Helsinki ICT network: Doctoral student positions in computer science–Open positions

Machine learning and AI are extensively used in the sciences. When modelling physical systems, the understandability and statistical robustness of the models is often more important than predictive accuracy. We are looking for talented postdoctoral researchers and doctoral students to study explainable and understandable AI and the uncertainty quantification of AI models. While the AI methods we develop are generic and not tied to any specific application domain, we work closely with scientists to build Virtual Laboratory for Molecular Level Atmospheric Transformations (https://wiki.helsinki.fi/display/VILMA). More information: https://bit.ly/edahelsinki2023jobs

Supervisor: Kai Puolamäki (Helsinki)

Keywords: explainable AI, uncertainty quantification

Research area: Algorithms and Machine Learning (AML)

GreenNLP addresses the problem of increasing energy consumption caused by modern solutions in natural language processing (NLP). Neural language models and machine translation require heavy computations to train and their size is constantly growing, which makes them expensive to deploy and run. In our project we will reduce the training costs and model sizes by clever optimizations of the underlying machine learning algorithms with techniques that make use of knowledge transfer and compression. Furthermore, we will focus on multilingual solutions that can serve many languages in a single model reducing the number of actively running systems. Finally, we will also openly document and freely distribute all our results to enable efficient reuse of ready-made components to further decrease the carbon footprint of modern language technology. The project will run in collaboration with the University of Turku and CSC IT center of science. The position is for 2 years with a possibility of an extension until the end of that project in December 2025.

This position is within the Doctoral Programme for Language Studies at the University of Helsinki. More information about the programme can be found https://www.helsinki.fi/en/admissions-and-education/apply-doctoral-programmes/doctoral-schools-and-doctoral-programmes/doctoral-school-humanities-and-social-sciences/doctoral-programme-language-studies 

Supervisor: Jörg Tiedemann (Helsinki) (in collaboration with TurkuNLP and CSC)

Keywords: Language technology, NLP, deep learning, large language models, machine translation, knowledge distillation

Research area: Algorithms and Machine Learning (AML)

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, we formalize and develop theory 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 modeler 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.  Similar workflow ideas can be used to support also analysis of networks of models, making it easier to illustrate similarities and important differences. By making applied scientific research and data analysis more reliable and reproducible, our understanding of the world and decision-making will be improved. Related paper

https://arxiv.org/abs/2011.01808 and video https://www.youtube.com/watch?v=ppKpwtGy8KQ

Supervisor: Aki Vehtari (Aalto)

Research area: Algorithms and Machine Learning (AML)

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 Doctoral and Postdoc 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 and publications in one or more of these areas is highly beneficial. Good interpersonal skills, collaborative research, conducting ethical research studies and participatory design with end users, and/or project coordination experience 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

Keywords: Human Computer Interaction (HCI), Natural Language Processing (NLP), conversational AI, speech/voice interaction, design research, human-AI interaction

Supervisor(s): Nitin Sawhney (nitin.sawhney@aalto.fi) and Tom Bäckström (tom.backstrom@aalto.fi)

Research area: algorithms and machine learning

Researchers interested in postdoctoral/doctoral positions are invited to apply for this project. Here, we shall study a new approach to synthesis of efficient communication schemes – including learning of novel concepts – in cooperative multi-agent systems, trained via reinforcement learning. We combine symbolic methods with machine learning, in what is referred to as a neuro-symbolic system, where a neural network learns to produce programs in a symbolic language to solve a task at hand.  Agents will not be restricted to use only primitives provided as input, but interactions will be interleaved with (symbolic) steps to extend and revise their current language with novel higher-level concepts, allowing for generalisation and more informative messages. We believe this combination of neural and symbolic methods will be an important next step in the development of AI beyond today’s capabilities, including bridging the gap between black-box neural models and classical symbolic methods. See [1] for a gentle introduction to the area.

The postdoc funding is assured for two years, while the PhD funding is for four years. Research for this project will be conducted at Aalto University, Finland in a world-class group known for its contributions to representation learning, generative models, and multiagent systems [2-8]  

The selected candidate will work closely with the group of Prof. Moa Johansson (Chalmers University, Sweden).  There might also be an opportunity to work with our collaborators at MIT.  

Contact: Vikas Garg (vgarg@csail.mit.edu  or  vikas.garg@aalto.fi)  

Supervisor Name: Vikas Garg

Research Area: Algorithms and Machine Learning

Machine learning (ML) is the foundation of artificial intelligence in today’s applications. The scope of ML is wide, including (but not limited to) speech synthesis and recognition, machine translation, and computer vision. This project involves designing and (or) applying ML techniques to the scenarios represented by wireless and mobile systems.

The related research encompasses: design of efficient deep neural network architectures for embedded devices; protocols and algorithms for distributed inference; lifelong learning for optimization of wireless systems; applications of ML to virtual and augmented reality. The project requires strong analytical skills, proficiency in using ML frameworks such as Tensorflow, and keen interest in applications of ML.

