i) ELIAS - European Lighthouse of AI for Sustainability: Join the team in the Sustainable Materials Innovation Hub led by Mike Shaver and the Centre for AI Fundamentals to transform the interface between AI and sustainability. The role involves using sustainability inputs to direct outcomes. ii) UKRI Turing AI World-Leading Fellowship: The role involves developing new principles and methods for Advanced User Modelling, sequential decision making, and Automatic Experimental Design, with and without a Human-in-the-Loop. iii) UKRI AI hub in Generative Models: The role involves developing principles and tools for generative models technology. iv) PhD student positions in the UKRI AI CDT in Decision Making for Complex Systems: Projects include 'Human-in-the-loop generative models for experimental design' and 'Learning theory and methods for novel types of distributional shifts'.

Thinking about the next position for your research career? I am hiring postdocs in my machine learning research group both in Helsinki, Finland and Manchester, UK. We develop new machine learning methods and study machine learning principles. Keywords include: probabilistic modelling, Bayesian inference, simulation-based inference, multi-agent RL and collaborative AI, sequential decision making and experimental design, active learning, human-in-the-loop learning and user modelling, privacy-preserving learning, Bayesian deep learning, generative models. We also solve problems of other fields with the methods – and use those problems as test benches when developing the methods. We have excellent collaborators in drug design, synthetic biology and biodesign, personalized medicine, cognitive science and human-computer interaction.

Thinking about the next position for your research career? I am hiring postdocs in my machine learning research group both in Helsinki, Finland and Manchester, UK. We develop new machine learning methods and study machine learning principles. Keywords include: probabilistic modelling, Bayesian inference, simulation-based inference, multi-agent RL and collaborative AI, sequential decision making and experimental design, active learning, human-in-the-loop learning and user modelling, privacy-preserving learning, Bayesian deep learning, generative models. We also solve problems of other fields with the methods – and use those problems as test benches when developing the methods. We have excellent collaborators in drug design, synthetic biology and biodesign, personalized medicine, cognitive science and human-computer interaction.

Neurofeedback provides a feedback display which is linked with on-going brain activity and thus allows self-regulation of neural activity in specific brain regions associated with certain cognitive functions and is considered a promising tool for clinical interventions. Recent reviews of neurofeedback have stressed the importance of applying the “double-blind” experimental design where critically the patient is unaware of the neurofeedback treatment condition. An important question then becomes; is double-blind even possible? Or are subjects aware of the purposes of the neurofeedback experiment? – this question is related to the issue of how we assess awareness or the absence of awareness to certain information in human subjects. Fortunately, methods have been developed which employ neurofeedback implicitly, where the subject is claimed to have no awareness of experimental purposes when performing the neurofeedback. Implicit neurofeedback is intriguing and controversial because it runs counter to the first neurofeedback study, which showed a link between awareness of being in a certain brain state and control of the neurofeedback-derived brain activity. Claiming that humans are unaware of a specific type of mental content is a notoriously difficult endeavor. For instance, what was long held as wholly unconscious phenomena, such as dreams or subliminal perception, have been overturned by more sensitive measures which show that degrees of awareness can be detected. In this talk, I will discuss whether we will critically examine the claim that we can know for certain that a neurofeedback experiment was performed in an unconscious manner. I will present evidence that in certain neurofeedback experiments such as manipulations of attention, participants display residual degrees of awareness of experimental contingencies to alter their cognition.

Remarkable advances in experimental neuroscience now enable us to simultaneously observe the activity of many neurons, thereby providing an opportunity to understand how the moment by moment collective dynamics of the brain instantiates learning and cognition. However, efficiently extracting such a conceptual understanding from large, high dimensional neural datasets requires concomitant advances in theoretically driven experimental design, data analysis, and neural circuit modeling. We will discuss how the modern frameworks of high dimensional statistics and deep learning can aid us in this process. In particular we will discuss: (1) how unsupervised tensor component analysis and time warping can extract unbiased and interpretable descriptions of how rapid single trial circuit dynamics change slowly over many trials to mediate learning; (2) how to tradeoff very different experimental resources, like numbers of recorded neurons and trials to accurately discover the structure of collective dynamics and information in the brain, even without spike sorting; (3) deep learning models that accurately capture the retina’s response to natural scenes as well as its internal structure and function; (4) algorithmic approaches for simplifying deep network models of perception; (5) optimality approaches to explain cell-type diversity in the first steps of vision in the retina.

Consciousness is a phenomenon unlike any other studied in natural science. Yet when building theories and designing experiments, we often proceed as if this were not the case. In this talk, I present two recent investigations of mine which explore the implications of consciousness' unique epistemic context for scientific theory building and experimental design. The first investigation is concerned with falsifications of theories of consciousness and identifies a rather deep problem in the usual scheme of testing theories. The second is an axiomatization and subsequent formalization of some of consciousness' more problematic epistemic features that allows to precisely quantify where the usual scientific methodology ceases to be applicable. For both cases, I indicate ways to resolve the problem.

Remarkable advances in experimental neuroscience now enable us to simultaneously observe the activity of many neurons, thereby providing an opportunity to understand how the moment by moment collective dynamics of the brain instantiates learning and cognition.  However, efficiently extracting such a conceptual understanding from large, high dimensional neural datasets requires concomitant advances in theoretically driven experimental design, data analysis, and neural circuit modeling.  We will discuss how the modern frameworks of high dimensional statistics and deep learning can aid us in this process.  In particular we will discuss: how unsupervised tensor component analysis and time warping can extract unbiased and interpretable descriptions of how rapid single trial circuit dynamics change slowly over many trials to mediate learning; how to tradeoff very different experimental resources, like numbers of recorded neurons and trials to accurately discover the structure of collective dynamics and information in the brain, even without spike sorting; deep learning models that accurately capture the retina’s response to natural scenes as well as its internal structure and function; algorithmic approaches for simplifying deep network models of perception; optimality approaches to explain cell-type diversity in the first steps of vision in the retina.