Quick Facts

Overview: Robotics

Creating autonomous robots that can assist humans in situations of daily life is a great challenge for machine learning. While this aim has been a long standing vision of robotics, artificial intelligence, and the cognitive sciences, we have yet to achieve the first step of creating robots that can accomplish a multitude of different tasks, triggered by environmental context or higher level instruction. Despite the wide range of machine learning problems encountered during the automatic acquisition of new abilities for robots, we have yet to fulfill the promise that modern learning approaches offer in this context. Creating complex learning systems that endow robots with the ability to autonomously learn new skills serves both as source of new ideas and benchmark for machine learning approaches.

Goals in the Near Future: It may among the most promising ways to take robots out of the research labs and bring them into real-world environments. Specific objectives are:

1. Bring robot learning problems to the attention of the machine learning community!
2. Efficient, scalable skill learning approaches based on model learning, imitation learning, apprenticeship learning, reinforcement learning.
3. Complex motor learning tasks requiring hierarchical structuring.
4. Application in high-dimensional, anthropomorphic robotics in real-time.

The goal of this thematic programme, funded by PASCAL2, is to foster such interactions by organizing a series of workshops at robotics as well as machine learning venues. These workshops aim to bring together people that are interested in robotics as a source and inspiration for new Machine Learning challenges, or which work on Machine Learning methods as a new approach to robotics challenges.

Concurrently, the thematic programme also provides the opportunity to apply for external funding of workshops in agreement with the scope of the programme.

Overview: Cognitive Architectures

This theme focuses on the design of intelligent systems, which deal with tasks normally requiring human intelligence. A core aim is to build links between machine learning and cognitive science. Both fields stand to gain substantially from this interaction. First, machine learning provides a powerful computational toolbox for building models of cognition. Indeed, ideas from machine learning and related fields in computer science have provided the foundation of the entire field of cognitive science since its inception over 50 years ago. It is therefore crucial importance for the health of the discipline that the relationship with current developments in machine learning is refreshed and renewed. Second, cognitive science provides insights into human and animal intelligence that may direct and stimulate machine learning research. This direction of influence is part of a more general movement, drawing on insights from biology to build engineering solutions. But the relationship between building intelligent artificial systems to carry out functions normally conducted by human intelligence, and the study of the cognitive processes underlie human intelligence, is especially rich. This is because many of the tasks of artificial systems are defined in relation to the outputs of the human mind: thus, many problems in machine vision are characterised as involving the reputation of a human description of visual scene, in terms of objects, colours, shapes, movement and even causality and intention. Similarly, machine learning methods applied to natural language typically aim to produce translations, search responses, or inferences, that are deemed appropriate by human users.

Success so far: The main activity of the theme has been the organization of a major summer school between 6-12 May, 2010, in Pula, Sardinia, bringing together leading researchers from the machine learning and cognitive science community. The aim of the summer school was two-fold: to provide tutorial introductions, to engage students and researchers in each of the two disciplines, with the problems and methods of the other (this was the focus of the first three days of the conference); and to provide an inspiring introduction to current leading research at the interface of cognitive science and machine learning, as presented by some of its leading practitioners. It proved to be possible to attract a truly impressive range of world leading speakers, from the US, Israel, and Europe. The conference website, with further details see here. Videos of the entire series of lectures are now online.

Upcoming Events:

1. Kick-Off of the EU Project ComPLACS
2. ICML 2011 Workshop on New Developments in Imitation Learning

Application

The thematic programme (TP) provides the opportunity for researchers (independently of their association with PASCAL2) to apply for workshop funding, if the topic of the proposed workshop is in agreement with the scope of the TP. Please note, however, that only in exceptional cases full workshop expenses can be covered through the TP, i.e., other sources of funding are mandatory. To submit an application for funding, please send a PDF to jrpeters (at) tuebingen.mpg.de, containing the following material:

1. Topic of and motivation for the proposed workshop (one page maximum)
2. Workshop organization committee
3. Venue & dates
4. Preliminary schedule
5. Invited speakers (if already available)
6. Itemized funding requests
7. List of other funding sources (if already available)

Please note that preference will be given to workshop proposals that emphasize open discussions rather than organizing mini-conferences!

Sponsor

PASCAL2 is a Network of Excellence funded by the European Commission under the 7th Framework Program. It stands for "Pattern Analysis, Statistical modelling and ComputAtional Learning 2", and unites European research groups with the purpose of advancing the state-of-the-art and promoting awareness of machine learning methods across scientific communities.

External Support

The PASCAL2 Thematic Programme on Machine Learning for Autonomous Skill Acquisition in Robotics is supported from the robotics community as well, e.g., through the IEEE RAS Technical Committee on Robot Learning.