Neural Abstract Reasoner
ISEF · 2021 Robotics and Intelligent Machines Fourth Award
Overview
Abstract reasoning and logic inference are difficult problems for neural networks, yet essential to their applicability in highly structured domains. In this work, the Neural Abstract Reasoner (NAR) is introduced, a memory-augmented architecture capable of learning and using abstract rules. It is shown that, when trained with spectral regularization, NAR achieves 61.13% accuracy on the Abstraction and Reasoning Corpus, while also implicitly learning a rigid domain-specific language structure. Intuition is provided for the effects of spectral regularization in the domain of abstract reasoning based on theoretical generalization bounds and Solomonoff's theory of inductive inference.
Awards (3)
- Fourth Award of $500 $500
- Innopolis University : Full tuition scholarships for the Bachelor program in Computer Science
- Association for the Advancement of Artificial Intelligence: Honorable Mention
Competition history
- ISEF 2021
Resources
Related projects
ISEF · 2026
MERIT: Mechanistic Explainability of Reasoning Integrity and Transparency
ISEF · 2017
Providing a Method for Neural Networks to Justify Their Conclusions in Both Prediction and Classification Problems
ISEF · 2025
Evolving Neural Networks to Exploit Problem Processing Regularities
ISEF · 2020
Memory-augmented Neural Agents for Deep Model-free Reinforcement Learning
Closest projects by meaning, across every fair and year in the corpus.
Source: Regeneron International Science and Engineering Fair