Case Studies
The research in this project is conducted within five case studies
[a-e]. Their topical relationship is illustrated in the matrix below:
[a] Knowing cities better through a New Urban Science
Vladimir Cvetkovic, Angela Fontan, Cecilia Katzeff, Marco Molinari, Jeremy
Pitt, and Stacy Vallis
The goal of this case study is to examine how digital tools are being developed and applied at various scales to produce new insights about urban dynamics.
[b] The smart sustainable city as social networks
Amr Alanwar, Vladimir Cvetkovic, Angela Fontan, Karl H. Johansson, Jeremy
Pitt, Mikael Skoglund, Elis Stefansson, and Yuhao Yi
The goal of this case study is to examine cities as social networks, how human-social choices define these networks and how digital tools applied at various levels shape urban interactions and change.
[c] Finding correlations and causality in Live-In Lab Testbed data
Mahsa Farjadnia, Karl H. Johansson, Hedvig Kjellström, Marco Molinari,
and Ruibo Tu
The goal of this case study is to apply to a real building scenario automated approaches to isolate correlations among large set of sensors and showcase the possibility to find causality.
[d] Cognitive models of human decision making and behavior in smart cities
Athanasios Charisoudis, Pawel Herman, Hedvig
Kjellström, and Ricky Molén
The aim of this case study is to gain better understanding how decision making and its behavioural consequences at a population level can be modelled as an emergent effect of different interactions between individuals with their own goals, motivations and experiences.
[e] Making sense of sensory input to brain's neural networks using information theoretic instruments
Pawel Herman, Arvind Kumar, Movitz Lenninger, and Mikael Skoglund
The aim of this case study is to gain novel insights into computational principles of representing and processing neural information in the brain. Despite a relatively large body of experimental data, there is no well-established theory how populations of neural cells in the brain reliably encode information about complex environmental stimuli and how neural network circuits are optimized to make this information behaviourally relevant.
|
|
Publications
Publications [1+] from this project are listed in cronological order, and their subjects are illustrated in the matrix below:
[8] A. Karvonen,
V. Cvetkovic, P. Herman, K. H. Johansson, H. Kjellström,
M. Molinari, and M. Skoglund. The 'New
Urban Science': Towards the Interdisciplinary and
Transdisciplinary Pursuit of Sustainable Transformations, under
review, 2021.
|
|
|
Digitalisation is an increasingly important driver of urban development through the deployment of a wide range of networked technologies. The so-called 'New Urban Science' provides new ways of knowing and managing cities more effectively. These practices tend to emphasise urban data analytics and modelling but there are multiple opportunities to broaden and deepen the New Urban Science through collaborations between the natural and social sciences as well as with public authorities, private companies, and civil society. In this article, we summarise the history and critiques of urban science and then call for a New Urban Science that embraces interdisciplinary and transdisciplinary approaches to scientific knowledge development and application. We argue that such an expanded version of the New Urban Science can be used to develop urban transformative capacity and achieve environmentally friendly, economically prosperous, and socially robust cities of the 21st century.
|
|
[7] M. Lenninger, M. Skoglund, P. Herman and A. Kumar. Bandwidth expansion in the brain: Optimal encoding manifolds for population coding. In Cosyne,
2021.
|
|
|
Stimuli in the brain are represented in the population activity of neurons, where individual neurons are tuned to respond to a small set of stimulus values. At the population level, tuning of neurons implies that every stim-ulus is mapped onto a D-dimensional encoding surface (D = dimensionality of the stimulus) embedded in an N-dimensional space (N= # of active neurons). Mathematically, the representation of stimuli in the neural population activity is identical to the framework of bandwidth expansion in communication theory. Here, we exploit the bandwidth expansion framework to address the question: given noise and correlations, what is the optimal shape of the encoding surface and tuning curves? The notion of encoding surface led us to distinguish between two types of neural noise: channel noise (noise in spiking mechanism, background input) and observation uncertainty (noise in the stimulus-related input). Both noise sources can result in local (weak distortion) or global estimation errors (threshold distortion). We show that, in the case of channel noise, minimizing weak or threshold distortion leads to contradictory surface shapes. The optimal shape of the encoding surface is a trade-off between locally curved and globally flat. On the other hand, to minimize the effects of observation uncertainty, the optimal encoding surface should be flat. Thus, we propose that the optimal shapes of tuning curves depend on the relative strength of channel noise and observation uncertainty. We show that sparse coding(small tuning width) is suboptimal when observation uncertainty is large. These considerations also suggest that multipeak tuning curves (grid cells) are more susceptible to threshold distortion than single peaked ones. Finally, taking the example of typical tuning curves seen in the early visual system, we show that threshold distortion becomes an acute problem when neurons have multi-modal tuning curves.
|
|
[6] S. Molavipour, G. Bassi, and M. Skoglund. On neural estimators for conditional mutual information using nearest neighbors sampling. IEEE Transactions on Signal Processing 69:766-780, 2021.
|
|
|
The estimation of mutual information (MI) or conditional mutual information (CMI) from a set of samples is a long-standing problem. A recent line of work in this area has leveraged the approximation power of artificial neural networks and has shown improvements over conventional methods. One important challenge in this new approach is the need to obtain, given the original dataset, a different set where the samples are distributed according to a specific product density function. This is particularly challenging when estimating CMI. In this paper, we introduce a new technique, based on k-nearest neighbors, to perform the resampling and derive high-confidence concentration bounds for the sample average. Then the technique is employed to train a neural network classifier and CMI is estimated accordingly. We propose three estimators using this technique and prove their consistency, make a comparison between them and similar approaches in the literature, and experimentally show improvements in estimating the CMI in terms of accuracy and variance of the estimators.
|
|
[5]
M. Sorkhei, G. Eje Henter, and H. Kjellström. Full-Glow:
Fully conditional Glow for more realistic image generation. In
DAGM German Conference on Pattern Recognition, 2021.
|
|
|
Autonomous agents, such as driverless cars, require large
amounts of labeled visual data for their training. A viable
approach for acquiring such data is training a generative model
with collected real data, and then augmenting the collected real
dataset with synthetic images from the model,generated with
control of the scene layout and ground truthlabeling. In this
paper we propose Full-Glow, a fully conditional Glow-based
architecture for generating plausible and realistic images of
novel street scenes given a semantic segmentation map indicating
the scene layout. Benchmark comparisons show our model to
outperform recent works interms of the semantic segmentation
performance of a pre-trained PSPNet. This indicates that images
from our model are, to a higher degree than from other models,
similar to real images of the same kinds of scenes and objects,
making them suitable as training data for a visual semantic segmentation or object recognition system.
|
|
[4] D. Rolando and M. Molinari. Development of a comfort platform for user feedback: The experience of the KTH Live-In Lab. In International Conference on Applied Energy,
2020.
|
|
|
This paper presents the comfort platform created within a research project carried out at KTH Live-In Lab in Stockholm, Sweden. The KTH Live-In Lab is a platform of buildings to test and promote innovation into the built environment. The Live-In Lab includes several buildings with state-of-the-art and expandable sensor infrastructure.
The comfort platform has been created to manage user feedbacks in buildings. The comfort platform includes a user-friendly web application and a cost-efficient sensor device that allow to exchange feedbacks with the building users.
The comfort platform is proposed as a possible solution to bridge the gap between modern smart buildings and existing buildings with limited sensor capability.
This paper describes the comfort platform and the environment where it has been tested. The paper also summarizes the preliminary findings and the potential large-scale implementation.
|
|
[3] M. Molinari and D. Rolando. Digital twin of the Live-In Lab Testbed KTH: Development and calibration. In Buildsim Nordic,
2020.
|
|
|
In the last decade, the development of Information and Communication Technology (ICT) has enabled unprecedented possibilities to tackle worldwide ambitious sustainability targets. Demonstration facilities like the KTH Live-In Lab are fundamental for the adoption of ICT solutions for energy efficiency and sustainability in buildings. The Live-In Lab monitoring infrastructure enables the creation of a digital-twin, which facilitates a cost effective development, testing and implementation of advanced control and fault detection strategies.
The paper proposes a calibration methodology for the thermal model (energy and comfort) of the Live-In Lab, developed in IDA-ICE, to be deployed as a digital twin. The methodology first screens the parameters with most impact on energy use and then calibrates the model minimizing the error in both indoor comfort and energy use with a weighting parameter β. Calibration results are then validated against the measured data.
The results of this paper will be instrumental to the improvement of control systems and it will facilitate the study of behavioral aspects of the energy use.
|
|
[2] Y. Yi, L. Shan, P. E. Paré, and
K. H. Johansson. Edge deletion algorithms for minimizing
spread in SIR epidemic models. arXiv preprint arXiv:2011.11087, 2020.
|
|
|
How to effectively reduce the number of infections in epidemic models?
Control interaction between agents/devices/humans.
Applications to virus spread but also many other (social) phenomena.
|
|
[1] E. Stefansson, F. J. Jiang, E. Nekouei,
H. Nilsson, and K. H. Johansson. Modeling the decision-making in
human driver overtaking. In IFAC World Congress,
2020.
|
|
|
Analyze risk-agnostic and risk-aware decision models.
Judge whether an overtaking is desirable or not.
Numerical and experimental evaluations.
|
|