The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index.
Neural network design and the complexity of learning
Open accessAbstract
We formalize a notion of learning that characterizes the training of feed-forward networks. In the field of learning theory, it stands as a new model specialized for the type of learning problems that arise in connectionist networks. The formulation is similar to Valiant's [Va184] in that we ask what can be feasibly learned from examples and stored in a particular data structure. One can view the data structure resulting from Valiant-type learning as a 'sentence' in a language described by grammatical syntax rules. Neither the words nor their interrelationships are known a priori. Our learned data structure is more particular than Valiant's in that it must be a particular 'sentence'. The position and relationships of each 'word' are fully specified in advance and the learning system need only discover what the missing words are. This corresponds to the problem of finding retrieval functions for each node in a given network. We prove this problem NP-complete and thus demonstrate that learning in networks has no efficient general solution. Corollaries to the main theorem demonstrate the NP-completeness of several sub-cases. While the intractability of the problem precludes its solution in all these cases, we sketch some alternative definitions of the problem in a search for tractable sub-cases. One broad class of subcases is formed by placing constraints on the network architecture; we study one type in particular. The focus of these constraints is on families of 'shallow' architectures which are defined to have bounded depth and unbounded width. We introduce a perspective on shallow networks, called the Support Cone Interaction (SCI) graph, which is helpful in distinguishing tractable from intractable subcases: When the SC1 graph has tree-width O (log n), learning can be accomplished in polynomial time; when its tree-width is n omega (1) we find the problem NP-complete even if the SC1 graph is a simple 2-dimensional planar grid.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Language identification in the limit
Gold, E. M. (1967) · Information and Control · doi:10.1016/s0019-9958(67)91165-5 · Cited by 3,572
Dynamic node creation in backpropagation networks
ASH, T. (1989) · Connection Science · doi:10.1080/09540098908915647 · Cited by 415
Cited by
Deep learning in neural networks: an overview
Schmidhuber, J. (2015) · Neural Networks · doi:10.1016/j.neunet.2014.09.003 · Cited by 18,570
Computational Intelligence: An Introduction
Fulcher, J. (2008) · Studies in Computational Intelligence · doi:10.1007/978-3-540-78293-3_1 · Cited by 629
Handbook of Perception and Cognition
(1994) · Animal Learning and Cognition · doi:10.1016/b978-0-08-057169-0.50001-2 · Cited by 572
Training a 3-node neural network is NP-complete
Blum, A. L., & Rivest, R. L. (1992) · Neural Networks · doi:10.1016/s0893-6080(05)80010-3 · Cited by 543
Statistical aspects of neural networks
Ripley, B. D. (1993) · Networks and Chaos — Statistical and Probabilistic Aspects · doi:10.1007/978-1-4899-3099-6_2 · Cited by 504
Metadata
- Venue
- Choice Reviews Online
- Publisher
- American Library Association
- Published
- 1991-02-01
- Volume
- 28
- Issue
- 06
- Pages
- 28-3325-28-3325
- ISSN
- 0009-4978 (ISSN-L)
- 1523-8253
- OpenAlex
- W2134331967
- Cited by
- 241
- References
- 2
- Type
- book review
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Neural network design and the complexity of learning. https://app.cochat.ai/papers/10.5860/choice.28-3325
- The paper
- Neural network design and the complexity of learning. (1991). Choice Reviews Online. https://doi.org/10.5860/choice.28-3325
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.