Supervisor: Mario Di Francesco

Keywords: machine learning, wireless communication, mobile computing, embedded systems, augmented reality, optimization, lifelong learning

Research Area: Algorithms and Machine Learning and Networks (AML), Networked Systems and Services (NNSS)

Cloud computing has introduced a new model for provisioning resources and services over the Internet on top of a virtualized, elastic infrastructure. Additional paradigms have also recently emerged: edge, fog, and serverless. Moreover, modern (namely, cloud-native) applications are generally composed of multiple microservices, realized as software containers and managed through an orchestrator such as Kubernetes. This project involves addressing different aspects of cloud-native systems, with focus on software systems and security. The related research encompasses: design and analysis of techniques for scalable and reliable computing; applying network economics to different computing paradigms and their applications; addressing open challenges in cloud network security; characterizing the performance and the security of applications based on microservices. The project requires strong analytical skills, proficiency in programming distributed systems, and solid knowledge on cloud-native software, in particular, Kubernetes.

Supervisor: Mario Di Francesco

Research Area: Networked Systems and Services (NNSS)

Keywords: cloud computing, edge computing, serverless, microservices, cloud security, network economics, distributed systems.

The project will study and develop new Transformer-based models for modeling of multispectral, hyperspectral and lidar based remote sensing data in varying resolutions. We will analyze the information content of modern and future hyperspectral earth observation and lidar data obtained over the boreal forest zone. The results will be used for studying the biodiversity of forests based on the distrubution of tree species and the spatial arrangements of individual trees. In addition, the models will be included in a digital twin of the Finnish forests.  The project aims at novel methodological and algorithmic improvements by using state-of-the-art Transformer models that have already been studied and developed in the research group.

Supervisor: Jorma Laaksonen, Senior University Lecturer Aalto University

Keywords: machine learning, deep learning, transformers, computer vision, data fusion, remote sensing, hyperspectral data

Research area: Algorithms and Machine Learning (AML)

Aalto University has launched a new collaborative initiative in transformative research to support cooperation between AI and Data Science researchers with researchers in other fields. The aim is to create new knowledge by developing AI methods to advance research in other domains.

We are especially seeking candidates interested in collaborations between Artificial Intelligence and/or Data Science with the following areas for Transformative AI collaborations:

1. AI-assisted modelling in economics: We are starting collaboration to address a few key questions in economics with new machine learning based approaches. Interesting topics include: 1. health economics: data-driven targeting of preventive treatments; 2. prior elicitation for economic models; 3. AI-assisted mechanism design and design of economic models. We welcome strong applicants with an interest in doing a PhD jointly supervised by a machine learning professor from the Finnish Centre for Artificial Intelligence FCAI and and an economics professor from Helsinki Graduate School of Economics GSE.
   Keywords: Machine learning, AI-assisted modelling, economics
   Supervisors: Otto Toivanen, Samuel Kaski
2. Automatic generation of production planning and scheduling optimization models from data: We aim at creating mathematical programming models by fitting a black box neural network model to the data collected from real or
   simulated processes and converting it into a “classical” (MILP, NLP) optimization model. The approach would have several benefits: 1) Semi-automation in model generation; 2) Automatic model calibration – represents reality, at least on average; 3) Ready-made general solvers can be used for solving
   Keywords: Explainable NNs, metamodeling, gray-box modeling
   Collaborating supervisor: Esko Niemi, Digital Production Group
3. AI and Digital Twins in Advanced Manufacturing: The physical manufacturing processes are supplemented by their Digital Twins. Digital Twin is a concept that includes all the analytical capability and the data generated by the physical process in real time.  It is like weather forecasting for optimal manufacturing processes.  For the most optimal process we want to add sensors and other data gathering concepts for optimization.  The most of manufacturing process optimization with Digital Twins is based on analytical understanding of the process.  However, the amount of data that will be available will probably be beyond the capability of analytical models and this calls for new AI-based models. More information about the group is available on the website.
   Collaborating supervisor: Jouni Partanen, Additive Manufacturing Technology and Applications Group
4. Deep learning and process-based model fusion for improved understanding of the earth system:  Earth system models are key to understanding the ongoing global change and potential mitigation strategies. Enhancing the capacity of the models underlying the predictions about the future conditions of our planet is crucial to mitigate and adapt to ongoing global environmental change. Towards this goal, deep learning methods have recently been applied also in the field of earth system modeling, contesting, or even surpassing the performance of traditional process-based models. These two methodologies approach the modeling problems from different perspectives; deep learning and other machine learning methods are data-driven, limited by data availability, whereas process-based models are mainly limited by model flexibility and our understanding of the underlying processes. This project focuses on understanding how to combine the strengths of these two types of models. Potential avenues to explore include, among others, 1) deep-learned model ensembles where deep learning informs an optimal model ensemble based on local conditions, 2) ensembles consisting of both data-driven and process-based models, 3) hybrid models integrating different aspects of DL and process-based models, or 4) identifying conditions where the strengths of each model type, data-driven or process-based, could be exploited. The desired outcome of this project is methodological advancements on how deep learning improves earth system modeling and practical applications leveraging the strengths of deep learning and process-based modeling in environmental and earth system sciences. More information about the research group is available on their website.
   Collaborating supervisors: Matti Kummu, Olli Varis, Water and Development Research Group
5. Multilevel optimization of autonomously interacting textiles: In the Multifunctional Materials Design group, we target a paradigm shift for smart soft materials into user-customizable textiles that autonomously interact with their environment and communicate through changes in color or shape. This will be done by coupling modular thermo- and photo-actuators to different textile constructions ranging from traditional fiber arts to industrial knitting and weaving techniques. Optimizing responses of different textile architectures composed out of actuating yarns opens an interesting question of how to predict the behavior of a dynamic textile combining multiple functionalities. We are especially interested in combining physical modelling of interacting/actuating networks to data science approaches in order to predict the dynamic behavior of textiles. We are a welcoming group of researchers that embraces transdisciplinary methods and diverse – often surprising contributions (Multifunctional Materials Design | Aalto University).
   Collaborating supervisor: Jaana Vapaavuori, Multifunctional Materials Design Group
6. Physically interpretable AI for power-conversion applications: AI can be used for condition monitoring of power converters and electrical machines that are used, for example, in renewable energy generation. For these kind of applications, physically interpretable AI methods are of interest. Furthermore, it would be important to be able to generalize data and models of one device design to other types of device designs.
   More information about the research group is available on their website
   Collaborating supervisor: Marko Hinkkanen, Electric Drives Group
   AI supervisor: Heikki Mannila
7. Biomaterials: The School of Chemical Engineering and School of Science host jointly the LIBER Centre of Excellence (https://www.aalto.fi/en/liber) with multiple research directions having potential for transformative research in collaboration with artificial intelligence. The focus is on processes leading to emergent materials structures, complex functions of materials, and dissipative nonequilibrium processes. Functions that are developed are for example self-organization, regeneration, adaptability, and self-repair. Other possible research directions include working with new imaging and data acquisition methods, and new computationally-aided design methods of new materials.
   Collaborating supervisor: Markus Linder, LIBER Centre of Excellence
8. Computational Design: Computational design applies AI methods to automate and assist professional designers in their tasks. Until recently, the field was dominated by optimization and model-based methods, but with deep nets has shifted into data-driven approaches. We believe that the next revolution will build on user models (models of designers), which are needed to make more relevant and timely suggestions for the human designer. We are looking for researchers interested in studying possibilities emerging in latest AI research, such as using LLMs to enable natural language interaction with AI assistants in design, and training multi-purpose transformers and diffusion models for design tasks. We are also interested in methods for inferring designers’ preferences interactively and using this knowledge to propose completions to design tasks.
   Supervisor: Antti Oulasvirta
9. Integrating Human Electrophysiological Models and Data using Bayesian methods: Bayesian workflow offers methods to constrain models with experimental data while retaining prior knowledge from e.g., neuroanatomical studies. In this project, we will use state-of-the-art Bayesian workflow tools (Kallioinen et al. (2022), Siivola et al. (2021), Gelman et al. (2020), Piironen et al. (2020)) to constrain BNMs with MEG data. Implementation: We will use (i) Bayesian Optimisation-based Likelihood-Free Inference (LFI) methods to
   estimate model parameters from MEG data, (ii) Projection Predictive Inference to inform decisions on parameters to include in the model, and (iii) Power-scaling approaches to diagnose sensitivity of the parameter estimates to model changes. 
   Supervisors: Aki Vehtari, Matias Palva
10. AI forecasting methods for renewable energy: Forecasting variable renewable energy generation and its impact on day ahead power prices. Especially prediction of wind power generation in a few coming days and modelling how the wind variation will impact power balance and thus the market prices.
    Supervisors: Simo Särkkä, Matti Lehtonen
11. Pattern discovery and structure discovery in sequential data: The project looks at sequential data (e.g., data from the electric grid), and combines structure discovery and search for rare phenomena.
    Supervisor: Heikki Mannila

