Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Rule Extraction from Trained ANN: A Survey - Darbari (2001) (Correct)
....hidden layer in between the input and the output layer. The rules for the individual units are then combined into a set of rules that describe the network as a whole. Local methods are computationally less expensive than global methods. A slightly di erent notion on rule extraction is suggested by [5], it is a generalization of the notions mentioned above. De nition 2 Given a particular set of weights W ij and biases i , resulting from a training process on a neural network, nd for each input vector i, all outputs o j in the corresponding output vector o such that o j A min , where A ....
....where A min 2 (0,1) is a prede ned value (we say that output neuron j is active for input vector i i o j A min ) A min is value guided by heuristics, and is such a value where the transition of a unit from a non active state to an active state takes place. It is interesting to observe that [5] has a slightly di erent formulation of the rule extraction problem and also a di erent approach towards the rule extraction task than [13] however both the groups who have been very successful have identi ed some key criteria for evaluation of algorithms and also they share the common notion of ....
[Article contains additional citation context not shown here]
A. S. d'Avila Garcez, K. Broda, and D. Gabbay. Symbolic Knowledge Extraction from Trained Neural Networks: A New Approach. Technical Report TR-98-014, Department of Computing, Imperial College, London, 1998.
Extracting Symbolic Rules from Trained Neural Network Ensembles - Zhou, Jiang, Chen (2003) (2 citations) (Correct)
....trees from trained neural networks with the help of an attribute selection criterion based on significance analysis. Tanaka et al. 40] extract linguistic rules through feeding rule antecedents to a network and then comparing the fuzzy outputs of the network with linguistic values. Garcez et al. [13] define a partial ordering on the set of input vectors and then extract nonmonotonic rules from the network. 3. REFNE 3.1. Motivation REFNE is a function analysis based approach. Suppose that there is a trained neural network ensemble E . If an input vector A k = a 1 , a 2 , a n ) is fed ....
A. S. D. Garcez, K. Broda, and D. M. Gabbay, Symbolic knowledge extraction from trained neural networks: A sound approach, Artificial Intelligence 125 (2001): 155-207.
HTRP II: Learning thematic relations from semantically sound.. - Rosa (2001) (Correct)
....is revised by the connectionist learning. The symbolic knowledge generated by the net can be extracted, in HTRP II, in a way comparable to initial symbolic knowledge insertion in BIW II. It had been proved that the set of rules and the network, from which it is extracted, are very equivalent [1]. Although many researchers believe that symbolic and connectionist systems are so different that they are irreconcilable, others emphasize that the integration of both is not only possible but also crucial for the systems understand cognition behind the computational implementations. Honavar ....
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence 125, 155-207, 2001.
Ontology Learning as a Use-Case for Neural-Symbolic.. - Hitzler, Bader, Garcez (2005) Self-citation (Garcez) (Correct)
No context found.
Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
A Connectionist Inductive Learning System For - Garcez, Gabbay (2002) Self-citation (Garcez Gabbay) (Correct)
No context found.
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
Combining Abductive Reasoning and Inductive Learning .. - Garcez, Russo.. (2003) Self-citation (Garcez) (Correct)
....provide explanation capability to neural networks in the form of symbolic rules. This task is known as rule extraction from trained neural networks [2] and although it is exponential on the number of input neurons in the worst case, it works well in practice for considerably large neural networks [7]. In addition, note that induction alone cannot guarantee that D satisfies P, as it normally includes some generalisation of the training examples. As a result, the (sound and complete) abductive mechanism of the analysis phase [40] should be responsible for verifying whether D satisfies P. This ....
....input sequences, and generating rules according to the output sequence obtained. The core of C IL2p s rule extraction algorithm is concerned with the selection of good candidate input sequences to be presented to iV, so that the network can be described by a correct and compact set of rules [7]. 4 Evolving Specifications In this section, we describe how abductive reasoning and inductive learning can be combined to, respectively, analyse and revise specifications. We present an automated formal reasoning process that interleaves analysis and revision phases, eventually stopping when no ....
[Article contains additional citation context not shown here]
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155-207, 2001.
First-Order rule learning through a Pulsed Neural Network - Darbari (2000) Self-citation (Broda) (Correct)
....One of the typical algorithm found very useful for training a network obtained from a propositional logic program is the backpropagation algorithm. Several groups have exploited the standard backpropagation or one of its variants to train such kind of Knowledge Based Networks [113] 121] [21], 7] After these networks are trained on a rather noisy data set, one focuses on the rule extraction part. At the level of propositional logic, there are several solutions proposed to the problem of rule extraction. The solutions are robust in the sense that they can also deal with noisy data ....
