A series of illustrated story books created in conjunction with the immersive machine intelligence 124C41 at MobileTimes & MobileTimesToday. The book series is available for the Kindle, Nook, and as a paperback.
Coevolutionary Algorithms and Coevolving Agents
13 July, 1999
Colin G. Johnson and Bjorn Olsson and Steve Romaniuk
Orlando, Florida, USA
GECCO-99WKS Part of wu:1999:GECCOWKS
Colin C. Johnson, Bjorn Olsson, Steve Romaniuk
Coevolutionary algorithms and coevolving agents
GECCO-99 Proceedings of the Genetic and Evolutionary Computation Conference
Bird-of-a-feather Workshops
July 13, 1999
Romaniuk, S.G., Extracting Knowledge from a Hybrid Symbolic Connectionist Network, Ph.D. Dissertation, Department of Computer Science and Enginnering University of South Florida, Tampa, 1991.
Romaniuk, S.G., "FUZZNET: A Fuzzy Connectionist Expert System", Master Thesis, Dept. of Computer Science and Engineering, Univ. of South Florida, Tampa FL, 1989
Parallel Connectionist Expert Systems, IASTED Conference on Expert Systems Theory and Applications, Zurich, Switzerland, June 1989. (With S. Romaniuk).
A Study of Machine Learning Approaches for some Classification Knowledge Bases, 4th Florida AI Research Symposium 1991, April, Cocoa Beach, pp. 125-129. (with S. Romaniuk and H. Lee)
Learning with Fuzzy Examples, The Fourth International Fuzzy Systems Association Symposium,Brussels, Belgium, pp. 50-53, 1991, (with S. Romaniuk and H. Lee).
The Use of Connectionist Networks to Recognize Airplanes from Radar Returns, Artificial Neural Networks in Engineering ’91, St. Louis, Mo., pp. 921-926, Nov. 1991. (with S. Romaniuk, J.Leonard, and R. Mitchell)
1991 · The Use of Connectionist Networks to Recognize Airplanes from Radar Returns Lawrence O. Hall and Steve G. Romaniuk Department of Computer Science and Engineering University of South Florida Tampa Intelligent Engineering Systems Through Artificial Neural Networks, Page 921
The LMHash Code - A World-Wide Computer Security Risk
Steve G. Romaniuk
National University of Singapore
Unpublished Technical Report
December 1993
Abstract
This report details the inherent limitations of the lmHash code, which is used in many computer systems around the world, as a user login authentication mechanism. The security community has since its inception held the view, that this authentication method is unbreakable. In this article, we will show - based on an empirical study - that a hybrid symbolic, connectionist network (SC-Net) can be trained to help speed-up the process of cracking lmHashes, significantly.
The experimental setup involves training SC-Net on a small sub-list of words drawn from an English language word list. Words would be selected in an equidistant manner (Every k-th word out of a total of n words would become part of the training set.). The lmHash of each drawn word would become the input and the actual word the output (of course properly encoded to match SC-Net's system requirements) for each training pattern. After training, the trained network would be evaluated on all words except those that are part of the training set. Upon presenting an lmHash as input, the trained network would output the corresponding guessed word (actually the encoded letters). To evaluate the quality of the response, the Hamming distance between the network guessed word and the actual word for an lmHash would be calculated. To measure the performance gain of the network guessed word over a traditional brute force search would be initiated.
Multiple iterations of the aforementioned algorithm revealed a significant speed-up in guessing an initial word from an lmHash and afterwards performing a comprehensive dictionary search (up/down interleaved) for the actual word over a traditional brute force only approach.
The findings clearly show that using SC-Net to initially guess the word associated with an lmHash and then using that word as a starting point for a final brute force search can significatly speed-up the overall search time, putting the world-wide computer system at risk.
NASA. Johnson Space Center, Proceedings of the Third International Workshop on Neural Networks and Fuzzy Logic, Volume 1
@inproceedings (romaniuk:1999:F, author = {Steve G. Romaniuk}, title = {From agent collaboration and communication to speciation and simplified software design}, booktitle = {Coevolutionary Algorithms and Coevolving Agents}, year = {1999}, editor = {Colin G. Johnson and Bjorn Olsson and Steve Romaniuk}, pages = {5--7}, address = {Orlando, Florida, USA}, month = {13 July}, note = {}, keywords = {}, url = {}, abstract = {}, notes = {GECCO-99WKS Part of wu:1999:GECCOWKS},
Victor, P., Romaniuk, S., Perez, R. and Hall, L.O., “Evaluation of some Inductive Algorithms for Automatic Knowledge Acquisition”, Third Florida Conference on Computer Integrated Eng. and Manufacturing, Tampa, Fl., pp. 51-57, Nov. 1990.