If you are particularly interested in one or more of the Transformative AI projects listed above, please indicate which project(s) you would be interested working with in your cover letter.

Research Area: Algorithms and Machine Learning (AML)

Recent progress in machine learning for generative and predictive models of molecules brings us towards computational, automatised drug design. We develop statistical methods and models for molecular structures, energies and interactions with the help of deep learning. A number of open problems reside in developing neural network models with physics-based inductive biases, in generative models in 3D spaces, in modelling the property landscapes of molecules, and in generalizing outside the training distribution in molecular design.
We are looking for motivated candidates with background in computational sciences, machine learning, statistics. You will join the Probabilistic Machine Learning (PML) research group at Aalto, https://research.cs.aalto.fi//pml

Supervision: Samuel Kaski, Markus Heinonen

Keywords: probabilistic modelling, drug design, deep learning

Research area: Algorithms and Machine Learning (AML)

 I am looking for a new doctoral student in my team which develops probabilistic modelling and Bayesian inference methods. The team has several exciting new machine learning formulations we work on, and opportunities for applying the methods with top-notch collaborators. But the core is always development of new methods, and with this call I am looking for talented researchers with background in machine learning, stats or CS (or other directly relevant topics) who are keen on developing the new methods. In the cover letter, let me know what you are interested in – if we are already working on it, all the better, but I am willing to listen to new ideas too. Keywords include: probabilistic modelling, Bayesian inference, simulator-based / likelihood-free inference, multi-agent RL and collaborative AI, sequential decision making and experimental design, privacy-preserving inference, Bayesian deep learning. Part of the team is in Finland/EU and part in the UK where I run a Turing AI World-Leading Researcher project. Plenty of collaboration opportunities exist between those, in ELLIS, and overseas.