....of rule extraction. The solutions are robust in the sense that they can also deal with noisy data and also fulfill some of the nice properties like expressivity, translucency etc [20] However, the issue of comprehensibility is not so well addressed. Even the very best and the latest work by [21] fails to reduce the size of the rule set. This in turn has serious implications because if the neural network was criticized for being abstruse, then the solution coming from the hybrid learning schemes is at least so huge in size that even though they are logical if then propositional rules, one ....
[Article contains additional citation context not shown here]
A. S. d'Avila Garcez, K. Broda, and D. Gabbay. Symbolic Knowledge Extraction from Trained Neural Networks: A New Approach. Technical Report TR-98-014, Department of Computing, Imperial College, London, 1998.
A Connectionist Inductive Learning System for Modal Logic.. - Garcez, Lamb, Gabbay Self-citation (Garcez Gabbay) (Correct)
....N computes the least fixp oi t of P . Thi s provi s a massi vely parallel model for computi ng the stable model semanti cs of P [28] In addi ti on, N can be trai4E wiE5 mples usi ng Bac ropagation [35] havi ng P as background knowledge. The knowledge acqui red bytrai ni can then be extracted [13], closi ng the learni ng cycle (asi n [39] Let us exempli fy how C IL P s Translation Algorithm works. Eachrule (r l ) s mapped from thei nput layer to the output layer of N through one neuron (N l)i nthesi nglehi ddenlayerofN .Intui ti vely,theTranslation Algorithm from P to N has to i ....
....learni ng possi ble world representati onsi n the C IL P ensemble. Thi s would lead us to anotheri nteresti ng avenue of research, namely, rule extracti n from neural networks ensembles, whi ch would need to cons i er extracti on methods for more expressi ve knowledge representati on formali ms [13]. In addi ti on, extensi ons of the basi modal C IL n thi s paperi nclude the study of how to represent properti es of other modal logi cs, e.g. S4 and S5, and ofi nference and learni ng of fragments of first order logi . Fi2 ly, the addi ti7 of modalih es to the C IL P system leads us towards ....
A. S. d'Avila Garcez, K. Broda, and D. M Gabbay. Symbolic knowledge extraction from trained neuralnetworks: A sound approach. Artificia Intelligence, 125:155--- 207, 2001.
Integrating First-Order Logic Programs and Connectionist.. - Bader, Hitzler, Witzel (2005) (Correct)
No context found.
Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
Extracting Reduced Logic Programs from Artificial Neural.. - Lehmann, Bader, Hitzler (2005) (Correct)
No context found.
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 126(1--2):155--207, 2001.
The Integration of Connectionism and First-Order.. - Bader, Hitzler.. (2004) (Correct)
No context found.
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
Dimensions of Neural-symbolic Integration - A Structured Survey - Bader, Hitzler (Correct)
No context found.
A. S. d'Avila Garcez, K. Broda and D. M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
Extracting Symbolic Rules from Trained Neural Network Ensembles - Zhou, Jiang, Chen (2003) (2 citations) (Correct)
No context found.
A. S. D. Garcez, K. Broda, and D. M. Gabbay, Symbolic knowledge extraction from trained neural networks: A sound approach, Artificial Intelligence 125 (2001): 155-207.
Extracting Symbolic Rules from Trained Neural Network Ensembles - Zhou, Jiang, Chen (2003) (2 citations) (Correct)
No context found.
A. S. D. Garcez, K. Broda, and D. M. Gabbay, Symbolic knowledge extraction from trained neural networks: A sound approach, Artificial Intelligence 125 (2001): 155-207.
Logic Programs and Connectionist Networks - Hitzler, Hölldobler, Seda (2004) (Correct)
No context found.
Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
Corollaries on the fixpoint completion: studying the stable.. - Hitzler (2004) (Correct)
No context found.
Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Arti cial Intelligence, 125:155-207, 2001.
Logic Programs, Iterated Function Systems, and Recurrent.. - Bader, Hitzler (Correct)
No context found.
Artur S. d'Avila Garcez, Krysia Broda, and Dov M. Gabbay. Symbolic knowledge extraction from trained neural networks: A sound approach. Artificial Intelligence, 125:155--207, 2001.
October 2003. Washington, District of Columbia, United.. - Abiologically.. (Correct)
No context found.
A. S. d'Avila Garcez, K. Broda, and D. M. Gabbay, "Symbolic Knowledge Extraction from Trained Neural Networks: A Sound Approach", Artificial Intelligence 125, pp. 155-207, 2001.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC