EGPAI 2017: 2nd International Workshop on Evaluating General-Purpose AI IJCAI 2017 Melbourne, Australia, August 20, 2017 |
Conference website | http://users.dsic.upv.es/~flip/EGPAI2017/ |
Submission link | https://easychair.org/conferences/?conf=egpai2017 |
Submission deadline | May 5, 2017 |
The 2nd international workshop on evaluating general-purpose AI (EGPAI 2017) will be held in conjunction with IJCAI 2017 (IJCAI 2017) in Melbourne, Australia (August 20, 2017).
Up to now, most AI systems are tested on specific tasks. However, to be considered truly intelligent, a system should exhibit enough flexibility to find a diversity of solutions for a range of tasks, some of which may not be known until after the system is deployed. Very recently there has been a large number of events, challenges and platforms that are giving a new perspective to how AI can be evaluated, such as the Arcade Learning Environment video games, the Video Game Definition Language (VGDL), OpenAI Gym, Microsoft Malmo, OpenAI Universe, Facebook TorchCarft, Facebook CommAI-env, GoodAI school, Google DeepMind Lab, etc. This workshop will welcome formalisations, methodologies and testbenches for evaluating the numerous aspects of this type of general AI systems. More specifically, we are interested in theoretical or experimental research focused on the development of concepts, tools and clear metrics to characterise and measure the intelligence, and other cognitive abilities, of general AI agents. Furthermore, EGPAI2017 will participate in the IJCAI2017 special theme on AI & Autonomy. Therefore, the workshop will welcome papers on the evaluation of autonomous agents of any kind, such as robots, software agents, artificial life agents, and any sort of autonomous systems capable of operating in long-term, real-world scenarios. There will be a panel dealing with this topic.
We are interested in questions such as: Can the various tasks and benchmarks in AI provide a general basis for evaluation and comparison of a broad range of such systems?, Can there be a theory of tasks, or cognitive abilities, that enables a more direct comparison and characterisation of AI systems? How much does the specificity of an AI agent relate to how fast it can achieve acceptable performance?, How does the structure of a cognitive system relate to how easy or difficult a task - or various classes of tasks - are for it to perform and learn?
Submission Guidelines
We solicit submissions (full or short papers) including:
- Original research contributions
- Applications and experiences
- Surveys, comparisons, and state-of-the-art reports
- Tool or demo papers
- Position papers related to the topics mentioned above.
- Work in progress papers.
Submitted papers must be formatted according to the camera-ready style for IJCAI 2017. Manuscripts must be submitted electronically via the easychair conference management system.
Authors of accepted papers will be asked to present the paper during the workshop. Online pre-proceedings containing all accepted papers will be prepared before the date of the conference. Depending on the number and quality of submissions, we will examine the possibility of targeting a volume or a journal special issue.
Papers are allowed a maximum of six (6) pages, references excluded. References can take up to one additional page. Formatting Guidelines, LaTeX Styles and Word Template can be downloaded from here.
List of Topics
We welcome regular papers, demo papers about benchmarks or tools, and position papers, and encourage discussions over a broad list of topics (not exhaustive):
- Analysis and comparisons of AI benchmarks and competitions. Lessons learnt.
- Proposals for new general tasks, evaluation environments, workbenches and general AI development platforms.
- Theoretical or experimental accounts of the space of tasks, abilities and their dependencies.
- Evaluation of development in robotics and other autonomous agents, and cumulative learning in general learning systems.
- Tasks and methods for evaluating: transfer learning, cognitive growth, structural self-modification and self-programming.
- Evaluation of social, verbal and other general abilities in multi-agent systems, video games and artificial social ecosystems.
- Evaluation of autonomous systems: cognitive architectures and multi-agent systems versus general components: machine learning techniques, SAT solvers, planners, etc.
- Unified theories for evaluating intelligence and other cognitive abilities, independently of the kind of subject (humans, animals or machines): universal psychometrics.
- Analysis of reward aggregation and utility functions, environment properties (Markov, ergodic, etc.) in the characterisation of reinforcement learning tasks.
- Methods supporting automatic generation of tasks and problems with systematically introduced variations.
- Better understanding of the characterisation of task requirements and difficulty (energy, time, trials needed..), beyond algorithmic complexity.
- Evaluation of AI systems using generalised cognitive tests for humans. Computer models taking IQ tests. Psychometric AI.
- Application of (algorithmic) information theory, game theory, theoretical cognition and theoretical evolution for the definition of metrics of cognitive abilities.
- Adaptation of evaluation tools from comparative psychology and psychometrics to AI: Item Response Theory (IRT), adaptive testing, hierarchical factor analysis.
- Evaluation methods for multiresolutional perception in AI systems and agents.
Apart from the technical sessions, we are planning to have a demo session presenting real platforms and ways to evaluate AI systems for several tasks in these platforms; and a panel with a more lively discussion about the research challenges around the workshop topics and future initiatives.
Committees
Program Committee
- Marco Baroni (Facebook AI Research)
- Jordi Bieger (CADIA, Reykjavik University)
- Angelo Cangelosi (Plymouth University)
- Emmanuel Dupoux (EHESS)
- Helgi P. Helgason (Activity Stream)
- Katja Hofmann (Microsoft Research)
- Sean B. Holden (Cambridge University)
- Estevam R. Hruschka (Carnegie Mellon University)
- Armand Joulin (Facebook AI Research)
- Jan Koutnik (IDSIA)
- Edward Keedwell (Exeter University)
- Tomas Mikolov (Facebook AI Research)
- Frans A. Oliehoek (University of Amsterdam)
- Ricardo B.C. Prudencio (Uni. Fed. de Pernambuco)
- Ute Schmid (Bamberg University)
- Bas Steunebrink (IDSIA)
- Pei Wang (Temple University)
Organizing committee
- Nader Chmait (Monash University)
- Jose Hernandez-Orallo (Technical University of Valencia)
- Fernando Martínez-Plumed (Technical University of Valencia)
- Claes Strannegård (Chalmers University of Technology)
- Kristinn R. Thórisson (Reykjavik University)
Invited Speakers
- David L. Dowe, Associate Professor at Monash University (Australia).
- Jan Feyereisl, Executive Director of the AI Roadmap Institute and a Senior Research Scientist at GoodAI.
More invited speakers to be announced.
Venue
The conference will be held in Melbourne, Australia, August 20, 2017 (satellite workshop of IJCAI 2017).
General AI challenge Special Track
Registered participants for the General AI Challenge are welcome to submit a short summary (2-4 pages) explaining their approach for solving the challenge and their experience so far (at the moment of submission). The deadline is June 10th, 2017 (note that the deadline for this special track is different from the main EGPAI2017 submission deadline (May 5th, 2017). EGPAI2017 will host a special session for this track, reporting on the state of the General AI Challenge, including an invited talk given from the challenge organisers (GoodAI) and a few selected short reports from the participants. Only registered participants for the challenge can submit to this special track. This must be done through the EGPAI2017 submission platform, but the title of the paper must start with the following text: "General AI Challenge Participant: ". Submission for the special track is compatible with the submission of other regular papers for EGPAI2017 (under its general deadline and instructions).
The papers for this special track will be lightly reviewed by the EGPAI organisers in coordination with the GoodAI people. Accepted papers of this track will be presented during the workshop and will appear in the workshop proceedings. Papers can focus on the learning methods the participants are using for training their agents but we especially welcome those submissions that touch upon evaluation-related issues (quantitative, validation of agents ability to learn gradually or qualitative, interpretability, white-box analysis of agents) and the main obstacles of the General AI challenge (non-stationarity, catastrophic forgetting, sample complexity, limited training data, generality, etc.).
Contact
All questions about submissions should be emailed to Nader Chmait (nader.chmait@monash.edu)