Subject: KDD Nuggets 95:16 KDD Nuggets 95:16, e-mailed 95-07-07 Contents: * Cen Li, new KDD related page http://www.vuse.vanderbilt.edu/~biswas/ResearchPages/kdd.html * W. Sarle, How are neural networks related to statistical methods? * GPS, useful pointers about Montreal, http://www.showboat.com/showboat/montreal.htm * R. Kohavi, MLC++ Utilities 1.2, http://robotics.stanford.edu:/users/ronnyk/mlc.html. * M. Pazzani, Postdoc at UC Irvine * R. Tucker, Renascence Partners -- http://www.rpl.com/ * S. Dixon, Data Mining related jobs at SmithKline Beecham The KDD Nuggets is a moderated mailing list for news and information relevant to Knowledge Discovery in Databases (KDD), also known as Data Mining, Knowledge Extraction, etc. Relevant items include tool announcements and reviews, summaries of publications, information requests, interesting ideas, clever opinions, etc. Please include a descriptive subject line in your submission. Nuggets frequency is approximately bi-weekly. Back issues of Nuggets, a catalog of S*i*ftware (data mining tools), references, FAQ, and other KDD-related information are available at Knowledge Discovery Mine, URL http://info.gte.com/~kdd/ or by anonymous ftp to ftp.gte.com, cd /pub/kdd, get README E-mail add/delete requests to kdd-request@gte.com E-mail contributions to kdd@gte.com -- Gregory Piatetsky-Shapiro (moderator) ********************* Official disclaimer *********************************** * All opinions expressed herein are those of the writers (or the moderator) * * and not necessarily of their respective employers (or GTE Laboratories) * ***************************************************************************** ~~~~~~~~~~~~ Quotable Quote ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ "The Universe is full of magical things, patiently waiting for our wits to grow sharper." -Eden Phillipotts (thanks to Susan Tafolla) >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Return-Path: Date: Tue, 27 Jun 95 11:26:50 CDT From: cenli@vuse.vanderbilt.edu (Cen Li) To: kdd%eureka@gte.com Subject: new WWW page to be included in the kdd home page Dr. Shapiro: I am a graduate student working on KDD at Vanderbilt University under Prof. Gautam Biswas. We have been putting together a web page in this area. We would like it to be included in your KDD page. The link is: "http://www.vuse.vanderbilt.edu/~biswas/ResearchPages/kdd.html" Thank you very much. Sincerely, Cen Li. >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Here is a very good excerpt from Warren Sarle that I picked up on CLASS-L list. GPS. Return-Path: Date: Tue, 27 Jun 1995 14:28:57 -0400 Reply-To: "Classification, clustering, and phylogeny estimation" Sender: "Classification, clustering, and phylogeny estimation" From: Warren Sarle Subject: Re: Neural Network Modeling Q's X-To: CLASS-L@ccvm.sunysb.edu X-Cc: dihost!fjurden@dirsrch.attmail.com To: Multiple recipients of list CLASS-L In-Reply-To: <199506271322.AA04755@lamb.sas.com> from "Art Kendall" at Jun 27, 95 09:16:44 am > does anybody have any info/citations on neural network modeling and its > relation to other stat techniques such as regression, structural equation > modeling, etc.? > > thanks. > Frank H. Jurden, Ph.D. > Decision Insight, Inc. > 2600 Grand > Kansas City, MO USA > dihost!fjurden@dirsrch.attmail.com Below is an excerpt that I wrote for the comp.ai.neural-nets FAQ, a discussion of neural nets and structural equation models, a compilation of neural net and statistical jargon, and directions for obtaining more information via ftp. ______________________________________________________________________ Q: How are neural networks related to statistical methods? A: There is considerable overlap between the fields of neural networks and statistics. Statistics is concerned with data analysis. In neural network terminology, statistical inference means learning to generalize from noisy data. Some neural networks are not concerned with data analysis (e.g., those intended to model biological systems) and therefore have little to do with statistics. Some neural networks do not learn (e.g., Hopfield nets) and therefore have little to do with statistics. Some neural networks can learn successfully only from noise-free data (e.g., ART or the perceptron rule) and therefore would not be considered statistical methods. But most neural networks that can learn to generalize effectively from noisy data are similar or identical to statistical methods. For example: * Feedforward nets with no hidden layer (including functional-link neural nets and higher-order neural nets) are basically generalized linear models. * Feedforward nets with one hidden layer are closely related to projection pursuit regression. * Probabilistic neural nets are identical to kernel discriminant analysis. * General regression neural nets are identical to Nadaraya-Watson kernel regression. * Kohonen nets for adaptive vector quantization are very similar to k-means cluster analysis. * Hebbian learning is closely related to principal component analysis. Some neural network areas that appear to have no close relatives in the existing statistical literature are: * Kohonen's self-organizing maps. * Reinforcement learning (although this is treated in the operations research literature as Markov decision processes). * Stopped training (the purpose and effect of stopped training are similar to shrinkage estimation, but the method is quite different). Feedforward nets are a subset of the class of nonlinear regression and discrimination models. Statisticians have studied the properties of this general class but had not considered the specific case of feedforward neural nets before such networks were popularized in the neural network field. Still, many results from the statistical theory of nonlinear models apply directly to feedforward nets, and the methods that are commonly used for fitting nonlinear models, such as various Levenberg-Marquardt and conjugate gradient algorithms, can be used to train feedforward nets. While neural nets are often defined in terms of their algorithms or implementations, statistical methods are usually defined in terms of their results. The arithmetic mean, for example, can be computed by a (very simple) backprop net, by applying the usual formula SUM(x_i)/n, or by various other methods. What you get is still an arithmetic mean regardless of how you compute it. So a statistician would consider standard backprop, Quickprop, and Levenberg-Marquardt as different algorithms for implementing the same statistical model such as a feedforward net. On the other hand, different training criteria, such as least squares and cross entropy, are viewed by statisticians as fundamentally different estimation methods with different statistical properties. It is sometimes claimed that neural networks, unlike statistical models, require no distributional assumptions. In fact, neural networks involve exactly the same sort of distributional assumptions as statistical models, but statisticians study the consequences and importance of these assumptions while most neural networkers ignore them. For example, least-squares training methods are widely used by statisticians and neural networkers. Statisticians realize that least-squares training involves implicit distributional assumptions in that least-squares estimates have certain optimality properties for noise that is normally distributed with equal variance for all training cases and that is independent between different cases. These optimality properties are consequences of the fact that least-squares estimation is maximum likelihood under those conditions. Similarly, cross-entropy is maximum likelihood for noise with a Bernoulli distribution. If you study the distributional assumptions, then you can recognize and deal with violations of the assumptions. For example, if you have normally distributed noise but some training cases have greater noise variance than others, then you may be able to use weighted least squares instead of ordinary least squares to obtain more efficient estimates. References: Balakrishnan, P.V., Cooper, M.C., Jacob, V.S., and Lewis, P.A. (1994) "A study of the classification capabilities of neural networks using unsupervised learning: A comparison with k-means clustering", Psychometrika, 59, 509-525. Chatfield, C. (1993), "Neural networks: Forecasting breakthrough or passing fad", International Journal of Forecasting, 9, 1-3. Cheng, B. and Titterington, D.M. (1994), "Neural Networks: A Review from a Statistical Perspective", Statistical Science, 9, 2-54. Geman, S., Bienenstock, E. and Doursat, R. (1992), "Neural Networks and the Bias/Variance Dilemma", Neural Computation, 4, 1-58. Kuan, C.-M. and White, H. (1994), "Artificial Neural Networks: An Econometric Perspective", Econometric Reviews, 13, 1-91. Kushner, H. & Clark, D. (1978), _Stochastic Approximation Methods for Constrained and Unconstrained Systems_, Springer-Verlag. Michie, D., Spiegelhalter, D.J. and Taylor, C.C. (1994), _Machine Learning, Neural and Statistical Classification_, Ellis Horwood. Ripley, B.D. (1993), "Statistical Aspects of Neural Networks", in O.E. Barndorff-Nielsen, J.L. Jensen and W.S. Kendall, eds., _Networks and Chaos: Statistical and Probabilistic Aspects_, Chapman & Hall. ISBN 0 412 46530 2. Ripley, B.D. (1994), "Neural Networks and Related Methods for Classification," Journal of the Royal Statistical Society, Series B, 56, 409-456. Sarle, W.S. (1994), "Neural Networks and Statistical Models," Proceedings of the Nineteenth Annual SAS Users Group International Conference, Cary, NC: SAS Institute, pp 1538-1550. White, H. (1989), "Learning in Artificial Neural Networks: A Statistical Perspective," Neural Computation, 1, 425-464. White, H. (1989), "Some Asymptotic Results for Learning in Single Hidden Layer Feedforward Network Models", J. of the American Statistical Assoc., 84, 1008-1013. White, H. (1992), _Artificial Neural Networks: Approximation and Learning Theory_, Blackwell. ______________________________________________________________________ > Is anyone familiar with both LISREL and neural networks who can explain (or > cite a publication which explains) how neural network analysis differs from > using LISREL to work with structural equations with latent variables? Linear regression is a special case of a structural equation model (LISREL) with no latent variables. Linear regression is also a special case of a feedforward neural net with no hidden layer and a linear activation function. Aside from that, there