Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Are multi-layer backpropagation networks catastrophically amnesic?

Makoto Yamaguchi¹

¹Waseda University, Tokyo 169-8050, Japan. yamag-psy@kurenai.waseda.jp

Scandinavian Journal of Psychology

|November 13, 2004

Summary

Connectionist models trained sequentially often forget previous learning, a problem called catastrophic interference. However, new simulations show this interference is modest with random inputs, suggesting the problem may be overstated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Real-world outcomes of avacopan beyond the first year in antineutrophil cytoplasmic antibody-associated vasculitis: a retrospective cohort study.

BMC rheumatology·2026

Same author

Editorial: Advances in limb-salvage surgery and reconstruction for pediatric bone and soft tissue tumors.

Frontiers in pediatrics·2026

Same author

Effects of glucagon-like Peptide 1, glucagon, and Glucose-Dependent Insulinotropic Polypeptide on Fluid Secretion in Guinea-Pig Pancreatic Duct Cells.

Pancreas·2026

Same author

Association Between Serum Syndecan-1 Levels and Relapse in Patients With Antineutrophil Cytoplasmic Antibody-associated Vasculitis: A Medical Record-based Cohort Study in Japan.

Journal of clinical rheumatology : practical reports on rheumatic & musculoskeletal diseases·2026

Same author

Single-cell transcriptomics reveal heat shock protein dysregulation in severe SARS-CoV-2-associated pediatric encephalopathy.

Scientific reports·2026

Same author

Long-term follow-up after en bloc resection of the distal radius with reconstruction using ulnar translocation.

Case reports in plastic surgery & hand surgery·2025

Area of Science:

Artificial Intelligence
Cognitive Science
Computational Neuroscience

Background:

Connectionist models using backpropagation learning are prone to catastrophic interference.
Sequential training can cause significant performance degradation on previously learned patterns.

Purpose of the Study:

To re-evaluate the severity of catastrophic interference in connectionist models.
To investigate the impact of input orthogonality and model architecture on sequential learning.

Main Methods:

Four simulations were conducted using connectionist models.
Models were trained sequentially on arithmetic facts with varying input properties.
The effect of reducing hidden units was also examined.

Related Experiment Videos

Main Results:

Sequential training resulted in only modest interference when inputs were random (orthogonal).
Adding irrelevant elements to inputs did not significantly increase interference.
Reducing the number of hidden units had minimal impact on interference levels.

Conclusions:

The catastrophic interference problem in connectionist models may be overstated, particularly with random inputs.
Input characteristics play a crucial role in mitigating sequential learning interference.
Further research is needed to fully understand the conditions under which interference occurs.