Learning fuzzy information in a hybrid symbolic, connectionist model
by SG Romaniuk · 1993 · Steve Romaniuk and Lawrence O. Hall
Discusses the application of SC-net to segmenting MRI images.
Romaniuk, Steve G., Hall, L. O., “Decision Making On Creditworthiness Using a Fuzzy Connectionist Expert System Development Tool”, Proc. INNC 90, Paris, France, 1990.
Through The Looking Glass
White paper
Steve G. Romaniuk
2002
Universal Problem Solvers, Inc.
Where do we go from here?
One interesting application of multi-shot learning is data mining. The goal is to extract implicit, previously unknown information from large databases. These databases can be spread over many different computer platforms across the global internet. Today about two million computers are hooked up to this vast network. Over the next years, these numbers will increase even more rapidly then they have in the past. As more people around the world join the information highway, they will contribute new data, but even more importantly, require the means to access the sheer abundance of available information. It will be necessary to supply tools with which people can readily tap into these new resources of stored data. Automatic retrieval of information on multitudes of subjects will be commonplace. One way of achieving this goal, is to supply end-users with intelligent expert networks.
Expert Networks
Expert networks represent a simple and efficient way of storing vast amounts of data (as can be found in large databases), within one or more neural networks. These networks can be utilized much like traditional expert system knowledge bases. The two main differences being: First, the knowledge base is automatically constructed from data bases by utilizing one or more autonomous agents (i.e., no human supervision). Second, these expert networks are integrated within a single, coherent network structure. This allows expert networks to share possibly common knowledge, much like human experts can chose to cooperate amongst one another and exchange their experiences. Sharing of information is not necessarily restricted to similar data, but can - like in TDL - encompass diverse domains. Let us next look at a possible scenario in which truly autonomous TDL-like agents scour the information highway in search of data that can be mined and extract relevant information, which is then stored in a central repository consisting of expert networks.
Autonomous Data Mining Agents
Imagine for a moment that a central computer system exists with large amounts of storage (beyond Tera Byte range). We will refer to this fictious place, as the Data Mining Headquarter (i.e., DMHQ). One of the tasks of the DMHQ is to send out so called Surveyors -programs whose sole intent is to discover new databases located anywhere on the Internet. Once a Surveyor discovers a database which has not been mined thus far, or one that has not been visited in a long time, it proceeds with taking small samples of the stored data. As soon as the sampling task has been completed, it relays its finds back to the DMHQ. Here, another process transforms the raw data into a collection of pattern files, not unlike those utilized by TDL. Next, the patterns are presented to the current Repository of Expert Networks for identification. In other words, the Expert Networks attempt to classify the patterns, by recognizing as many of them as they possibly can. Just like in TDL, those network units which display the highest degree of recognition are selected. Besides maintaining the Expert Network Repository, the DMHQ also keeps informed about which Data Miner (a program much like TDL) was responsible for mining the data which gave rise to the earlier selected network units. In the final step, the DMHQ instructs the chosen Data Miners to start mining the database at their earliest convenience. If one or more Data Miners are swamped by work, it is trivial to create a copy of them, which can help lessen the work load. Finally, whenever a Data Miner completes learning some database relation, it sends its newly created Expert Network to the DMHQ for integration in the Central Expert Network Repository. The stored Expert Networks can of course be utilized by users in search of answers to specific expert questions. The most interesting aspect of this thought experiment is the fact that thousands of autonomous agents will need to cooperate within an artificial landscape - the information highway - without any human intervention.
Fuzzy Quantifiers and Quantifying Operators in a Connectionist Expert System Development Tool,International Joint Conference on Neural Networks, Singapore, November, pp. 134-139, Nov. 1991.(With S. Romaniuk)
Inductive Learning For Expert Systems In Manufacturing, 25th Hawaii International Conference onSystems Sciences, Jan. 1992. (with R. A. Perez, S. Romaniuk and J. T. Lilkendey)
A Connectionist Architecture for Production Rules with Variables, Iizuka’92, 2nd International Conference on Fuzzy Logic and Neural Networks, July. (with S.G. Romaniuk and K. Sanou).
Evaluation of Machine Learning Tools Using Real Manufacturing Data, International Journal of Expert Systems: Research and Applications, (1992) V. 5, No. 4, pp. 299-318, (With R. Perez, S.Romaniuk and J.T. Lilkendey).
An Encoding of Production Rules in a Connectionist Network, Journal of Intelligent and Fuzzy Systems, 4 (1), pp. 1-18, Feb. 1996, (with K. Sanou, S. Romaniuk)
A Hybrid Symbolic, Connectionist Production System, Tools for Artificial Intelligence, 1992, McLean,Va. (With K. Sanou, and S.G. Romaniuk).
A Production System based on a Connectionist Architecture, International Joint Conference on Neural Networks, Nov. 1992, Beijing, China. (With K. Sanou, and S.G. Romaniuk).
A Connectionist Production System with Approximate Matching Function, FUZZ-IEEE, (1993), pp.415-421, (With K. Sanou and S.G. Romaniuk).
Learning Fuzzy Rules an Instance Based Approach, 5th International Fuzzy Systems Association World Congress, 1993, pp. 171-174, Seoul, Korea. (With S. Romaniuk).
S. Romaniuk and L. Hall
Injecting Symbol Processing Into a Connectionist Model
Book
Neural and intelligent systems integration: fifth and sixth generation integrated reasoning information systems
Branco Soucek and the IRIS Group
New York : Wiley, c1991
Lawrence O. Hall Steve G. Romaniuk
Performance Issues of a Hybrid Symbolic, Connectionist Algorithm
Chapter 6
In Hybrid architectures for intelligent systems
Eds. Abraham Kandel, Gideon Langholz
CRC Press, 1992
Encoding of Production Rules in a Connectionist Network
Article
Journal of Intelligent and Fuzzy Systems
Jan 1996
Katsuaki Sanou
Steve G. Romaniuk
Lawrence O. Hall
The task of implementing a simple rule-based production system in a connectionist architecture is discussed. To implement a connectionist-based production system, it is essential to be capable of representing explicit rules by means of simple neuronlike computing units. The architecture described here uses a local representation based upon the inst...
A Lossless Image Compression Algorithm Using Variable Block Size Segmentation
IEEE Transactions on Image Processing
Article
Nov 1995
Nagarajan Ranganathan
Steve G. Romaniuk
Kamesh Namuduri
The redundancy in digital image representation can be classified into two categories: local and global. In this paper, we present an analysis of two image characteristics that give rise to local and global redundancy in image representation. Based on this study, we propose a lossless image compression scheme that exploits redundancy both at local a...
Application of Learning to Learn to Real-World Pattern Recognition
Artificial Neural Nets and Genetic Algorithms (pp.198-201)
Chapter
Jan 1995
Steve G. Romaniuk
Pattern recognition has for decades played an important role in the development of intelligent systems and numerous algorithms have been proposed to deal with this important task. Though much interest and research has been invested into the construction of advanced pattern recognition systems, almost all of these algorithms are incapable of meta-le...
Applying constructed neural networks to lossless image compression
IEEE Xplore
Conference Paper
Dec 1994
S.G. Romaniuk
The ability to employ neural networks to the task of image compression has been pointed out in research. The pre-dominant approach to image compression is centered around the backpropagation algorithm used to train on overlapping frames of the original picture. Several deficiencies can be identified with this approach. First, no potential time boun...
A lossless image compression algorithm using variable block size segmentation
IEEE Transactions on Image Processing
Conference Paper
Nov 1994
N. Ranganathan
S.G. Romaniuk
Kamesh Namuduri
In this paper, we present an analysis of two image characteristics which give rise to local and global redundancy in image representation. Based on this study, we propose a lossless image compression scheme which exploits both types of redundancy. The algorithm segments the image into variable size blocks and encodes them depending on characteristi...
Theoretical results for applying neural networks to lossless image compression
Network Computation in Neural Systems
Article
Nov 1994
Steve G Romaniuk
The ability to employ neural networks to the task of image compression has been pointed out in recent research. The pre-dominant approach to image compression is centered around the backpropagation algorithm to train on overlapping frames of the original picture. Several deeciencies can be identiied with this approach: First, no potential time boun...
Applying crossover operators to automatic neural network construction
IEEE Xplore
Conference Paper
Jul 1994
S.G. Romaniuk
The ability to automatically construct neural networks is of importance, since it supports reduction in development time and can lead to simpler designs than traditionally handcrafted networks. Automation is further required to take the step towards a more autonomous learning system. In this paper, we report further results involving the automatic...
Towards minimal network architectures with evolutionary growth networks
Conference: Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Volume: 3
IEEE Xplore
Conference Paper
Jan 1994
S.G. Romaniuk
This paper points out how simple learning rules such as perceptron and delta can be re-introduced as local learning techniques to yield an effective automatic network construction algorithm. This feat is accomplished by choosing the right training set during network construction. The choice of partitions can have profound affects on the quality of...
Theoretical results for applying neural networks to lossless image compression
Network Computation in Neural Systems
Article
Jan 1994
Steve G Romaniuk
The fact that neural networks may be employed in the task of image compression has been pointed out in recent research. The predominant approach to image compression centres around the back-propagation algorithm for training on overlapping frames of the original picture. Several deficiencies of this approach can be identified. First, no potential t...
A general class of neural networks
Conference: Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Volume: 3
IEEE Xplore
Authors:S.G. Romaniuk
Conference Paper
Jan 1994
S.G. Romaniuk
Striving to derive minimal network architectures for neural networks has been at the center of attention for several years now. To this date numerous algorithms have been proposed to automatically construct networks. Unfortunately, these algorithms lack a fundamental theoretical analysis of their capabilities and only empirical evaluations on a few...
Learning to Learn: Automatic Adaptation of Learning Bias
Conference: Proceedings of the 12th National Conference on Artificial Intelligence, Seattle, WA, USA, July 31 - August 4, 1994, Volume 2.
Conference Paper
Jan 1994
Steve G. Romaniuk
Traditionally, large areas of research in machine learning have concentrated on pattern recogni- tion and its application to many diversified prob- lems both within the realm of AI as well as out- side of it. Over several decades of intensified re- search, an array of learning methodologies have been proposed, accompanied by attempts to eval- uate...
Fuzzy Rule Extraction for Determining Creditworthiness of Credit Applicants
Article
Dec 1993
Steve G. Romaniuk
The main objective of this research paper is to provide an empirical analysis of the hybrid symbolic/connectionist expert system development tool SC-net to act as a viable system for acquiring expert system knowledge by means of learning. The task to be studied is the prediction of creditworthiness for credit seeking applicants. The creditworthines...
Representing Complex Fuzzy Membership Functions in a Connectionist Network
Article
Dec 1993
Steve G. Romaniuk
The problem of deriving membership functions as a means for describing linguistic variables (for some control process) and the choice of fuzzy inference operators and connectives is at the heart of developing fuzzy control systems. Over the years connectionist systems have obtained prominence as a means to solve complicated learning tasks. More rec...
Towards minimal network architectures with evolutionary growth perceptrons
Conference: Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Volume: 1
Authors:Conference Paper
Nov 1993
S.G. Romaniuk
The purpose of this paper is twofold: First, it will show how the perceptron learning rule can be re-introduced as a local learning technique within the general framework of automatic network construction. Second, it will be pointed out how choosing the right training set during network construction can have profound affects on the quality of the c...
Divide and Conquer Neural Networks
November 1993
Neural networks: the official journal of the International Neural Network Society
6(8):1105-1116
Article
Nov 1993
Steve G. Romaniuk
Lawrence O. Hall
Determining an effective architecture far a multi-layer feedforward back propagation neural network can be a time-consuming effort. We describe an algorithm called Divide and Conquer Neural Networks (DCN), which creates a feedforward neural network architecture during training, based upon the training examples. The first cell introduced on any laye...
SC-net: A hybrid connectionist, symbolic system
Article
July 1993
Information Sciences
71(3):223-268
Steve G. Romaniuk
Lawrence O. Hall
This paper describes the SC-net system that has been developed to provide expert systems capability augmented with learning in a hybrid connectionist/symbolic approach. A distributed connectionist representation of cells connected by links is used to represent symbolic knowledge. Rules may be directly encoded in the connectionist network or learned...
Trans-Dimensional Learning.
June 1993
International Journal of Neural Systems
4(02):171-185
Article
Jun 1993
Steve G. Romaniuk
The main objective of this research paper is to introduce a novel approach to pattern classification across multiple dimensions. Trans-dimensional learning is concerned with automatically determining network architectures that can generalize not only for fixed-dimensional problems but take a profound step by exemplifying how learning an unrestricte...
A connectionist production system with approximate matching function
Conference: Fuzzy Systems, 1993., Second IEEE International Conference on
Conference Paper
Feb 1993
K. Sanou
S.G. Romaniuk
L.O. Hall
A hybrid symbolic/connectionist implementation of a production system with an approximate matching function is presented. The authors demonstrate the conversion of sequential production rules into a parallel firing connectionist network consisting of simple cells and providing an approximate matching function. The hybrid symbolic/connectionist prod...
Learning fuzzy control with hybrid symbolic, connectionist networks
Conference: Fuzzy Systems, 1993., Second IEEE International Conference on
Conference Paper
Feb 1993
S.G. Romaniuk
The author shows, by means of a real-world example of controlling a steam engine, how hybrid symbolic/connectionist learning systems can be employed for automating the design of fuzzy controllers. Deriving the necessary linguistic variables and accompanying membership functions from raw data by use of machine learning is addressed. It is stressed t...
Prove of Convergence of Extended Divide & Conquer Networks
ICANN ’93 (pp.494-497)
Chapter
Jan 1993
Steve G. Romaniuk
The task of determining an effective architecture for multi-layer feed forward backpropagation like neural networks can be a time consuming effort. Over the past couple years several algorithms were proposed for dynamically constructing network architectures. Some of these algorithms have been shown to converge for binary data. Unfortunately, the r...
A hybrid/symbolic connectionist production system
Conference: Tools with Artificial Intelligence, 1992. TAI '92, Proceedings., Fourth International
Conference Paper
Dec 1992
K. Sanou
S.G. Romaniuk
L.O. Hall
The task of implementing a simple rule-based production system in terms of a hybrid symbolic/connectionist architecture is discussed. The aim is to achieve parallel execution of rules. The architecture uses a local representation and builds upon prior work, which resulted in the symbolic/connectionist expert system development tool SC-net. The Hybr...
Dynamic neural networks with the use of divide and conquer
Conference
Conference: Neural Networks, 1992. IJCNN., International Joint Conference on
Volume: 1
Jul 1992
S.G. Romaniuk
L.O. Hall
An algorithm called divide and conquer neural networks that creates a feedforward neural network during training based upon the training examples is described. In addition to learning the weights for connections, it learns an architecture that enables it to learn the examples. Training is done on the inputs to one cell at a time with learned weight...
Decision making on creditworthiness, using a fuzzy connectionist model
992
Fuzzy Sets and Systems
48(1):15-22
Article
May 1992S
Steve G. Romaniuk
Lawrence O. Hall
In this paper we present an unpolished expert system development tool, based on a connectionist architecture for knowledge representation. Our work is centered around a connectionist expert system, which can be expanded, and updated through learning of sample domain specific cases [1,4]. A cell recruitment learning algorithm [2] capable of forgetti...
Learning fuzzy information in a hybrid connectionist, symbolic model
Conference: Fuzzy Systems, 1992., IEEE International Conference on
Conference Paper
Apr 1992
S.G. Romaniuk
L.O. Hall
An implementation of fuzzy variables using pi-shaped membership functions is shown in a hybrid symbolic connectionist expert system tool that uses fuzzy logic to implement reasoning with uncertainty and imprecision and that can learn from imprecise data. A method of dynamically modifying the arms, or fuzzy part of the membership functions, during l...
Inductive learning for expert systems in manufacturing
Conference: System Sciences, 1992. Proceedings of the Twenty-Fifth Hawaii International Conference on
Volume: iii
Conference Paper
Feb 1992
R.A. Perez
L.O. Hall
S. Romaniuk
Jim Lilkendey
The authors evaluate several inductive learning techniques using semiconductor wafer failure data gathered during its manufacturing process and where there is currently an expert system in use with rules derived from experts. The learning systems include symbolic (ID3, GID3, CN2), connectionist (Quickprop) and a hybrid model (SC-net). A year's wort...
Performance issues of a hybrid symbolic, connectionist algorithm
Conference Paper, Conference: Hybrid architectures for intelligent systems
Jan 1992
Lawrence O. Hall
Steve G. Romaniuk
Fuzzy quantifiers and quantifying operators in a connectionist expert system development tool
Conference: Neural Networks, 1991. 1991 IEEE International Joint Conference on
Conference Paper
Dec 1991
S.G. Romaniuk
L.O. Hall
The authors give information pertaining to the implementation of fuzzy quantifiers and quantifying operators within a connectionist network model. The operators described can be extended to arbitrary input size, by retaining similar overall behavior. Examples are given to show the responses one would obtain when modifying the belief in the inputs...
A Hybrid Connectionist, Symbolic Learning System.
Conference: Proceedings of the 8th National Conference on Artificial Intelligence. Boston, Massachusetts, July 29 - August 3, 1990, 2 Volumes.
Conference Paper
Jan 1990
Lawrence O. Hall
Steve G. Romaniuk
This paper describes the learning part of a system which has been developed to provide expert systems capability augmented with learning. The learning scheme is a hybrid connectionist, symbolic one. A network representation is used. Learning may be done incrementally and requires only one pass through the data set to be learned. Attribute, value pa...
Fuzzy connectionist expert systems
Conference: Neural Networks, 1989. IJCNN., International Joint Conference on
Conference Paper
Feb 1989
S.G. Romaniuk
L.O. Hall
Summary form only given, as follows. A fuzzy connectionist expert system with learning capabilities is described. The system uses a recruitment-of-cells learning algorithm for knowledge acquisition, and also allows the translation of rules into an equivalent connectionist network. The system contains conventional expert system features such as expl...
Multi-pass instance based learning
DISC Article NUS Technical Report
Steve G Romaniuk
This paper introduces a new modiied approach to the instance based learning the-ory. Instance based learning is augmented by neighborhood spheres and multi-pass training to improve both on generalization capabilities and storage requirements. Two models for creating neighborhood spheres are investigated and put in perspec-tive with the IBL instance...
Learning fuzzy control rules from examples
Article
Steve G. Romaniuk
Lawrence O. Hall
Towards A Practical Estimate Of Training Sample Size
Article
Steve G Romaniuk
The purpose of this paper is to introduce a simple learning model that allows one to draw conclusions about the number of distinct training examples required to learn some boolean function with at least accuracy and probability across a general class of learning algorithms. The motivation for this work stems from the inability of l
Steve G. Romaniuk
Using Intelligent Agents to Identify Missing and Exploited Children.
IEEE Intell. Syst. 15(2): 27-30 (2000)
Katsuaki Sanou, Steve G. Romaniuk, Lawrence O. Hall
Encoding of Production Rules in a Connectionist Network.
J. Intell. Fuzzy Syst. 4(1): 1-18 (1996)
Steve G. Romaniuk
Theoretical results for a class of neural networks.
Int. J. Neural Syst. 6(4): 463-472 (1995)
Nagarajan Ranganathan, Steve G. Romaniuk, Kameswara Rao Namuduri
A lossless image compression algorithm using variable block size segmentation.
IEEE Trans. Image Process. 4(10): 1396-1406 (1995)
Steve G. Romaniuk
Application of Learning to Learn to Real-World Pattern Recognition.
ICANNGA 1995: 198-201
Steve G. Romaniuk
Evolutionary Grown Semi-Weighted Neural Networks.
ICGA 1995: 444-451
Steve G. Romaniuk
Learning to Learn: Automatic Adaptation of Learning Bias.
AAAI 1994: 871-876
Steve G. Romaniuk
Applying Co-Evolution to the Construction of Neural Networks.
ECAI 1994: 226-230
Steve G. Romaniuk
Applying Crossover Operators to Automatic Neural Network Construction.
International Conference on Evolutionary Computation 1994: 750-752
Steve G. Romaniuk
Applying Constructed Neural Networks to Lossless Image Compression.