Supervision: Samuel Kaski (Aalto University)

Research area: Algorithms and Machine learning (AML)

We develop AI assistants which help people make better decisions, with ongoing applications in science and engineering. We use multi-agent formalisms to define the assistance problems these assistants solve, including the human being assisted, and employ models of human behavior to (pre-)train them in silico. We build models of human behavior using (multi-agent) reinforcement learning, based on theories of human behavior from cognitive science.
In this project you will develop novel multi-agent formalizations of assistance and create new models of human behavior for these formalizations. The focus will be on maximally autonomous assistants — assistants that automate a person’s task as much as possible, while using a minimal interaction to learn to solve said person’s task well.
We are looking for a postdoc and doctoral student with experience in probabilistic machine learning and reinforcement learning. No formal experience with cognitive science is required. Additional knowledge in any of the following will be helpful: game theory, multi-agent RL, Bayesian RL, computational rationality, and inverse reinforcement learning.
This project will involve collaborations with Prof. Ville Kyrki (robotics), Prof. Andrew Howes (cognitive science), TU Delft, MIT, and pharma and self-driving car companies.

Supervision: Samuel Kaski (Aalto University) , Andew Howes (University of Birmingham)

Keywords: user modeling, multi-agent RL, human-AI interaction, cooperative AI

Research area: Algorithms and Machine Learning (AML)

Current machine learning approaches have shown outstanding success in various tasks. However, they generally require an explicitly defined goal, for instance as a reward or objective function. Defining these goals in real-world applications is laborious and error-prone, often leading to misaligned and undesirable behavior. We develop assistants that infer people’s goals through interaction, and thus avoid the need for well-specified goals.
In this project, you will develop principles and methods for AI assistants to be maximally autonomous. The goal is to automate to the greatest extent possible, while using a minimal interaction with the person being assisted. Interaction is important to learn the person’s goal, but should be done sparingly. It should only happen when necessary, i.e. when it can help reduce uncertainty about the goal and improve the assistant’s long-term decision making.
We are looking for a postdoc and doctoral student with experience in probabilistic machine learning and reinforcement learning. No formal experience with cognitive science is required. Additional knowledge in any of the following will be helpful: game theory, multi-agent RL, Bayesian RL, computational rationality, and inverse reinforcement learning.
This project will involve collaborations with Prof. Ville Kyrki (robotics), Prof. Andrew Howes (cognitive science), TU Delft, MIT, and pharma and self-driving car companies.

Supervision: Samuel Kaski, Ville Kyrki

Keywords: user modeling, multi-agent RL, human-AI interaction, cooperative AI

Research area: Algorithms and Machine Learning (AML)

Differential privacy allows developing machine learning algorithms with strong privacy guarantees. Recent work shows it is possible to combine strong privacy and high accuracy by pre-training models on public data and only fine-tuning the model with the sensitive data. However, high accuracy still requires care for example in hyperparameter tuning. The aim of this project is to develop methods that make it easier to train high accuracy private models. The project will benefit from a very large grant of compute time on LUMI, 3rd fastest supercomputer in the world. The project requires a background in deep learning.

Supervision: Antti Honkela, Samuel Kaski

Keywords: Deep learning, image classification, hyperparameter learning, differential privacy

Research area: algorithms and machine learning (AML)

Speech synthesis has advanced dramatically with the emergence of generative deep learning methods, such as WaveNets, GANs, Diffusion Models, and Transformer-based acoustic modeling. These methods can achieve human-level naturalness, but they currently lack intuitive means of controlling and interacting with the synthesizer. A major contributor to this issue is the present text-to-speech synthesis paradigm: speech simply contains information that cannot be uniquely inferred from text.

Explicit features based on signal processing and linguistics are currently under-utilized in state-of-the-art systems, but these kinds of features have high potential for interactive guided synthesis in a deep learning synthesis system. For example, a user could explicitly adjust the synthesizer’s accent placement, pitch contour, and formant frequencies.

Furthermore, speech has a large amount of variability that is impractical to annotate, difficult to describe in terms of signal features, and hard to infer from text. This kind of variation is exemplified in expressive speech, varying vocal effort levels, and speaker identity in multi-speaker systems. Handling such latent variability calls for the use of self-supervised methods for learning control with disentangled representations and zero-shot adaptation.

This project aims to develop an interactive and intuitive neural speech synthesizer that encompasses both explicit control using signal processing and latent control using learning-based methods. Easy-to-use control beyond text will enable users, such as voice artists, creators, and linguists to interactively modify and adjust their voice synthesizer output.

The ideal candidate for this doctoral research project has a strong background in machine learning and a working knowledge of deep learning Python libraries (such as PyTorch). Further background in speech and/or audio signal processing, C++ DSP programming, and a general interest in sound synthesis is beneficial.

Supervisor: Lauri Juvela Assistant Professor in Machine Learning for Speech and Language Technology, Aalto University, ELEC/DICE

Accumulation of massive amounts of 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. Successful applicants are expected to have outstanding skills in machine learning, statistics, applied mathematics, or a related field. The goal of the position is to develop novel methods for challenging biomedical applications. Examples of our recent research can be found in https://users.ics.aalto.fi/~pemartti/.

Supervisor: Prof. Pekka Marttinen (pekka.marttinen@aalto.fi)

We develop amortized experimental design and inference techniques that take into account the down-the-line decision making task. For example, this may include delayed-reward decision making where data has to be measured, at a cost, before making the decision. This problem occurs in the design-build-test-learn cycles which are ubiquitous in engineering system design, and experimental design in sciences and medicine. The solutions need Bayesian experimental design techniques able to work well with simulators, measurement data and humans in the loop, who are both information sources and the final decision makers. For online and real-time tasks, algorithmic recommendations need to come near-instantly, thus requiring amortization of both experimental design and of the decision-making suggestions. The assistive methods need to account for uncertainty in the inference process and possibly in the utility function itself.

