Important Event Information

Key Dates  
AAAI Early Registration Deadline December 20th, 2023
Course Materials Posted to Website February 1st, 2024
AAAI-24 Tutorial Takes Place in Vancouver, BC, Canada February 21st, 2024 from 8:30am-12:30pm

You can register for AAAI and the Tutorial and Lab Forum at the conference website. We recommend registering early, but it is also possible to register anytime up until the event itself (AAAI-24 allows for late and on-site registration).

Tutorial Description

Overview

A detailed illustration depicts a modern classroom or workshop setting with a large digital display at the front. The screen showcases various machine learning concepts, including Bayesian Models, Decision Trees, Nearest Neighbor, and Clustering. Each concept is visually represented by corresponding graphs and models. A tree-like network structure with vibrant nodes radiates from the center of the screen. In front of the display, an instructor, holding a pointer, explains the concepts to an older individual. Below them, a group of diverse individuals sits around a horseshoe-shaped table, equipped with advanced touch-screen monitors, attentively listening and interacting with their devices. The ambiance is studious, with a futuristic touch symbolizing the advanced nature of the subject matter.

This tutorial will offer an introduction to the Cobweb family of concept formation approaches (Fisher, 1987). This class of methods supports incremental, unsupervised learning from a continuous stream of examples (Fisher, Pazzani, & Langley, 2014). Throughout the tutorial, we explore the core theory underlying Cobweb as well as several extensions and applications (e.g., to language and vision tasks). We will also demonstrate its use through practical examples to real-world scenarios. This tutorial will provide participants with the knowledge and hands-on experience required to effectively apply Cobweb to their own tasks.

Keywords: Concept formation, Incremental learning, Unsupervised Classification, Cobweb

Why have a tutorial on Cobweb?

As humans, we excel at learning knowledge and skills across different domains, forming concepts based on our observations. We then use our concept knowledge to guide classification and learning over future experiences. Creating a model of human concept formation, has been one of the primary areas of focus in machine learning (Gennari, Langley, & Fisher, 1989). The aim of this tutorial is to offer an overview of the Cobweb family of approaches, which adopt a probabilistic formalism and support incremental, unsupervised learning. Our goal is to introduce this approach to a broader audience of researchers in machine learning and demonstrate its practical applications to real-world scenarios.

Throughout this tutorial, we will introduce multiple Cobweb approaches and provide practical guidance on obtaining and running code for each approach. More generally, we wish to present a new framework for continual, human-like learning, offering a unique perspective on how machines can utilize concepts for classifying observations. We will illustrate this through practical applications and conclude by sharing our thoughts on the future of research in Cobweb and human-like learning.

Our tutorial seeks to bring together researchers from diverse backgrounds (e.g., cognitive modeling and machine learning) and applications (e.g., language and vision). By fostering interdisciplinary collaboration, we envision integrating methodologies and ideas from different fields, potentially leading to novel approaches and breakthroughs in AI research. The tutorial will provide an enriching and collaborative learning experience that enhances attendee understanding of incremental concept formation and human-like learning. It will also promote interdisciplinary collaboration–encouraging the integration of different of machine learning paradigms and methodologies.

Schedule

The tutorial will be structured into three parts, with additional Q&A sessions and hand-on activities:

Start End Session Speaker
8:30am 9:15am Overview of Basic Cobweb (35 min talk + 10 min Q&A) Doug Fisher
9:15am 9:30am Hands-On Example of Cobweb with Tabular Data Christopher J. MacLellan
9:30am 10:15am Early Extensions to Cobweb (35 min talk + 10 min Q&A) Pat Langley
10:15am 10:30am Hands-On Example of Cobweb with Continuous Data Christopher J. MacLellan
10:30am 11:00am Break —
11:00am 11:45am New & Future Extensions to Cobweb (35 min talk + 10 min Q&A) Christopher J. MacLellan
11:45am 12:00pm Hands-On Example of Cobweb on a Vision Task Christopher J. MacLellan
12:00pm 12:15pm Hands-On Example of Cobweb on a Language Task Christopher J. MacLellan
12:15pm 12:30pm General Q&A —

Tutorial Example Code

We’ve created a repository specifically for this tutorial here. Please check out the repo before coming and consider trying to install the package (via pip install) beforehand. Links to the specific tutorial activities from this repo are included in the schedule above. You can also find another relevant repo here, however, the main branch of this repo is still being updated and so does not reflect the latest cobweb extensions we plan to discuss (e.g., for language and vision tasks).

Prerequisite Knowledge

This tutorial is tailored to participants with a basic background in computer science and AI/ML. Anyone with intro-level coursework in Algorithms and Machine Learning will be able to fully engage with the tutorial. The emphasis here lies in understanding the concept formation approach, which is based on psychological studies of how humans incrementally develop and utilize concepts.

Organizing Committee

References

  • Fisher, D. 1996. Iterative optimization and simplification of hierarchical clusterings. Journal of artificial intelligence research, 4: 147–178.
  • Fisher, D. H. 1987. Knowledge acquisition via incremental conceptual clustering. Machine learning, 2: 139–172.
  • Fisher, D. H. 1988. A computational account of basic level and typicality effects. In AAAI, 233–238. Fisher, D. H.; Pazzani, M. J.; and Langley, P. 2014. Concept formation: Knowledge and experience in unsupervised learning. Morgan Kaufmann.
  • Gennari, J. H.; Langley, P.; and Fisher, D. 1989. Models of incremental concept formation. Artificial intelligence, 40(1- 3): 11–61.
  • Langley, P. 2022. The computational gauntlet of human- like learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 12268–12273.
  • Lebowitz, M. 1987. Experiments with incremental concept formation: UNIMEM. Machine learning, 2: 103–138.
  • MacLellan, C. J.; Harpstead, E.; Aleven, V.; and Koedinger, K. R. 2015. Trestle: Incremental learning in structured do- mains using partial matching and categorization. In Proceedings of the 3rd Annual Conference on Advances in Cognitive Systems.
  • MacLellan, C. J.; Matsakis, P.; and Langley, P. 2022. Efficient Induction of Language Models Via Probabilistic Concept Formation. arXiv preprint arXiv:2212.11937.
  • MacLellan, C. J.; and Thakur, H. 2022. Convolutional Cob- web: A Model of Incremental Learning from 2D Images. arXiv preprint arXiv:2201.06740.
  • McKusick, K.; and Thompson, K. 1990. Cobweb/3: A portable implementation. Technical report.
  • McKusick, K. B.; and Langley, P. 1991. Constraints on Tree Structure in Concept Formation. In IJCAI, volume 91, 810– 816. Citeseer.
  • Thompson, K.; and Langley, P. 1991. Concept formation in structured domains. In Concept formation, 127–161. Elsevier.