (Chris Brzuska and Russell Lai)
  - Distributed and Parallel Computing (Jukka Suomela and Jara Uitto)
  - Natural Computation (Pekka Orponen)
  - Quantum Computing (Alexandru Paler)

We welcome applications from students interested in pursuing research work in theoretical topics in computer science based on your interests and strengths, which you are encouraged to highlight and explore in your cover letter to best match you to our teams.

A successful applicant in general has a strong background in computer science and mathematics as evidenced by excellent academic performance and possible other merits such as prior research experience and success in competitive activities. Good programming skills are a further asset as many of our research activities involve using computers to gain insight and drive our understanding of the mathematics of computation.

See https://research.cs.aalto.fi/theory/ for more information about the group and faculty members.

17. Learning Deep Tractable Models

Supervisor: Assistant Professor Arno Solin
Academic contact person: Martin Trapp
Email: [email protected]
Number of open positions: 1

Deep generative models have gained increasing attention within and outside the machine learning research community. For example, recent techniques can simulate human-like interactions with users or generate high-quality images. However, these approaches fall short in general-purpose applications as they do not allow probabilistic inference to be computed efficiently. Probabilistic circuits (PCs) are a unifying computational framework based on sparsely structured deep neural networks to represent tractable probability distributions. Consequently, PCs allow us to efficiently and exactly answer probabilistic queries like the one mentioned earlier. This is made possible by structural constraints enforced on the sparse deep neural network. Therefore, PCs are increasingly used in general-purpose applications in which probabilistic inference is key. Examples include enforcing algorithmic fairness, anomaly detection, and predictions under missing or noisy data.

This project will aim to contribute to the PC community by investigating some of the pressing research questions in the field. Possible topics include data-driven structure learning with partial prior knowledge, Bayesian parameter and structure learning, and approximate inference with PCs. The concrete topic will be decided together with the student.

The successful student should have good math and programming skills (preferably in Julia or Python) and experience with related machine-learning topics. Prior knowledge in probabilistic circuits and Bayesian learning is a plus but not required. The project will be carried out in person in Helsinki together with members of the research group.

Recommended further readings:
[1] Trapp et al. (2019) Bayesian Learning of Sum-Product Networks. NeurIPS. Pre-print available at: https://arxiv.org/abs/1905.10884 [2] Choi et al. (2020). Probabilistic Circuits: A Unifying Framework for Tractable Probabilistic Models. Pre-print available at: http://starai.cs.ucla.edu/papers/ProbCirc20.pdf

18. Deep Model-Based Reinforcement Learning Under Uncertainty

Supervisor: Assistant Professor Arno Solin
Academic contact person: Martin Trapp
Email: [email protected]
Number of open positions: 1

Model-based reinforcement learning (RL) methods have the potential to be more sample efficient than their model-free counterparts. However, learning a dynamics model from high-dimensional observations, such as images, is challenging. To this end, recent work has proposed to learn dynamics models in low-dimensional (latent) spaces, by simultaneously learning an encoder which maps the high-dimensional observations to the latent space. Typically, these methods do not quantify their uncertainty, so RL agents are not aware of what they do not know. That is, they are not aware when their models are not able to predict accurately, due to being far away from their training data.

In this project, we will improve deep model-based RL by enhancing it with state-of-the-art uncertainty quantification. The selected candidate can choose from two exciting research ideas: 1. Improving uncertainty quantification in the encoder by combining recent advances in variational auto-encoders (VAE) with a deep model-based RL. 2. Developing latent-space dynamics models that principally handle uncertainty by leveraging Gaussian process (GP) priors.

A successful candidate is expected to have knowledge of probabilistic modelling and approximate inference, machine learning, and reinforcement learning, as well as experience with programming in Python (e.g., TensorFlow, JAX, PyTorch, etc.)

19. Deep Representation Learning, Reinforcement Learning, Generative Models, Quantum ML, or UI/UX

Supervisor: Assistant Professor Vikas Garg
Email: [email protected]
Number of open positions: Around 5

Applications are invited for various internship positions in our group, widely known for its contributions to representation learning, generative models, and multiagent systems [1-10].  An ideal student would be eager to push the frontiers of science; have strong mathematical, theoretical, statistical, or algorithmic background; and be comfortable programming in a deep learning library (e.g., PyTorch).

Topics of particular interest include but are not limited to: 3D Generative Models;  Graph Neural Networks;  Neural ODEs/PDEs/SDEs, Deep Equilibrium Models, Implicit Models;  Differential Geometry/Information Geometry/Algebraic Methods for Deep Learning;  Learning under limited data, distributional shift, and/or uncertainty;  Bayesian Neural Networks, Causal Modeling, Probabilistic Graphical Models;  Fair, diverse, and interpretable representations;  Off-policy reinforcement learning, inverse reinforcement learning, and causal reinforcement learning; (9) Multiagent systems and AI-assisted human-guided models;  Compression and learning on the edge (i.e., resource constrained settings such as IoT devices); Applications in NLP, drug discovery, material design, synthetic biology, quantum chemistry, etc.; Quantum Machine Learning.

Representative publications:
(1) J. Ingraham, V. Garg, R. Barzilay, and T. Jaakkola. Generative Models for Protein Design. NeurIPS (2019).
(2) V. Garg, S. Jegelka, and T. Jaakkola. Generalization and Representational Limits of Graph Neural Networks. ICML (2020).
(3) V. Garg and T. Jaakkola. Solving graph compression via Optimal Transport. NeurIPS (2019).
(4) V. Garg, L. Xiao, and O. Dekel. Learning small predictors. NeurIPS (2018).
(5) V. Garg and T. Jaakkola. Predicting Deliberative Outcomes. ICML (2020).
(6)  Y. Verma, S. Kaski, M. Heinonen, and V. Garg. Modular Flows: Differential Molecular Generation. NeurIPS (2022).
(7) A. Souza, D. Mesquita, S. Kaski, and V. Garg. Provably expressive temporal graph networks. NeurIPS (2022).
(8) G. Mercatali, A. Freitas, and V. Garg. Symmetry induced disentanglement on graphs. NeurIPS (2022).
(9) D. Alvarez-Melis(*), V. Garg(*), and A. Kalai(*). Are GANs overkill for NLP? NeurIPS (2022).
(10) V. Garg and T. Pichkhadze. Online Markov Decoding: Lower Bounds and Near-Optimal Approximation Algorithms.  NeurIPS (2019).

20. Massively Parallel Algorithms for Graph Problems

Supervisor: Assistant Professor Jara Uitto
Email: [email protected]
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.

21. Machine Learning for Health (ML4H)

Supervisor: Associate Professor Pekka Marttinen
Email: [email protected]
Number of open positions: 1-2

Comprehensive health data has enabled researchers to address questions such as: how to accurately predict the risk of disease, how to personalize treatments based on real-time data from wearable devices, or how to use genomic data to understand disease or antibiotic resistance. Central challenges in ML4H include noisy data, multiple heterogeneous data sources including images and text, learning about causality, interpreting the models, and quantifying the uncertainty, to name a few. We tackle these by developing models and algorithms which leverage modern machine learning principles: Bayesian machine learning, deep latent variable models, Gaussian processes, transformers, reinforcement learning, and natural language processing.

We are looking for summer interns with an outstanding study record in computer science, data science, statistics, applied mathematics, or a related field, and a passion to put these skills to use in interdisciplinary research to address some of the most burning challenges in today’s society.

22. Creative Decision-Making in Artificial Agents – Bridging Frameworks

Supervisor: Assistant Professor Christian Guckelsberger
Email: [email protected]
Number of open positions: 1

AI research has brought forward powerful algorithms to drive the decision-making of artificial agents. Many of these have been put to the test in the domain of games, e.g. in Go or Starcraft II. Especially here, the agents’ behaviour has often been recognised as “creative”, though without any conceptual grounding. The lack of such grounding makes it hard for us to identify (different types of) creativity in AI decision-making, and to develop artificial agents that can leverage creative decision-making to their benefit. This project aims to support the conceptual grounding of creativity in mainstream AI research by establishing formal mappings between AI frameworks of decision-making and theories of creativity.

To this end, the student should have (i) some prior experience in reinforcement learning (e.g. from one related course); (ii) solid skills in mathematics (especially basic set- and probability theory); and (iii) the willingness to deeply engage with literature beyond Computer Science, especially from Creativity Research and Cognitive Science.

