Skip to main content

Machine Learning day

Save to calendar

Jan 17

Date and time: 17 January 2022, 08:50 – 17:00 CET
Title: Machine Learning day
Where: online via Zoom

Zoom-link to MAIN STAGE:
Meeting ID: 625 3552 9876

Register for the event HERE.


The main objective behind this event is to map the rich landscape of machine learning research at KTH, Stockholm University and RISE, and inform each other about collaborative opportunities. For that purpose, we would like to invite Digital Futures faculty, their colleagues, industrial and societal partners, postdocs, and PhD students who want to share their machine learning research interests and network.

Both morning and afternoon poster sessions are divided into three parallel groups. Each group has its own Zoom room dedicated and is composed of 5-7 poster presentations, which are held one after another. The average time for each poster is around 7-8 mins and we expect a lot of informal discussions to take place. For each poster group, there is a moderator assigned to facilitate more focused discussions in dedicated break-out rooms (created on a spontaneous request). It is expected that posters will be presented either in the form of a few slides or a single pdf shared in the respective Zoom room with the accompanying live commentary by the authors in the background.


Zoom-link to MAIN STAGE:

08:50 – 09:00  Welcome and introduction

Aristides Gionis, Pawel Herman, Alexandre Proutiere and Karl H Johansson, Director Digital Futures

09:00 – 10:00  KEYNOTEProfessor Mikael Johansson, Decision and Control Systems, KTH

Addressing the algorithmic bottlenecks to large-scale learning: asynchrony, communication-efficiency and adaptivity

Abstract: Machine learning problems are rapidly expanding in size and scope. Increasingly often, we face problems where data, computations, and decisions need to be distributed on multiple nodes. These nodes may be individual cores in a CPU, different processing units in a multi-processor platform, or devices in a federated learning system. In these systems, training times are no longer limited by our access to raw processing power. Instead, they are bound by data transfer times and hampered by a lack of understanding of how to design and tune asynchronous optimization algorithms. In this talk, I will describe a few recent results that allow relieving such bottlenecks.

First, I will present novel convergence results for asynchronous iterations that appear in the analysis of parallel and distributed optimisation algorithms. The results are simple to apply and give explicit estimates for how the degree of asynchrony impacts the convergence rates of the iterates. Our results shorten, streamline and strengthen existing convergence proofs for several asynchronous optimization methods, and allow us to establish convergence guarantees for popular methods that were thus far lacking a complete theoretical understanding.

Next, I will describe our efforts to develop communication-efficient optimization algorithms. I will review techniques for gradient compression and quantify how the degree of compression affects the accuracy of the optimization process. I will then demonstrate how to compensate for the errors introduced by the gradient compression, leading to higher accuracy guarantees at the same low level of communications.

Finally, I will describe a framework for resource-adaptive algorithms that adjusts the amount of gradient compression online, maximizing the expected optimization progress per communicated bit. Experimental deployments demonstrate communication savings of several orders of magnitude compared to the state-of-the-art gradient compression strategies.

10:00 – 10:15  Coffee break

10:15 – 11:15  Short 5-minutes presentations

Time Presenter Presentation title
10:15-10:20 Tony Lindeberg Covariant and invariant deep networks
10:20-10:25 Hossein Azizpour Beyond Standard Deep Learning
10:25-10:30 Borja Rodriguez Galvez & Mikael Skoglund Information-theoretic bounds on the learning generalization error
10:30-10:35 Florian Pokorny Machine Learning & Robotic Manipulation at Scale
10:35-10:40 Guangyi Zhang & Aristides Gionis A family of decision-tree algorithms with complexity guarantees
10:40-10:45 Tobias Oechtering On privacy-preserving learning PATE
10:45-10:50 Martin Monperrus Machine Learning for Automated Program Repair
10:50-10:55 Raksha Ramakrishna & Gyorgy Dan Property inference attacks against neural network models
10:55-11:00 Kim Hammar & Rolf Stadler Intrusion Prevention through Optimal Stopping
11:00-11:05 Henrik Boström Reliable machine learning
11:05-11:10 David Broman Probabilistic Programming: Expressive Probabilistic Modeling with Efficient Bayesian Inference
11:10-11:15 Martina Scolamiero Topological data analysis


11:15 – 12:15  Poster session (moderators in bold)

Group 1:  Zoom  Meeting ID: 657 0579 6617

Group 1

Foundations, Security

Borja Rodriguez Galvez & Mikael Skoglund Information-theoretic bounds on the learning generalization error
Heng Fang & Hossein Azizpour Multi-temporal Consistency Regularization for Change Detection
Federico Baldassarre & Hossein Azizpour DeepFake Detection Explanations
Tobias Oechtering On privacy-preserving learning PATE
Mojtaba Eshghie, Cyrille Artho & Dilian Gurov Dynamic Vulnerability Detection on Smart Contracts Using Machine Learning

Group 2:  Zoom:  Meeting ID: 693 8968 9999

Group 2


Lissy Pellaco & Joakim Jalden Deep Unfolding for Wireless Communications
Xuechun Xu & Joakim Jalden Model Supported Deep Learning for Nanopore DNA Sequencing
Mojtaba Eshghie, Cyrille Artho & Dilian Gurov Dynamic Vulnerability Detection on Smart Contracts Using Machine Learning
Tiziana Fuoco, Kateryna Morozovska & Federica Bragone Physics Informed NNs for modelling temperature and loss distribution in power transformers
Christian Pek & Jana Tumova Data-driven safe set approximation

