https://www.dagstuhl.de/21362

September 5 – 10 , 2021, Dagstuhl Seminar 21362

Structure and Learning

Organizers

Tiansi Dong (Universität Bonn, DE)
Achim Rettinger (Universität Trier, DE)
Jie Tang (Tsinghua University – Beijing, CN)
Barbara Tversky (Columbia University – New York, US)
Frank van Harmelen (VU University Amsterdam, NL)

For support, please contact

Susanne Bach-Bernhard for administrative matters

Shida Kunz for scientific matters

Dagstuhl Reports

As part of the mandatory documentation, participants are asked to submit their talk abstracts, working group results, etc. for publication in our series Dagstuhl Reports via the Dagstuhl Reports Submission System.

Documents

List of Participants
Shared Documents
Dagstuhl Seminar Wiki

(Use personal credentials as created in DOOR to log in)

Motivation

Structure and learning are among the most prominent topics in Artificial Intelligence (AI) today. Structure is traditionally centered in symbolic AI, learning is currently centered in neural networks. How structures and learning can be integrated has been a grand challenge for decades. This Dagstuhl Seminar is intended to discuss the roadmap of the integration of structure and learning from an interdisciplinary perspective.

In this seminar, we will start with a talk about Mind in Motion, give examples on how cognitive collages are used for spatial thinking, and how abstract thinking is rooted in spatial thinking. We will then demonstrate how a novel explainable unsupervised neural network, equipped with a meta spatial structure, can rigorously carry out all types of syllogistic reasoning. This introduces a new perspective to the relation between symbolic AI and connectionist AI. We will revisit the two competing and complementary AI paradigms, showing that each simulates one aspect of the mind, System 1 and System 2. Recent progress in geometrical embeddings, either hyperbolic, euclidean, or neural, has shown that structures can be precisely embedded as spatial relations among ball embeddings, while features learned by neural networks can be represented as axis vectors around which balls can rotate (ball in motion). We plan to discuss Cognitive Graph — a form of neuro-symbolic AI, the cognitive System 1 corresponds to the machine learning-based AI and the System 2 corresponds to symbolic-based AI. This will shape “the third-generation of AI” with merits of both symbolic AI and neural networks: explainable, reliable, rigorous, robust, safe, and trustworthy. These features separate “the third-generation of AI” from direct neural-symbolic integration that builds rough bridges between symbolic and subsymbolic worlds. Machine learning will be dedicated to acquiring geometric embeddings from the triad of knowledge (background or interactive knowledge), data (ranging from sensors, to texts, audio, and images), and computing architectures, and realizes a human-centered AI ecosystem. It constructively learns a universe of vectors, static or rotating manifolds in an embedding space, solves problems that neither symbolic AI approach nor neural approach can solve alone, and finally realizes Motion in Mind of Machines, which, together with Mind in Motion, schematize a collage of hybrid human-machine intelligence (Hybrid Intelligence).

We will discuss research questions in a variety of fields of AI, Computer Science, and Cognitive Science, such as Knowledge Graph Reasoning, Natural Language Understanding, Computational Humor, Formal Ontologies, Qualitative Spatial Representation (QSR), Semantic Web, StarAI, Software Engineering, IoT, Computer Vision, and Robotics.

Motivation text license
  Creative Commons BY 3.0 DE
  Tiansi Dong, Achim Rettinger, Jie Tang, Barbara Tversky, and Frank van Harmelen

Classification

  • Artificial Intelligence
  • Machine Learning
  • Symbolic Computation

Keywords

  • Neural-symbol unification
  • Knowledge graph
  • Machine learning

Documentation

In the series Dagstuhl Reports each Dagstuhl Seminar and Dagstuhl Perspectives Workshop is documented. The seminar organizers, in cooperation with the collector, prepare a report that includes contributions from the participants' talks together with a summary of the seminar.

 

Download overview leaflet (PDF).

Publications

Furthermore, a comprehensive peer-reviewed collection of research papers can be published in the series Dagstuhl Follow-Ups.

Dagstuhl's Impact

Please inform us when a publication was published as a result from your seminar. These publications are listed in the category Dagstuhl's Impact and are presented on a special shelf on the ground floor of the library.