ICIP (3) 1994: 948-952
N. Ranganathan, Steve G. Romaniuk, Kameswara Rao Namuduri
A lossless image compression algorithm using variable block size segmentation.
ICPR (3) 1994: 40-44
Steve G. Romaniuk
Efficient Storage of Instances: The Multi-Pass Approach.
IEA/AIE 1994: 237-244
and7th Int. Conf. Industrial & En. edited by Frank D. Anger, Rita V. Rodriguez, Moonis Ali
Steve G. Romaniuk
Trans-Dimensional Learning.
Int. J. Neural Syst. 4(2): 171-185 (1993)
Steve G. Romaniuk, Lawrence O. Hall
SC-net: A hybrid connectionist, symbolic system.
Inf. Sci. 71(3): 223-268 (1993)
Steve G. Romaniuk, Lawrence O. Hall
Divide and Conquer Neural Networks.
Neural Networks 6(8): 1105-1116 (1993)
Steve G. Romaniuk
Evolutionary Growth Perceptrons.
ICGA 1993: 334-341
Katsuaki Sanou, Steve G. Romaniuk, Lawrence O. Hall
A Hybrid/Symbolic Connectionist Production System.
ICTAI 1992: 44-53
Lawrence O. Hall, Steve G. Romaniuk
A Hybrid Connectionist, Symbolic Learning System.
AAAI 1990: 783-788
Genetic Algorithms for Pattern Recognition (CRC Press Revivals)
Chapter
by Sankar K. Pal , Paul P. Wang, et al.
20 September 2017
Pruning Divide & Conquer Networks
Steve G. Romaniuk
National University of Singapore, Department of Information Systems and Computer Science,
1993 - Neural networks (Computer science) - 25 pages
Abstract:
"Determining an effective architecture for multi-layer feed forward back propagation neural networks can be a time-consuming effort. In general it requires human intervention in determining the number of layers, number of hidden cells, the learning rule, and the learning parameters. Over the past few years several approaches to dynamically configure neural networks have been proposed, which remove most of the responsibility for chosing the correct network configuration from the user. As important as finding a viable network architecture for some given learning problem, is the need to obtain a minimal configuration. The total time required to emulate or simulate neural networks is largely dependent on the number of connections present in a network. Therefore, it is essential to provide pruning methods to reduce network complexity
Proceedings/Papers from the 1999 AAAI Spring Symposium/Intelligent Agents in Cyberspace
Evolutionary Agent Societies Applied to Knowledge Discovery and Predictive Data Mining
Authors
Steve G. Romaniuk
Abstract:
This paper reports ongoing efforts in developing a Knowledge Discovery and Predictive Data Mining (KD & PDM) system at ANSER as part of an National Institute of Justice (NIJ) grant to locate missing and exploited children. The here in described system represents a first pilot project for exploring the general use of Evolutionary Agent Societies (EAS). EAS seeks to expand current efforts in IT software development, deployment, and maintenance by addressing such important factors as reliability, scalability, and adaptability of intelligent software.
Towards a Practical Estimate of Training Sample Size
Steve Romaniuk
URI: https://dl.comp.nus.edu.sg/xmlui/handle/1900.100/1441
Date: 1993-11-01
Abstract
The purpose of this paper is to introduce a simple learning model that allows one to draw conclusions about the number of distinct training examples required to learn some boolean function with at least accuracy ${\alpha}$ and probability ${\delta}$ across a general class of learning algorithms. The motivation for this work stems from the inability of learning theoretical models to suggest reasonable sample bounds. Reducing sample size is essential in the wake of expected costs for labeling patterns by a teacher (e.g. human expert). The derived results are then extended from learning functions to learning concepts to make the analysis more realistic. The importance of domain specific knowledge in learning concepts is discussed and incorporated into the model in form of identifying impossible training patterns. Several possible sources for these impossibilities are pointed out. The paper is concluded with 2 representative examples.
S. G. Romaniuk and L. O. Hall, "Parallel Connectionist Expert Systems", IASTED Applications and Theory Track, pp. 241-244, 1989-June.Published in: International 1989 Joint Conference on Neural Networks
S. G. Romaniuk and L. O. Hall, "FUZZNET: Towards a Fuzzy Connectionist Expert System Development Tool", IJCNN-90, 1989-January.