Group 3:  Zoom:  Meeting ID: 642 4424 7263

Group 3

Learning to Control

John S. Baras The One Learning Algorithm Hypothesis: Towards a Proof
John S. Baras Risk Sensitive Reinforcement Learning
Rijad Alisic & Karl H Johansson Learning Covert and Zero-Dynamics Data-Injection Attacks on Cyber-Physical Systems
Yu Xing & Karl H Johansson Learning Linear Systems with Multiplicative Noise from Multiple Trajectory Data
Yassir Jedra Learning and Controlling Dynamical Systems with Guarantees
Po-An Wang Frank Wolfe based algorithms in bandit problems

12:15 – 13:15  Lunch


Zoom-link to MAIN STAGE:

13:15 – 14:15  KEYNOTEDr Mounia Lalmas-Roelleke, Head of Tech Research @ Personalization at Spotify

Personalizing and diversifying the listening experience

Abstract: The aim of the Personalization mission at Spotify is to “connect listeners and creators in a unique and enriching way”. In this talk, Mounia will describe some of the (research) work to achieve this, from using machine learning to understanding listening diversity.

Bio: Mounia Lalmas-Roelleke is a Director of Research at Spotify, and the Head of Tech Research in Personalisation, where she leads an interdisciplinary team of research scientists, working on personalization and discovery. Mounia also holds an honorary professorship at University College London. In January 2022, she took an additional appointment as a Distinguished Research Fellow at the University of Amsterdam. Before that, she was a Director of Research at Yahoo, where she led a team of researchers working on advertising quality. She also worked with various teams at Yahoo on topics related to user engagement in the context of news, search, and user-generated content. Prior to this, she held a Microsoft Research/RAEng Research Chair at the School of Computing Science, University of Glasgow. Before that, she was Professor of Information Retrieval at the Department of Computer Science at Queen Mary, University of London.

14:15 – 15:30 Short 5-minutes presentations

Time Presenter Presentation title
14:15-14:20 Pawel Herman Brain-like perspective on AI and ML
14:20-14:25 Jörg Conradt Spiking Neuronal Networks for Real-Time Systems
14:25-14:30 Jan Kronkqvist Analyzing DNNs by mixed-integer optimization
14:30-14:35 Miguel Campos Pinto Coelho de Aguiar & Karl H Johansson Physics-informed learning for identification and state estimation in traffic applications
14:35-14:40 Carlo Fischione Machine Learning over Wireless Networks
14:40-14:45 Liane Colonna Introduction to visuAAL (Privacy-Aware and Acceptable Video-Based Technologies and Services for Active and Assisted Living)
14:45-14:50 Filip Cornell & Sarunas Girdzijauskas Representing graphs using Random Indexing
14:50-14:55 Karl Meinke, Arvind Nair & Rachael Sugars Graph Learning: from autonomous vehicles to digital pathology
14:55-15:00 Oskar Kviman & Jens Lagergren Ensembles of variational approximations
15:00-15:05 Ricardo Vinuesa Towards a more sustainable aviation through computer simulations and artificial intelligence
15:05-15:10 Martin Jacobsson Deep Learning-Based Early Prediction of Intraoperative Hypotension
15:10-15:15 Emrah Karakaya & Mats Engwall AI and Industrial transformation
15:15-15:20 Qianwen Xu AI for sustainable power systems
15:20-15:25 Bob Sturm MUSAiC: Music at the Frontiers of Artificial Creativity and Criticism
15:25-15:30 Zahra Kalantari Development of novel hybridized models for urban flood susceptibility mapping


15:30 – 15:45  Coffee break

16:10 – 16:55  Poster session (moderators in bold)

Group 4:  Zoom  Meeting ID: 666 6738 6245

Group 4

Learning in Networks

Filip Cornell & Sarunas Girdzijauskas Representing graphs using Random Indexing
Karl Meinke, Arvind Nair & Rachael Sugars Graph Learning of Tissue Models for Digital Pathology
Ruochun Tzeng & Aristides Gionis Discovering conflicting groups in signed networks
Sijing Tu & Stefan Neumann A viral marketing-based model for opinion dynamics in online social networks
Lodovico Giaretta & Sarunas Girdzijauskas Machine Learning on Decentralized Networks
Ahmed Emad & Sarunas Girdzijauskas Machine Learning on Heterogeneous Networks

Group 5:  Zoom  Meeting ID: 639 7301 4901

Group 5


Xiaoxuan Wang & Rolf Stadler Online feature selection for efficient learning in networked systems
Stefano Markidis Automatic Data Exploration and Analysis of Space Physics Data
Asta Kizyte & Ruoli Wang Robust Ankle Torque Estimation Using Electromyography and Long Short-Term Memory Network
Jörg Conradt Spiking Neuronal Networks for Real-Time Systems
Giovanni Calzolari Deep learning to aid CFD simulations in Built environment
Naresh Balaji Ravichandran & Pawel Herman Unsupervised representations learning in brain-like networks with Hebbian-like plasticity
Erik Fransén Machine learning-based unsupervised feature extraction from magnetoencephalography and eye-tracking data

Group 6:  Zoom  Meeting ID: 618 2300 9421

Group 6


Martin Isaksson & Sarunas Girdzijauskas Adaptive Expert Models for Federated Learning
Debaditya Roy & Sarunas Girdzijauskas Confidence Calibrated Human Activity Recognition
Xavier Weiss & Qianwen Xu Energy management of smart home with the electric vehicle using deep reinforcement learning
Mengfan Zhang & Qianwen Xu Data-driven control of smart converters for sustainable power systems using deep reinforcement learning
Hao Hu & Hossein Azizpour Improving Transferability for Self-Supervised Vision Transformers on Small Datasets
Zahra Kalantari Environmental modelling (weather extremes in a changing climate)

Zoom-link to MAIN STAGE:

16:55 – 17:00 Concluding remarks

End of day


Contact persons: