Machine Learning

Research Highlights

Mastering the game of Go

Mastering the game of Go with deep neural networks and tree search

The game of Go has long been viewed as the most challenging of classic games for artificial intelligence. The journal Nature reported on 28 January that researchers at Google Deepmind developed AlphaGo, a program based on neural networks, policy gradient and Monte Carlo tree search that achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0. This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away.

Thore Graepel, one of the authors on the paper, will present the results during the Intelligent Machines day on 22 March 2016 in de Flint in Amersfoort.

http://www.snn.ru.nl/v2/ml2016.php

www.ictopen.nl

http://www.nature.com/nature/journal/v529/n7587/abs/nature16961.html

More about Mastering the game of Go

VIDI grant vor Ming Cao

NWO award a VIDI grant to Ming Cao for his research proposal “An evolutionary approach to coordination of self-interested agents”. Robot and sensor networks, distributed energy grids and many facets of society can be modelled as complex networks of agents making self-interested decisions that often conflict with group objectives. Ming Cao will seek methods for coordinating these networks and ultimately resolving such social dilemmas through a closely-coupled approach of theory and experimentation. More here.

More about VIDI grant vor Ming Cao

VIDI Ivan Titov: Scaling Semantic Parsing to Unrestricted Domains

NWO awarded Ivan Titov (UvA) a VIDI grant for the proposal “Scaling Semantic Parsing to Unrestricted Domains”.

Enabling a machine to understand human language – that is, to process any input text or utterance and be able to answer questions or perform actions on its basis – is one of the main goals of natural language processing. The lack of accurate methods for predicting meaning representations of texts is the key bottleneck for many natural language processing applications such as question answering, text summarization and information retrieval. Although state-of-the-art semantic parsers work fairly well on closed domains (e.g., interpreting natural language queries to databases), accurately predicting even shallow forms of semantic representations for less restricted texts remains a challenge. The reason for the unsatisfactory performance is reliance on supervised learning from human-annotated text collections, with the amounts of annotation required for accurate open-domain parsing exceeding what is practically feasible. In order to deal with this challenge, in this project, we will introduce methods for inducing semantic parsers primarily from un-annotated data (e.g., text on the Web). More here.

More about VIDI Ivan Titov: Scaling Semantic Parsing to Unrestricted Domains
A-Virtual-Neural-Network

VIDI voor Marcel van Gerven

NWO has awarded a VIDI grant to Marcel van Gerven (RU)

Marcel van Gerven and his lab will use computational techniques from Artificial Intelligence to read out internal representations from the brain. He illustrates his approach by reconstructing stimuli that participants were exposed to from brain activity that is recorded with MRI. This line of research will form the basis of new applications within neuroscience. More here.

More about VIDI voor Marcel van Gerven
images-1

ERC Starting Grant for Joris Mooij

Joris Mooij researcher at the Intelligent Autonomous Systems-group of the Informatics Institute, has obtained an ERC Starting Grant for the project ‘Causal Analysis of Feedback Systems (CAFES)’.

Building on recently established connections between dynamical systems and causal models, CAFES will develop theory and machine learning algorithms for modeling, reasoning, discovery and prediction for causal systems involving feedback loops. This will enable large-scale applications of causal inference in various challenging domains in science, industry and decision making. The work will be done with a strong focus on applications in molecular biology, one of the most promising areas for automated causal inference from data.

More about ERC Starting Grant for Joris Mooij
images

ERC Starting Grant Shimon Whiteson

Shimon Whiteson receives an ERC Starting Grant on the topic of coevolutionary policy search.

In order to make intelligent systems (e.g., robots and search engines) autonomous, we need algorithms that can automatically discover high-performing control policies for such systems. However, in many cases, evaluating the performance of candidate policies is complicated by the presence of rare events whose effects on performance are hard to measure. In this project, Shimon and his team will develop new algorithms that exploit the principle of coevolution to simultaneously optimise both control policies and the manner in which those policies are evaluated on rare events. The resulting methods will be applied to realistic tasks in robot control and information retrieval.

More about ERC Starting Grant Shimon Whiteson
imgres

NWO Top Grant

Prof. dr. M. Loog, Technische Universiteit Delft
Linear, Discriminative, Semi-Supervised Classifiers

Learning methods are at the heart of almost any modern computer application. Supervised learning algorithms [e.g. classifiers and decision rules] are able to generalize from examples and predict the desired output to unseen input. A major obstacle in their successful use is the need for sufficient expert-labeled examples to learn from. Semi-supervised learning [SSL] promises to improve radically upon this situation by exploiting both labeled and unlabeled data. To this date, however, SSL has not lived up to this promise, often even deteriorating instead of improving performance. Current methods can be difficult to handle, especially by the non-expert, and they are not as widely used as their supervised counterparts. We therefore need SSL methods that are reliable and can be readily substituted for the supervised classifiers that are en vogue in the different research domains and application areas.
There are vacancy’s as result of this grant, see http://recruitment2.tudelft.nl/vacatures/index.php?lang=en&id=544901&type=a.

More about NWO Top Grant
causal_explorer_logo

NWO Top Grant, Comp 1

Prof. dr. T.H. (Tom) Heskes, Radboud Universiteit Nijmegen
Causal Discovery from High-Dimensional Data in the Large-Sample Limit

Discovering causal relations from data lies at the heart of most scientific research today. Controlled experimentation, the standard and most popular method for causal discovery, is in many cases practically impossible, ethically undesirable, or too costly. About twenty years ago, scientists realized that there is an alternative: under appropriate assumptions, causal knowledge can also be derived from purely observational data. In the ‘big data’ era, such observational data is abundant and being able to actually derive causal relationships from very large data sets would open up a wealth of opportunities for improving business, science, government, and healthcare.
Sadly, existing algorithms for causal discovery from observational data are not very well suited to big data: small changes in the data or in the algorithmic details can lead to significantly different causal conclusions, in particular for data sets containing many different variables and even in the limit of a large number of samples. In this project we aim to tackle these issues through a much better mathematical understanding of the appropriate asymptotic statistics. Effective causal discovery, in one way or another, hinges upon the ability to accurately and efficiently infer sparse models. We will therefore translate and extend existing work in mathematical statistics on sparse model estimation to the domain of causal inference. Improved mathematical understanding guides the development of novel, more reliable algorithms for causal discovery from big data that can control the false causal discovery rate. We demonstrate the usefulness of these algorithms on real-world problems in genomics and ecology.

More about NWO Top Grant, Comp 1
nai_robot

NAI Program

In the Natural Artificial Intelligence call, NWO EW awarded six research teams grants for advancing state-of-the-art AI. The NAI program has a multidisciplinary aim, and consists of research-projects that aim to strengthen the link between natural intelligence and artificial intelligence, combining insights either area to reinforce the other. The projects specifically target advances in deep learning, cognitive robotics, and models of efficient learning in brain-like structures.

The awarded projects include:

  • Learning the Fundamental Symmetries in Video Data (M. Welling (UvA) and L.P.J. van der Maaten (TUD))
  • Reward-based learning of Subroutines by Neural networks (P.R. Roelfsema (NHI) and S.M. Bohté (CWI))
  • Deep Learning for Robust Robot Control (R. Babuska (TUD) and K. Tuyls (TUD))
  • Learning to Communicate via Social and Linguistic Interaction (P.A. Vogt (UvT) and A. Alishahi (UvT))
  • Deep Spiking Vision: Better, Faster, Cheaper (S.M. Bohté (CWI) and S. Ghebreab (UvA), H.S. Scholte (UvA))
  • Decentralized UAV Control (G.C.H.E. de Croon (TUD) and H.J. Kappen (RU), K. Tuyls (TUD))
More about NAI Program

Dr. S.A. (Shimon) Whiteson (m), UVA – Informatica – VIDI -2014

Coevolutionary Policy Search: in order to make intelligent systems (e.g., robots and search engines) autonomous, we need algorithms that can automatically discover high-performing control policies for such systems. However, in many cases, evaluating the performance of candidate policies is complicated by the presence of rare events whose effects on performance are hard to measure. In this project, Dr. Shimon Whiteson and his team will develop new algorithms that exploit the principle of coevolution to simultaneously optimise both control policies and the manner in which those policies are evaluated on rare events. The resulting methods will be applied to realistic tasks in robot control and information retrieval.

More about Dr. S.A. (Shimon) Whiteson (m), UVA – Informatica – VIDI -2014

News & Events

News


You can register for e-mail list for academic announcements about machine learning in the Netherlands:

https://lists.leidenuniv.nl/mailman/listfo/machine_learning_nederland

The goal is too improve communication within our community: sharing announcements about departmental talks, and for interesting international visitors giving a talk, please do drop a note to the list.
Announcements about other meetings, about job openings or about any related topics are also appropriate.
Subscribing to the list is open to everyone, and posting is open to all subscribers.

Events

Recorded Events

OPENING QUVA-Lab. On Tuesday September 15, the University of Amsterdam UvA and Qualcomm Technologies Inc. (QTI) opened a joint research lab focused on advancing the state of the art in machine learning techniques for mobile computer vision. The opening will include a festive program and a lecture by deep learning guru Yoshua Bengio. More information here: https://ivi.fnwi.uva.nl/quva/.

Intelligent Machines 2015 On Tuesday March 17, SNN organized a one day symposium entitled Intelligent Machines. Slides of Zoubin Ghahramani (Cambridge University), Sethu Vijayakumar (Edinburgh University) and Ralf Herbrich (Amazon Research Berlin) can be found on the website. Daan Wierstra (Google Deep Mind) had an interview with Bennie Mols (in Dutch), and Deep mind published a paper in Nature.

Intelligent systems track at ICT Open 2016.During the ICT.OPEN2016 congress there was a novel track Intelligent Systems.
This track was motivated by the spectacular recent developments in machine learning and artificial intelligence, both in science and in society.
Large-scale Online Learning and Decision Making Workshop, (LSOLDM),Cumberland Lodge 2013.

Big data and machine learning: Inaugural lecture Max Welling,
[slides ppt][slides pdf][written text][spoken text][video]
UvA, 29-01-2014 (In Dutch).