We are looking for a machine learning researcher with familiarity with probabilistic modelling, amortized inference via deep learning techniques, and/or Bayesian experimental design, interested in developing the new methods, with options on applying the techniques to improve modelling in the FCAI’s Virtual Laboratories.

Keywords: Sequential design of experiments, Bayesian experimental design, active learning, amortized inference

Supervision: Luigi Acerbi (University of Helsinki), Samuel Kaski (Aalto University)

Research Area: Algorithms and Machine Learning (AML)

Recent advances in machine learning have shown how powerful emulators and surrogate models can be trained to drastically reduce the costs of simulation, optimization and Bayesian inference, with many trailblazing applications in the sciences. In this project, the candidate will join an active area of research within several FCAI groups to develop new methods for simulation, optimization and inference that combine state-of-the-art deep learning and surrogate-based kernel approaches – including for example deep sets and transformers; normalizing flows; Gaussian and neural processes – with the goal of achieving maximal sample-efficiency (in terms of number of required model evaluations or simulations) and wall-clock speed at runtime (via amortization). The candidate will apply these methods to challenging problems involving statistical and simulator-based models that push the current state-of-the-art, be it for number of parameters (high-dimensional amortized inference), number of available model evaluations (extreme sample-efficiency) or amount of data. The ideal candidate has expertise in both deep learning and probabilistic methods (e.g., Gaussian processes, Bayesian optimization, normalizing flows).

References:

• Acerbi (2018); NeurIPS: https://arxiv.org/abs/1810.05558
• Acerbi (2020); NeurIPS: https://arxiv.org/abs/2006.08655
• Järvenpää & Corander (2021); arXiv: https://arxiv.org/abs/2104.03942
• Cranmer et al. (2020); PNAS: https://doi.org/10.1073/pnas.1912789117

Keywords: Emulators, amortized inference, Bayesian optimization, normalizing flows, simulator-based inference

Supervision: Profs. Luigi Acerbi (University of Helsinki), Jukka Corander (University of Helsinki)

Research Area: Algorithms and Machine Learning (AML)

The goal of the project is to develop novel deep learning algorithms to answer open questions in nanoscale physics such as predicting molecular structure in liquids or developing accurate, but very fast models of water. In particular we want to take advantage of the capabilities of Graph Neural Networks to provide robust and highly efficient simulators for molecular systems. These models will be coupled to state-of-the-art experimental characterisation, ultimately including a dynamic interaction where simulations are actively used to focus on information rich regions during experiments. We are looking for applicants with a strong background in deep learning and/or physical simulations.

Keywords: Deep learning, graph neural networks, material science, physics simulations

Supervision: Profs. Adam Foster (Aalto University), Alexander Ilin (Aalto University)

Research Area: Algorithms and Machine Learning (AML)

Machine learning (ML) is now used widely in sciences and engineering, in prediction, emulation, and optimization tasks. Contrary to what we would like to think, it does not work well in practice. Why?

Because the conditions during deployment may radically differ from training conditions. This has been conceptualized as distribution shift or sim-to-real gap, and a particularly interesting challenge which we will be tackling is changes due to unobserved confounders. Solving this challenge is imperative for widespread deployment of ML. In this project, we will consider deployment as a fundamental machine learning challenge, developing new principles and methods for tackling this problem which can be argued to be the main show-stopper in making machine learning seriously useful in solving the real problems we are facing, in sciences, companies and society. We have exciting test cases in FCAI’s highlight problems, in drug design, materials science, and health applications. We are looking for candidates with strong background in probabilistic machine learning.

Keywords: ML deployment, distribution shift, out-of-distribution, unobserved confounders

Supervision: Profs. Samuel Kaski (Aalto University), Vikas Garg (Aalto University)

Research Area: Algorithms and Machine Learning (AML)

Probabilistic models are essential for capturing the stochastic properties in speech and audio signals. Speech synthesis in particular has recently emerged as a proving ground application for deep generative models. Autoregressive (AR) density models, such as WaveNet, achieve high quality, can be trained effectively using maximum likelihood, and have low algorithmic latency suitable for real-time applications. However, available implementations of AR inference using GPUs and popular Python libraries are slow, which has shifted research to feedforward models for parallel inference. These models can similarly achieve high quality, whether trained as Generative Adversarial Networks (GANs), Diffusion Models or Energy Based Models (with contrastive estimation). However, these training recipes often involve a complex mixture of training objectives, while feedforward models can further be relatively inefficient and introduce algorithmic latency. Meanwhile, Digital Signal Processing (DSP) methods for speech and audio contain valuable knowledge both in perceptually relevant metrics for similarity, and in design and implementation of models with feedback. First, we propose to leverage this knowledge to develop efficient DSP-based recurrent building blocks (with forward and backward stability guarantees), and integrate them into a deep learning system. Second, we aim to formulate a unified framework for evaluating the various probabilistic models for speech using Score Matching Networks, experiment on what is really needed to make a model work, and extend the current approaches to use perceptually relevant score matching functions. We are looking for applicants with a strong background in deep learning and probabilistic modeling.

Keywords: Deep generative models, digital signal processing, probabilistic modeling, speech

Supervision: Profs. Lauri Juvela (Aalto University), Alexander Ilin (Aalto)

Research Area: Algorithms and Machine Learning (AML)

Both MCMC and distributional approximation algorithms (variational and Laplace approximations) often struggle to handle complex posteriors, but we lack good tools for understanding how and why. We study diagnostics for identifying the specific nature of the computational difficulty, seeking to identify e.g. whether the difficulty is caused by narrow funnels or strong curvature. We also develop improved inference algorithms that account for these challenges, e.g. via automated and semi-automated transformations for making the posterior easier or by better accounting for the underlying geometry. We are looking for applicants with a strong inference background and interest in working on improving inference for the hardest problems.

Keywords: MCMC, variational approximation, differential geometry, inference diagnostics, Bayesian workflow

Supervision: Profs. Aki Vehtari (Aalto University) and Arto Klami (University of Helsinki); primary/secondary depending on the candidate’s interests

Research Area: Algorithms and Machine Learning (AML)

FCAI is actively developing methods and software for virtual laboratories to enable AI assistance in the research process. We are looking for a candidate to research explainable AI and uncertainty quantification. Efficient human-AI collaboration requires methods that are either inherently capable of providing explanations for the decisions or can explain the decisions of other AI models. For instance, the user needs to know why AI recommends a particular experiment or predicts a specific outcome. They should always be aware of the reliability of the AI models. You will conduct the project with a team of AI researchers with access to researchers specialised in various application areas. The applicant should be interested in incorporating the techniques as part of virtual laboratory software developed at FCAI for broad applicability.

Keywords: Virtual laboratory, explainable AI, uncertainty quantification, human-AI collaboration

Supervision: Profs. Arto Klami (University of Helsinki), Kai Puolamäki (University of Helsinki)

Research Area: Algorithms and Machine Learning (AML)

Foundation models such as GPT-3 and CLIP are revolutionizing how AI is developed and applied, by providing reusable and general-purpose building blocks with unprecedented capabilities. Instead of training large-scale models from scratch for thousands or millions of GPU hours, one can solve novel tasks by combining pretrained foundation models in novel ways, or finetuning them for downstream tasks. The release of OpenAI’s CLIP, for instance, soon led to the emergence of CLIP-guided image generation models such as Disco Diffusion, Stable Diffusion, Midjourney, and DALL-E 2, as well as smaller-scale experiments such as ClipDraw and StyleClipDraw. CLIP also increasingly empowers various semantic search solutions across multiple industries.

The objective of this research topic is to study, develop, and test/validate foundation models for interactive computing, where they have received relatively less attention so far. The specific research foci depend on the hired researchers’ interests, but might include:

• Large language models (LLMs) as textual human simulacra, e.g., in generating synthetic user research data, role-playing research participants in rapid research exploration and piloting.
• Foundation models for simulated human movement control, such as AI “user simulators” that can be used to test interactive systems, or virtual AI motion capture actors that can follow choreographer instructions to generate animations or suggest ways to solve movement problems.
• Multimodal extensions of the above, e.g., video-language reward models that determine how well an AI agent solves a task defined using natural language, based on visual observations, or user motivation and emotion models that can provide synthetic “think aloud” narrative based on user simulation data.
• Generative models for designing interactive systems, e.g., generating visual designs, personas, user stories, or wireframes.

Aalto FCAI teams have already made progress in many of the directions above (e.g., https://github.com/aikkala/user-in-the-box, https://dl.acm.org/doi/abs/10.1145/3490100.3516464, https://dl.acm.org/doi/abs/10.1145/2858036.2858233, https://github.com/NVlabs/stylegan3), providing an excellent foundation for future research.

Keywords: Large Language Models, Foundation Models, User simulation, Generative models

Supervision: Profs. Perttu Hämäläinen (primary, Aalto University); Robin Welsch (Aalto University), Christian Guckelsberger (Aalto University), Jaakko Lehtinen (Aalto University, NVIDIA)

Research area: Algorithms and Machine Learning

Incorporating accurate causal knowledge about an environment in terms of objects and their relationships into the world model of a reinforcement learning agent can yield significant improvements in reducing the amount of exploration required to solve new tasks and to generalize to new environments. Recently, causal representation learning has been proposed as a way of extracting and representing such causal knowledge from previous experience in the latent space. However, in practice data are extremely correlated and it is not straightforward to come up with ways to even disentangle relevant objects, not to mention the causal relationships between them. On the other hand, large language models (LLMs) encompass a lot of information about objects and causal relationships, for example: “User: How can I turn the lights on? GPT-3: Depending on the type of light fixture in the room, you can usually turn the lights on by flipping a switch near the entrance to the room.” Leveraging this type of information when learning world models for reinforcement learning (RL) and causal inference is still mostly underexploited, although the combination of LLMs with RL is a rapidly surfacing new topic in the machine learning community, see https://larel-workshop.github.io/. We are looking for a postdoc who would leverage and expand our previous work on RL agents with language understanding abilities, to address this highly topical research theme. In particular, we focus on improving the s-o-t-a on language guided RL benchmarks by empowering the agent with deep latent variable based probabilistic world models, which accurately encompass uncertainty that serves as the foundation for reliable planning.