23. Deep generative modeling for precision medicine

Supervisor: Associate Professor Harri Lähdesmäki
Email: [email protected]
Number of open positions: 1

We are looking for summer interns to develop novel probabilistic machine learning methods for large-scale health datasets from biobanks and clinical trials. This project aims to develop novel deep generative modeling methods to (i) predict adverse drug effects or other end points using longitudinal/time-series data from large-scale biobanks and clinical trials, and to (ii) harmonize large-scale health data sets for AI-assisted decision making to revolutionize future clinical trials. Methodologically this project can include e.g. VAEs, GANs, Bayesian NNs, domain adaptation, Gaussian processes and causal analysis. Experience/Studies on (probabilistic) machine learning is expected. Tasks for summer internship can be adapted to fit student's skills. The work will be done in collaboration with other research groups from the Finnish Center for Artificial Intelligence, and the novel methods will be tested using unique real-world data sets from our collaborators in university hospitals and big pharma company. Work can be continued after the summer.

Our selected recent work:
[1] http://proceedings.mlr.press/v130/ramchandran21b.html
[2] http://proceedings.mlr.press/v130/ramchandran21a.html
[3] https://arxiv.org/abs/2203.01218
[4] https://arxiv.org/abs/2111.02019
[5] https://academic.oup.com/bioinformatics/article/37/13/1860/6104850

24. Deep generative modeling for dynamical systems

Supervisor: Associate Professor Harri Lähdesmäki
Email: [email protected]
Number of open positions: 1

Recent machine learning breakthroughs include black-box modeling methods for differential equations, such as Gaussian process ODEs [1], neural ODEs, neural PDEs [2], etc. These methods are particularly useful in learning arbitrary continuous-time dynamics from data, either directly in the data space [1,3] or in a latent space in case of very high-dimensional data [4,5]. We are looking for summer interns to join our current efforts to (i) develop efficient yet calibrated Bayesian methods to learn such black-box differential equation models directly from data (e.g. in robotics, biology, physics or video applications) using a low-dimensional latent space representation, and (ii) to further developing these methods for reinforcement learning [6] and causal analysis e.g. in health applications. Methodologically this project can include e.g. VAEs, neural ODEs, variational inference, reinforcement learning, and causal analysis. Experience/Studies in (probabilistic) machine learning is expected. Tasks for summer internship can be adapted to fit student's skills. Work can be continued after the summer.

Our selected recent work:
[1] http://proceedings.mlr.press/v80/heinonen18a.html
[2] https://openreview.net/forum?id=aUX5Plaq7Oy
[3] https://arxiv.org/abs/2106.10905
[4] https://papers.nips.cc/paper/9497-ode2vae-deep-generative-second-order-odes-with-bayesian-neural-networks
[5] https://arxiv.org/abs/2210.03466
[6] https://proceedings.mlr.press/v139/yildiz21a.html

25. Deep learning for protein interaction and function prediction

Supervisor: Associate Professor Harri Lähdesmäki
Email: [email protected]
Number of open positions: 1

Deep learning provides important tools to analyze, predict and even design properties for proteins or their interactions. Applications of these methods are numerous e.g. in health and synthetic biology. Our group has recently developed new deep learning methods to (i) predict which T cell receptors (TCRs) recognize (and later destroy) various pathogens, such as SARS-CoV2 virus and melanoma cells [1,2,3], as well as to (ii) predict signal peptides that facilitate the translocation of proteins [4]. We are looking for summer interns to further develop deep learning methods e.g. to improve TCR-epitope prediction accuracy, to use patients' TCR repertoire for diagnostic purposes, to predict immunogenic/disease-associated peptides, and to design protein properties, to name a few. Methodologically this project involves various neural network architechtures (attention, GCN), language models and probabilistic machine learning. Experience/Studies in (probabilistic) machine learning is expected. Tasks for summer internship can be adapted to fit student's skills. Work can be continued after the summer.

Our selected recent work:
[1] https://academic.oup.com/bioinformatics/advance-article/doi/10.1093/bioinformatics/btac788/6881078
[2] https://www.nature.com/articles/s41467-022-33720-z
[3] https://www.aalto.fi/en/news/an-ai-model-reveals-how-the-bodys-defence-system-recognises-skin-cancer
[4] https://www.biorxiv.org/content/10.1101/2022.06.02.493958v1

26. Kesätyöntekijöitä ja pääassareita Ohjelmointi 1 -kurssille

Supervisor: Senior University Lecturer Juha Sorva
Email: [email protected]
Number of open positions: Approximately 2

These positions require a reasonably fluent command of written and spoken Finnish.

Ohjelmointi 1:n (O1:n) pääassaritiimi uudistuu osittain syksyksi 2023; tarjolla on hommia uusille tekijöille. Vähimmäisvaatimuksena on kurssin asioiden hyvä osaaminen, vastuullisuus sekä innokkuus panostaa laadukkaaseen opetukseen ja opetella uutta tarpeen mukaan.