S. G. Romaniuk and L. 0. Hall, "The Use of Fuzzy Variables in a Hybrid Connectionist Expert System", NAFIPS'90, 1990-June.
S. G. Romaniuk and L. O. Hall, "Injecting Symbol Processing into a Connectionist Model" in Neural and Intelligent Systems Integration, N.Y.:John Wiley, 1991.
K. Sanou, S. G. Romaniuk and L. O. Hall, "A Connectionist Implementation of a Production System on a Hypercube Multiprocessor", 93 Korea/Japan Joint Conference on Expert Systems Seoul, 1993-February.
Romaniuk, S.G. (1994) Efficient Storage of Instances: The MultiPass Approach, in Procee. 7th Inter. Conf. On Industrial and Engineering Applications of Artificial Intelligence and Expert Systems, Austin, Texas, June, 1994.
Romaniuk, S.G. (1995) TDL Agents applied to Autonomous Data Mining, White Paper: Universal Problem Solvers, Inc. (Also see Through the Looking Glass... in TDL's Windows Help System, http://pages.prodigy.com/upso).
Evolutionary Agent Societies, Steve G. Romaniuk, ANSER, Fairmont, WV 26554, USA, Email: romanius@anser.org, https://web-archive.southampton.ac.%75%6B/cogprints.org/461/1/EAS.html
Multi-Pass Instance Based Learning by Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg Keywords: Instance-based Learning, Complexity Analysis, Empirical Evaluation, Neighborhood ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1992/TR20-92.ps.gz, 19931124
Fuzzy Rule Extraction for Determining Creditworthiness of Credit Applicants by Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Singapore, 0511
Fuzzy Logic, Rule Extraction, Machine Learning, Hybrid Symbolic/Connectionist
ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TRE6-93.ps.gz, 19931124
Learning Fuzzy Control with Hybrid Symbolic, Connectionist Networks Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Singapore, 0511 June 25, 1993 ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1992/TR12-92.ps.gz, 19931124
Representing Complex Fuzzy Membership Functions in a Connectionist Network by Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore, 0511ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TR20-93.ps.gz, 19931124
Pruning Divide & Conquer Networks Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg Abstract Determining an effective architecture for multi-layer feed forward back propagation neural ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TRF4-93.ps.gz, 19931124
Prove of Convergence of Extended Divide & Conquer Networks Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg Abstract The task of determining an effective architecture for multi-layerftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TRC1-93.ps.gz, 19931124
A Feature-Based Heuristic Algorithm for Lossless Image Compression S. G. Romaniuk K. R. Namuduri N. Ranganathan DISCS Dept. Computer Science Center for National University of Singapore and Engineering Microelectronics Research 10 Kent Ridge Univ. of South Florida Univ. of South Florida Singapore 0511ftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TR21-93.ps.gz, 19931216
Towards A Practical Estimate Of Training Sample Size by Steve G. Romaniuk Department of Information Systems & Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg Abstract The purpose of this paper is to introduce a simple learning model thatftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1993/TR11-93.ps.gz, 19931216
Learning To Learn : Automatic Adaptation of Learning Bias Steve G. Romaniuk Department of Information Systems and Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg Abstract Traditionally, large areas of research in machine learning haveftp://ftp.nus.sg/pub/NUS/ISCS/techreports/1994/TRA1-94.ps.gz, 199403
Theoretical Results for Applying Neural Networks to Lossless Image Compression Steve G. Romaniuk Department of Information Systems & Computer Science National University of Singapore 10 Kent Ridge Crescent Singapore 0511 e-mail: stever@iscs.nus.sg
Romaniuk, S.G., “FUZZNET/SC-net Users Manual”, V. 1.0, Department of Computer Science and Engineering, University of South Florida, Tampa, 1989.
Perez, R.A., Hall, L.O., Romaniuk, S.G, and Lilkendey, J., “Induced rules vs. expert derived rules in a manufacturing environment”, Tech. Rept. ISL-91-01, Dept. Of Computer Science and Engineering, University of South Florida, Tampa, FL, 1991.
Feel Free To Drop Me A Line.
MobileTimesToday / Revelation Engine, New York, NY | San Jose, CA | Clearwater, FL