Keywords: Causality, language modeling, reinforcement learning, world models

Supervision: Pekka Marttinen (Aalto); Alexander Ilin (Aalto)

Research area: Algorithms and Machine Learning

In many complex sequential decision making tasks, online planning is crucial for high-performance. Monte Carlo Tree Search (MCTS) is an efficient online planning tool which employs a principled mechanism for trading off between exploration and exploitation. Following the success of MCTS in discrete control problems (such as the games of Go, Chess, and Shogi), various MCTS extensions have been proposed to continuous domains. However, the inherent high branching factor and the resulting explosion of the search tree size is limiting existing methods. In this project, we investigate novel extensions of MCTS for continuous domains based on search graphs. Our approach is built on the idea that sharing the same action policy between several states can yield efficient planning and thus high performance. This results in a limited number of stochastic action bandit nodes to produce a layered graph instead of an MCTS search tree allowing for long term planning. The designed algorithms can be used for robotic manipulation and navigation, for example, with a Boston Dynamics Spot robot. We are looking for applicants with a strong background in reinforcement learning (especially model-based) and tree search algorithms.

Supervision: Profs. Joni Pajarinen (Aalto University), Alexander Ilin (Aalto University)

Keywords: Continuous control, graph search, Monte Carlo tree search, online planning, reinforcement learning

Research area: Algorithms and Machine Learning

Machine learning is increasingly being used as a key element in research, for instance to efficiently approximate computationally costly simulations, automate design of experiments, and for integrated analysis of experimental results and multi-fidelity simulations. Much of the practical work is done in the context of specific applications in science, but our interest lies in the more general question of how ML could be used as part of the research process, essentially to improve the results and the scientific process itself. We seek solutions that work across multiple disciplines and applications. We are looking for candidates interested in aspects such as (a) how to best incorporate domain knowledge into probabilistic ML models, (b) how to integrate ML models as a part of the research process that involves e.g. also empirical experimentation, and (c) how to assist the research process itself using AI solutions. The work relates closely to our Virtual Laboratories initiative: realizing that most fields now use computational tools, essentially doing experiments first virtually with simulations, we can have scale advantages with AI methods that are cross-usable across the fields. The virtual laboratories give opportunities to demonstrate and validate the research contributions in several different natural science applications. An ideal candidate has expertise in both ML and some science domain, but candidates with strong background in either one are also considered.

Keywords: Natural science, virtual laboratory, AI-assisted research

Supervision: Profs. Samuel Kaski (Aalto University), Arto Klami (University of Helsinki), other professors (e.g. Patrick Rinke, Aalto University) depending on candidate’s interests/qualifications

Research Area: Algorithms and Machine Learning

We develop the new ML principles and methods needed by AI assistants to help people make better decisions, with ongoing applications in science and engineering. We use multi-agent formalisms to define the assistance problems these assistants solve, including the human agent being assisted, and develop new multi-agent RL solutions for the problem. We are particularly interested in (1) how to build and (pre-)train models of human behaviour based on cognitive science, and (2) how to solve new ad-hoc teamwork problems with multi-agent RL.

We are looking for new members in our team, with experience in probabilistic machine learning and multi-agent reinforcement learning. No formal experience with cognitive science is required. Additional knowledge in any of the following will be helpful but not necessary – we have a great team to work with: game theory, Bayesian RL, computational rationality, and inverse reinforcement learning.

Recent publications by the team:

1. https://arxiv.org/abs/2211.16277 (best paper award; HiLL@Neurips-22)
2. https://arxiv.org/abs/2202.07364 (AAAI-23)
3. https://arxiv.org/abs/2204.01160 (AAMAS-22)

Keywords: User modeling, multi-agent RL, human-AI interaction, cooperative AI

Supervision: Samuel Kaski (Aalto University) and other professors

Research Area: Algorithms and Machine Learning

Many applications of machine learning require training on distributed data while keeping the data private. Private federated learning enables this, but its communication requirements can be impractical. The aim of this project is to develop new approaches and methods for private learning on distributed data. A strong background in differential privacy and/or federated learning is an asset for this project.

Keywords: Differential privacy, federated learning, deep learning

Supervision: Profs. Antti Honkela (University of Helsinki), Samuel Kaski (Aalto University)

Research Area: Algorithms and Machine Learning

Sustainability is important in the context of AI and in particular there exist AI/ML/DS methods to improve sustainability of a given system using say AI methods. However, the computational methods used in AI are not always sustainable as they might require a lot of data or a lot of computational power produce the results. The aim of this research project is to take a look at sustainable AI methods and in particular sustainable computational methods whose energy fingerprint is minimal. One possible approach that has recently been studied is the clever use of parallel and distributed algorithms to decrease the amount of energy per flop. We are looking for applicants with interests in energy efficient or otherwise sustainable computational methods in general.

Keywords: Sustainability in AI, energy-efficient machine learning, data-light computing, parallel computing

Supervision: Profs. Simo Särkkä (Aalto University), Laura Ruotsalainen (University of Helsinki)

Research Area: Algorithms and Machine Learning

Bayesian models rely on prior distributions that encode knowledge about the problem, but specifying good priors is often difficult in practice. We are working on multiple fronts on making it easier, with contributions to e.g. prior elicitation, prior diagnostics, prior checking, and specification of priors in predictive spaces. We welcome applicants looking to work on any of these aspects and contribute to both theoretical development and practical software for aiding the prior specification process.

Keywords: Prior elicitation, Bayesian workflow, priors on predictive space, default priors

Supervision: Profs. Aki Vehtari (Aalto University) and Arto Klami (University of Helsinki); primary/secondary depending on the candidate’s interests

Research Area: Algorithms and Machine Learning

Applications need to be submitted through the online electronic application system of Aalto University. Applications sent through any other means will not be processed. You can find the application form through the website

All materials should be submitted in English in a PDF format. Note: files should be 5MB max. You can upload multiple files to the eRecruitment system, each 5MB.

1. Letter of motivation (two pages maximum)Please describe your background and future plans, and in particular the reasons for selecting the project(s) (you can get more information on the projects and supervisors through their web pages). Try to make your motivation letter as convincing as possible, so that the potential supervisors get interested. You do not have to write several motivation letters in case you apply for multiple projects, but if you prefer you can attach separate letters for individual projects.
2. A curriculum vitae and list of publications with complete study and employment history (please see an example CV at Europass pages)
3. A study transcript provided by the applicant’s university that lists studies completed and grades achieved.
4. A copy of the M.Sc. degree certificate. In the Finnish university system, a person must have a Master’s degree in order to enroll for doctoral studies. If the degree is still pending, then a plan for its completion must be provided. (The letter describing the completion plan can be free-format)
5. Contact details of possible referees from 2 senior academic people. We will contact your referees, if recommendation letters are required.

The HICT calls are targeted to prospective new doctoral students who are willing to start their doctoral studies in Aalto University or University of Helsinki under one of the HICT supervisors. Current funded doctoral students in Aalto University or University of Helsinki cannot participate in this call. While all applicants who have submitted an application by the deadline will be appropriately considered, Aalto University and the University of Helsinki reserve the right to consider also other candidates for the announced positions.

If you become selected as a new doctoral student, you need to apply for a study right for doctoral studies either at Aalto University or the University of Helsinki. Your new supervisor will assist you with the application. HICT itself does not award doctoral study rights or doctoral degrees. All doctoral students within the HICT network are doctoral students of either Aalto University or the University of Helsinki.

In the Finnish university system, a person must have a Master’s degree in order to enroll for doctoral studies. In case you wish to pursue graduate studies with a B.Sc. background, please apply first to one of the participating units’ Master’s programmes (Aalto University School of Science (SCI) or School of Electrical Engineering (ELEC), and University of Helsinki). A number of these programmes provide special “doctoral tracks” with some financial support and study plans oriented towards continuing to doctoral education after the M.Sc. degree.

A successful applicant must have an excellent command of Finnish, Swedish, or English. The universities participating in HICT have strict language skills requirements for doctoral students (Aalto University, University of Helsinki). International applicants applying for doctoral studies must demonstrate their proficiency in English. For example,an English language proficiency certificate (TOEFL, IELTS, CAE/CPE) is required later in case you will proceed to the recruitment process and apply for a doctoral study right.

Only the following applicant groups can be exempted from the language test requirement: applicants who have completed a higher education degree:

1. taught in Finnish, Swedish or English in a higher education institution in Finland.
2. in an English-medium programme at a higher education institution in an EU/EEA country, provided that all parts of the degree were completed in English.
3. an English-medium higher education degree requiring a physical on-site presence at a higher education institution in the United States, Canada, Great Britain, Ireland, Australia or New Zealand.

More information on minimum language requirements and language test scores can be found at the programs doctoral programme information pages (Aalto Doctoral Programme in Science (SCI), Aalto Doctoral Programme in Electrical Engineering (ELEC), University of Helsinki) (see “Language requirements” and “demonstrating proficiency in English”). Please be prepared to check the eligibility requirements for doctoral studies and present additional documents in case you will proceed to the recruitment and apply for doctoral study right in Aalto University or University of Helsinki.

Please find below some links with more detailed information about the eligibility requirements and doctoral studies in general at Aalto University and University of Helsinki:

Living and studying in Finland

For the Aalto Doctoral Programme in Science (SCI)

For the Aalto Doctoral Programme in Electrical Engineering (ELEC)

For the University of Helsinki

The exact starting date can be negotiated between the student and the supervisor. The student must have completed his/her M.Sc. degree by the time of starting doctoral studies.

New students also need to go through the standard doctoral student enrolment process of the hosting university/school before the start of the funding period. The supervisors will help in this process, once the best candidates have been identified and linked to a supervisor.

The maximum length of the funding period is four years.

The exact amount of monthly salary depends on the stage of the doctoral studies and varies between 2,000 and 3,000 euros/month. The level of the salary is sufficient for a funded student to focus on his or her doctoral studies full-time, without the need to resort to other sources of income.

Funded doctoral students are typically hired as full-time employees for the duration of their doctoral studies. The contract includes the normal occupational health benefits of the employing university. Finland has a comprehensive social security system.

The annual total workload of research and teaching staff at Finnish universities is 1624 hours per year. In addition to doctoral studies, persons hired are expected to participate in the supervision of students and teaching, following the standard practices of the recruiting unit.

Finland is among the best countries in the world with respect to education and many quality of life indicators. Both HICT partner universities (Aalto University and the University of Helsinki) have a wide range of courses offered in English. Boasting the largest technology hub of the Nordic countries, Finland is a world leader in information technology, business, design and many other academically centred fields.

Read more about living and studying in Finland.

You do not need to learn Finnish. The working environment of doctoral students is highly international, and the working language is English. You can normally also cope in English outside work as most Finns have a very good command of English.