Kesällä 2023 pääassistentit osallistuvat kurssin kehittämiseen koko- tai osa-aikaisesti. Työ jatkuu syyslukukaudelle, jolloin sille tulee varata noin 50 % viikosta ja tehtävissä painottuvat opettaminen ja kurssijärjestelyt. Aluksi tarjoamme työsopimuksen kalenterivuoden loppuun saakka, mutta oikein mielellämme jatkamme pestejä pidemmiksikin.

Päivä- ja viikkotasolla työajat voidaan sopia joustavasti. Työn aloittaminen osa-aikaisena jo keväällä on myös mahdollista.

Kurssikehityssuunnitelmat syksyksi 2023 ovat joustavat, ja kehitystehtävät voidaan sovittaa kunkin pääassarin omiin toiveisiin ja osaamiseen. Ne voivat olla ohjelmointia, oppimateriaalin laatimista (esim. työkaluvideot), automaattisen arvioinnin ja muiden verkko-opetustyökalujen konfigurointia yms. Työssä voi oppia uusia taitoja ja teknologioita, ja opinnäyteaiheitakin voimme keksiä, jos se puoli kiinnostaa.

O1-kurssi on esitelty oppimateriaalin ensimmäisessä luvussa:
https://plus.cs.aalto.fi/o1/2022/w01/ch01/. Lisätietoja saa vastuuopettajilta.

27. Ohjelmointikurssien oppimateriaalin kehittäminen / Programming course resource development

Supervisor: Professor Lauri Malmi and University Lecturer Otto Seppälä
Email: [email protected]
Number of open positions: 3-4

Kesäharjoittelijoiden tehtävät keskittyvät uusien oppimisresurssien kehittämiseen kursseille Ohjelmointistudio 1, Ohjelmointistudio 2 ja Ohjelmointistudio A. Tämä voi sisältää monenlaisen oppimateriaalien kehittämistä, kuten uusia ohjelmointiharjoituksia testitapauksineen, uusia esimerkkiohjelmia, visualisointeja kurssin sisältöön liittyen, uusia projektiaiheita jne.

Kesäharjoittelijat työskentelevät tiiminä ja tehtävät sovitaan tiimin jäsenten kanssa heidän taustansa, kiinnostuksen kohteiden ja kohdekurssien tarpeiden perusteella.

Edellytykset: Hyvä Scala-ohjelmoinnin taito, kiinnostus kehittää oppimisresursseja, joista on apua junioriopiskelijoille.

The summer trainee positions focus on developing new learning resources for the courses Programming Studio 1, Programming Studio 2 and Programming Studio A. This can include many different types of learning content, such as, new programming exercises with test cases, implementing new example programs, visualisations for course topics, new project topics etc.

Summer trainees will work as a team, and the tasks will be agreed on with the team members based on their background, interests and needs of the target courses.

Prerequisites:  Good command of Scala programming, interest in developing learning resources that would be helpful for junior students.

28. Lockdown on the Web: The unequal impact of the COVID19 pandemic on people’s web browsing behaviour

Supervisor: Assistant Professor Juhi Kulshrestha
Email: [email protected]
Number of open positions: 1

The COVID-19 pandemic and the challenges resulting from associated lockdowns have introduced social, psychological, and economic hardships into people's lives and have disproportionately harmed vulnerable social groups, exacerbating preexisting inequalities. However, it is yet unclear how the pandemic has altered people's information-seeking behaviour on the Web---an essential part of our lives today. We still only have a limited understanding of how these changes might have enduring effects on people's well-being, especially for socially disadvantaged groups. This project will quantify and understand how the lockdowns in the physical space changed people's browsing behaviour on the Web and investigate whether these behavioural changes persist after the lockdown(s) were lifted.

First, we will statistically characterise individuals' web browsing traces before, during, and after lockdowns, using individual-level data. Then, we will focus our analyses of behavioural changes of different groups of individuals based on: (i) socio-demographic characteristics (e.g., gender, family status, economic status, size of place of residence), and (ii) self-reported experience of the lockdown.

Required skills: good python programming skills and interest in the topic

Desired skills: familiarity with data science methods and statistical models

29. Information overload: How much is too much? Examining inequities in information processing capacities of individuals on the Web

Supervisor: Assistant Professor Juhi Kulshrestha
Email: [email protected]
Number of open positions: 1

How much information can we absorb before it becomes a burden? Can web browsing data help answer this question? While the vast amount of information available online aids people in their everyday decisions, from researching health issues to deciding who to vote for, too much of it can paradoxically cognitively overwhelm individuals and impair decision-making. Though information overload negatively affects our decisions on varied matters, it is still unclear how this phenomenon emerges on the Web and intertwines with our online browsing behavior. In this project, we take steps in these directions by answering questions such as, how limited is the information processing capacity of individuals online? How does this capacity vary across individuals? How do browsing contexts affect this capacity?

Required skills: good python programming skills and interest in the topic

Desired skills: familiarity with data science or complex systems methods

30. NLP methods of characterising news articles to study polarization

Supervisor: Assistant Professor Juhi Kulshrestha
Email: [email protected]
Number of open positions: 2

Within a project focused on studying political polarisation in individualised online environments, we aim to analyse news articles to quantify constructs such as topics, sentiments, named entities, stance, sensationalism and linguistic complexity.

Required skills: good python/R programming skills, experience in machine learning and deep learning methods, familiarity with NLP

Desired skills: knowledge of German language is beneficial since the articles are in German

31. Online information consumption and mental health

Supervisor: Assistant Professor Juhi Kulshrestha & Talayeh Aledavood            
Email: [email protected]
Number of open positions: 1

In this project, we will investigate how individuals’ mental health states relates with the health of their information diets that they consume on the Web. We will combine passive web navigation traces with repeated surveys for this project.

Required skills: good python programming skills and interest in the topic

Desired skills: familiarity with data science methods, and complex systems methods or statistical models

32. Designing for Developer Experience

Supervisor: Assistant Professor Fabian Fagerholm
Email: [email protected]
Number of open positions: 1-2

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.

In this project, you will contribute to one or more of the following themes:
- Software developers' mental models of modern programming frameworks
- Developer experience of low-code development platforms
- Longitudinal measurement of developer experience

You could take either a more research-oriented or a more technical focus. In a research-oriented position, you participate in planning and conducting empirical studies with software developers, collect data through interviews, observation, or instrumentation, analyse data using qualitative and quantitative methods, and review existing scientific literature. In a technically oriented position, you participate in developing software for developer experience measurement. The position can be combined with a Master's thesis if you are enrolled at Aalto University.

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).

In addition, we particularly appreciate skills and/or interest in:
- Understanding of research instrument development (e.g., questionnaire design).
- Visual design skills for web and/or mobile.
- Academic writing skills (in English).

33. Methods and models for Continuous Experimentation

Supervisor: Assistant Professor Fabian Fagerholm
Email: [email protected]
Number of open positions: 1-2

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. Open research questions revolve around how to choose what to test, how to produce representative experiment objects to use in the tests, and how to make decisions based on experiment results. We develop methods and models for supporting CE in various kinds of organisations.

In this project, you would contribute to ongoing research to support development of methods and models for organising CE as well as automated design and decision support. 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.
- Basic knowledge of software product requirements engineering, user-centred design, user studies, or similar.
- Basic knowledge of the theory of experimental design.
- Ability to read and summarise scientific literature.
- Academic writing skills (in English).

Desired skills (one or more):
- Conceptual modelling (e.g., using UML or any other modelling technique).
- Understanding and familiarity with statistical modelling such as regression or multilevel modelling, or Bayesian networks.

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.

34. DNAforge: Design tools for DNA nanotechnology

Supervisor: Professor Pekka Orponen
Email: [email protected]
Number of open positions: 1-2

The area of DNA nanotechnology, or more broadly nucleic acid nanotechnology [1] employs the nucleic acids DNA and RNA as generic building material for assembling nanoscale objects with dimensions in the order of 10-100 nanometres. In this area, our group is one of the world leaders in developing automated design tools for general 3D wireframe structures folded from DNA [2] or RNA [3].

The summer internships are part of a broader initiative to develop a distribution platform “DNAforge” for these tools that will bring them together in a common framework and provide the research community with an open, extendible, browser-based interface to access this powerful design methodology. In later stages of the project the platform will also be integrated to the oxDNA molecular dynamics simulation engine [4], so that the designed nanostructures can be directly exported for simulation. A preliminary version of the platform is already available, and new complements to it will be added during Spring 2023. The details of the summer internships will depend on the topical needs of the project and the competences and personal interests of the available developers.

The project requires familiarity with basic algorithm design techniques, facility with combinatorial thinking, and good programming skills. Previous knowledge of biomolecules is not necessary, although it is an asset. For further information about our work, please see the research group webpage at http://research.cs.aalto.fi/nc/.

[1] https://en.wikipedia.org/wiki/DNA_nanotechnology
[2] Benson et al., Nature 2015, https://doi.org/10.1038/nature14586
[3] Elonen et al., ACS Nano 2022, https://doi.org/10.1021/acsnano.2c06035
[4] https://dna.physics.ox.ac.uk/

35. Computerised assignments for course CS-C2160 Theory of Computation

Supervisor: Professor Pekka Orponen
Email: [email protected]
Number of open positions: 1-2

An essential learning tool on the course CS-C2160 Theory of Computation have for several years been the "Astra" computerised assignments, which according to course feedback are also highly appreciated by the students. This software provides each student a personalised set of basic design tasks that cover the key course topics of finite automata, regular expressions and context-free grammars. Solutions are submitted via a graphical user interface and validated by automated checkers that also provide targeted feedback to help improve incorrect solutions.

This currently very stable and well-functioning software is however now in need of two types of revisions: (1) the Astra platform will be deprecated some time in the near future, and the software needs to be migrated to the A+ platform widely used on other courses at the CS department; and (2) new problems and their checkers should be developed to cover also the more advanced topics on the course, in particular Turing machines but possibly also pushdown automata and/or general grammars.

The project requires familiarity with basic algorithm design techniques, facility with combinatorial thinking, and good programming skills. The developers should be comfortable with the material covered on the course and ideally also familiar with the A+ platform. Some practical experience on software development beyond basic programming is an asset (particularly for topic (1)), as well as some experience in GUI design (particularly for topic (2)).

36. Developing a virtual reality environment to study mental rotation in humans

Supervisor: Assistant Professor Stephane Deny
Email: [email protected]
Number of open positions: 1

Topic: Mental rotation is the ability that humans have to imagine objects in various poses, allowing them to recognise and compare objects in their everyday life. It is still unknown what algorithms are employed by the brain to perform mental rotation. This project consists in building a virtual reality environment (using the Unity3D framework) in order to study how humans perform mental rotation of 3D objects. Part of the project will consist in developing the 3D environment and part of the project will consist in designing and running pilot experiments to study mental rotation.

About the lab:  The 'Bidirectional Research in AI and Neuroscience' (BRAIN) lab is joint between the Department of Neuroscience and Biomedical Engineering and the Department of Computer Science at Aalto University. The objective of the lab is to deliver fundamental advances in our understanding of the brain algorithms and inspire the next generation of models for AI.

Preferred skills: Some experience in programming. Some interest for the topic.

37. Conformal Bayesian computation with Pareto smoothed importance sampling

Supervisor: Associate Professor Aki Vehtari
Email: [email protected]
Number of open positions: 1

Conformal inference can provide better calibrated predictive intervals in case of model misspecification. In case of Bayesian models and Markov chain Monte Carlo methods, add-one-observation-in importance sampling can be used for efficient conformal inference. The project involves a short review of the method, implementation of the method in Stan probabilistic programming ecosystem, and making experiments and a case study to illustrate the usefulness of the approach for model checking.

Prerequisites: Bayesian inference, Monte Carlo, and R.

38. Central composite design integration

Supervisor: Associate Professor Aki Vehtari
Email: [email protected]
Number of open positions: 1

Central composite design integration is a deterministic quadrature integration method useful for low dimensional posteriors with costly log posterior density evaluations. A typical use case is when the latent variables in a latent Gaussian variable model are integrated out using Laplace's method. If the number of the latent values is high, running the Laplace's method makes the computation of the lower dimensional marginal posterior slow. However, this marginal posterior has often nice shape, and central composite design integration can provide reasonable accuracy with much less evaluations than, e.g. MCMC. The project involves implementing CCD algorithm and diagnostics in R.

Prerequisites are Bayesian inference, Monte Carlo, and R.

39. Bayesian workflows for safe iterative model building

Supervisor: Associate Professor Aki Vehtari
Email: [email protected]
Number of open positions: 1

Statistical analysis is critical when it comes to obtaining insights from data. Despite the practical success of iterative Bayesian statistical model building, it has been criticized to violate pure Bayesian theory and that we may end up with a different model had the data come out differently. In this project, you will participate in developing methods and diagnostics for iterative Bayesian model building. The practical workflow recommendations and diagnostics guide the modeller through the appropriate steps to ensure safe iterative model building or indicate when the modeller is likely to be in the danger zone. Sometimes the user wants also to build and compare models of different complexity or based on different assumptions.

40. Robust and efficient multivariate model comparison

Supervisor: Associate Professor Aki Vehtari
Academic contact person for the topic: Dr. David Kohns
Email: [email protected]
Number of open positions: 1

The concepts of Bayesian prediction, model comparison, and model selection have developed significantly over the last decade. As a result, the Bayesian community has witnessed a rapid growth in theoretical and applied contributions to building and selecting predictive models. The projection predictive model selection framework in particular has shown promise to this end, and has recently been applied to medical statistics and deep learning. Indeed, it is understood to be less prone to over-fitting than naïve selection based purely on cross-validated performance metrics. Over the course of this project, we endeavor to extend of projection predictive inference to multivariate models, and present a fast and stable method of approximate cross-validation for such multivariate models. Use cases include vector-autoregressive (VAR) models and multivariate linear models more broadly. Students can expect their work to be an integral part of an academic publication.

Pre-requisite: Bayesian inference, R or Python.

41. Practical automatic reparametrizations for Bayesian inference

Supervisor: Associate Professor Aki Vehtari
Academic contact person for the topic: Dr. Nikolas Siccha
Email: [email protected]
Number of open positions: 1

Probabilistic programming languages and packages such as Stan, PyMC, Turing.jl and brms have done a lot to make Bayesian inference more accessible to applied researchers. However, there are still several roadblocks to more "automatic" reliable Bayesian inference for general models, such as multilevel hierarchical models or discretized Gaussian process models. We aim to remove one of the roadblocks by experimenting with, testing and implementing automatic reparametrizations for a subset of generalized non-linear multivariate multilevel models in the popular brms package. Your contributions would help make Bayesian inference easier, more reliable, and potentially faster for a wide range of applied researchers.

Prerequisites: Bayesian inference and MCMC, R+Stan knowledge, brms knowledge beneficial.

42. Summer internship positions in computer vision and machine learning

Supervisor: Associate Professor Juho Kannala
Email: [email protected]
Number of open positions: 3

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 localization) 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/dgc-net-site/
https://aaltoml.github.io/GP-MVS/
https://github.com/SpectacularAI/HybVIO
https://aaltovision.github.io/hscnet/
For more papers and further information visit: https://users.aalto.fi/~kannalj1/

43. Trust-M: Designing Trustworthy Conversational AI Services for Migrants

Supervisor: Professor of Practice Nitin Sawhney
Academic contact person for the topic: Lucy Truong
Email: [email protected]
Number of open positions: 1-2

The Trust-M research project aims to improve the integration of migrants in Finland by devising hybrid and trustworthy digital services based on conversational AI. Finnish public services may not always be accessible, inclusive or trustworthy for all migrants. Improving such services can strengthen social cohesion, resilience of the labor market, and economic vibrance in Finnish Society. The project is a partnership between Aalto University, University of Helsinki, Tampere University, and City of Espoo, supported by the Academy of Finland’s Strategic Research Council (SRC) program in Security and Trust in the Age of Algorithms (SHIELD).

Project objectives include: (1) understanding how the socially and culturally constructed notions of trust, inclusion and equality are manifest in present-day digital public sector services, (2) devising alternatives for novel digital public sector services that could nurture trust and respect human rights, particularly considering migrant women, and (3) designing pilot versions of hybrid digital services based on conversational interaction, in conjunction with the City of Espoo.

We are seeking motivated researchers to join the Trust-M team to conduct research and design of novel trustworthy conversational AI systems using multimodal voice-based interaction. Candidates must ideally have interests and expertise in at least 2-3 relevant areas including Human Computer Interaction (HCI), Natural Language Processing (NLP), conversational AI chatbots, speech/voice interaction, rapid prototyping, design research, user evaluation, and ethical/responsible AI. Evidence of prior work or publications in one or more of these areas is highly beneficial. Good interpersonal skills, collaborative research, conducting ethical research studies or participatory design with end users is helpful. Diverse international candidates with multi-lingual backgrounds are encouraged to apply.

You would join the CRAI-CIS research group in the Computer Science department at Aalto University. The transdisciplinary group explores the impact of technology in critical societal contexts, working at the intersection of computational and social sciences engaging HCI and participatory design. More here: https://crai-cis.aalto.fi

44. Countering AI-infused Disinformation in the Finnish News Ecosystem

Supervisor: Professor of Practice Nitin Sawhney
Academic contact person for the topic: Henna Paakki
Email: [email protected]
Number of open positions: 1-2

The research aims to examine the increasing emergence of AI-infused disinformation and the challenges faced by news media practitioners and fact-checking organizations. It seeks to devise computational tools, social processes, and cooperative practices that can be used to counter disinformation among the Finnish news ecosystem. While it’s clearly a difficult and rapidly changing problem domain, we hope this pilot research will allow us to engage with news media partners in Finland to explore novel socio-technical approaches for information resilience.

We are seeking motivated researchers to conduct research on computational methods for identification of linguistic cues to pre-emptively identify disinformation in news content. You will also fine-tune existing solutions for use with Finnish language content with an aim to develop a
prototype (set of tools and methods) that would learn from journalists and media experts who use the tools to annotate and identify disinformation-related content. We will utilize state-of-the art NLP methods including pre-trained large language models for Finnish (e.g. FinBERT), methods that allow training or enhancing models even with very little resources (e.g. using few-shot learning), and smart data selection for human annotation. The qualitative and computational insights from this research has the potential to increase the preparedness of news media practitioners to address the liquid character of disinformation in the future.

You would join the CRAI-CIS research group in the Computer Science department at Aalto University. The transdisciplinary group explores the impact of technology in critical societal contexts, working at the intersection of computational and social sciences engaging HCI and participatory design. More here: https://crai-cis.aalto.fi

45. Civic Agency in AI (CAAI): Examining Responsible Practices & Critical Discourses for Public Sector AI

Supervisor: Professor of Practice Nitin Sawhney
Academic contact person for the topic: Karolina Drobotowicz [email protected]
Number of open positions: 1-2

Algorithmic tools are increasingly being incorporated into public sector services in cities today. The CAAI project aims to understand citizens’ algorithmic literacy, agency and participation in the design and development of AI services in the Finnish public sector in order to advance more democratic and citizen-centric digital infrastructures. This new project has the following research objectives: 1) understanding the values, narratives and discourses embedded in public sector data-centric and algorithmic services, 2) understanding citizens’ level of literacy and perceived agency with regards to algorithmic public services, 3) empowering citizens to critically engage with algorithmic public services, and 4) transforming design of public sector AI services to ensure civic participation.

Applicants must show a keen interest in this topic and bring a mix of technical and soft skills in at least one of these aspects: programming and rapid prototyping of web-based platforms, using NLP and textual data processing for analysing content and data visualization, and/or conducting interviews and qualitative research with potential participants as part of our team.

You would join the CRAI-CIS research group in the Computer Science department at Aalto University. The transdisciplinary group explores the impact of technology in critical societal contexts, working at the intersection of computational and social sciences engaging HCI and participatory design. More here: https://crai-cis.aalto.fi

46. Deep learning with differential equations

Supervisor: Dr. Markus Heinonen
Email: [email protected]
Number of open positions: 1

We are looking for a motivated student to push the boundaries of deep learning with differential equations. In conventional neural networks the inputs are transformed by a sequence of layers, while an alternative paradigm emerged recently interpreting deep learning as continuous signal of a differential equation system (ODE, SDE, PDE). We aim at developing new ways to perform machine learning by repurposing differential equations.

Possible topics include (i) accelerating neural network ODE optimisation with Lagrangians, (ii) studying the capacity of generative models (see eg. https://distill.pub/2020/growing-ca/), or (iii) applying ODE models for the task of classification. In these works we build on top of existing, international research.

Background with machine learning, statistics, math or physics learning will be useful. This work builds on our previous work, eg.

* https://arxiv.org/abs/2206.13397

* https://arxiv.org/abs/2106.10905

* https://yogeshverma1998.github.io/ModFlow/

47. Prediction of blood demand in Helsinki University Hospital

Supervisor: Dr. Markus Heinonen
Email: [email protected]
Number of open positions: 1

Helsinki University Hospital (HUS) uses over 50 000 bags of red blood cells each year. Variation in blood demand is the largest source of uncertainty in the Finnish blood supply chain. This HUS and Finnish Red Cross Blood Service joint project builds predictors of blood demand using HUS blood transfusion data. Machine learning with real-world data and real-world impact!

We are looking for candidates with good understanding of statistics, data science or machine learning. This work is jointly supervised by Aalto and FRCBS.