is no connection between structural equation models and neural networks. Latent variables are things that are assumed to exist in the real world but can't be measured directly. Hidden units are computational conveniences--they are not latent variables. Note that in a network diagram for a neural network, the arrows go from the inputs to the hidden units. In a path diagram for a latent variable model, the arrows go in the opposite direction from the latent variables to the manifest variables. You can set up a neural net with linear hidden units that is equivalent to certain principal component or maximum redundancy models. However, principal components and maximum redundancy are not latent variable models. If you use principal components to estimate common factors (which are latent variables), you will get wrong answers. Similarly, if you use a neural net to estimate a superficially similar latent variable model, you will get wrong answers. This is because standard estimation methods such as ordinary least squares (OLS) apply to various linear and nonlinear models such as principal components and feedforward neural nets, but OLS does _not_ produce consistent estimates for latent variable models. ______________________________________________________________________ Neural Network and Statistical Jargon ===================================== Warren S. Sarle saswss@unx.sas.com May 12, 1995 The neural network (NN) and statistical literatures contain many of the same concepts but usually with different terminology. Sometimes the same term or acronym is used in both literatures but with different meanings. Only in very rare cases is the same term used with the same meaning, although some cross-fertilization is beginning to happen. Below is a list of such corresponding terms or definitions. Particularly loose correspondences are marked by a ~ between the two columns. A < indicates that the term on the left is roughly a subset of the term on the right, and a > indicates the reverse. Terminology in both fields is often vague, so precise equivalences are not always possible. The list starts with some basic definitions. There is disagreement in the NN literature on how to count layers. Some people count inputs as a layer and some don't. I specify the number of hidden layers instead. This is awkward but unambiguous. Definition Statistical Jargon ========== ================== generalizing from noisy data Statistical inference and assessment of the accuracy thereof the set of all cases one Population wants to be able to generalize to a function of the values in Parameter a population, such as the mean or a globally optimal synaptic weight a function of the values in Statistic a sample, such as the mean or a learned synaptic weight Neural Network Jargon Definition ===================== ========== Neuron, neurode, unit, a simple linear or nonlinear computing node, processing element element that accepts one or more inputs, computes a function thereof, and may direct the result to one or more other neurons Neural networks a class of flexible nonlinear regression and discriminant models, data reduction models, and nonlinear dynamical systems consisting of an often large number of neurons interconnected in often complex ways and often organized into layers Neural Network Jargon Statistical Jargon ===================== ================== Statistical methods Linear regression and discriminant analysis, simulated annealing, random search Architecture Model Training, Learning, Estimation, Model fitting, Optimization Adaptation Classification Discriminant analysis Mapping, Function Regression approximation Supervised learning Regression, Discriminant analysis Unsupervised learning, Principal components, Cluster analysis, Self-organization Data reduction Competitive learning Cluster analysis Hebbian learning, Principal components Cottrell/Munro/Zipser technique Training set Sample, Construction sample Test set, Validation set Hold-out sample Pattern, Vector, Case Observation, Case Reflectance pattern an observation normalized to sum to 1 Binary(0/1), Binary, Dichotomous Bivalent or Bipolar(-1/1) Input Independent variables, Predictors, Regressors, Explanatory variables, Carriers Output Predicted values Training values Dependent variables, Responses, Target values Observed values Training pair Observation containing both inputs and target values Shift register, Lagged variable (Tapped) (time) delay (line), Input window Errors Residuals Noise Error term Generalization Interpolation, Extrapolation, Prediction Error bars Confidence interval Prediction Forecasting Adaline Linear two-group discriminant analysis (ADAptive LInear NEuron) (not Fisher's but generic) (No-hidden-layer) perceptron ~ Generalized linear model (GLIM) Activation function, > Inverse link function in GLIM Signal function, Transfer function Softmax Multiple logistic function Squashing function bounded function with infinite domain Semilinear function differentiable nondecreasing function Phi-machine Linear model Linear 1-hidden-layer Maximum redundancy analysis, Principal perceptron components of instrumental variables 1-hidden-layer perceptron ~ Projection pursuit regression Weights, < (Regression) coefficients, Synaptic weights Parameter estimates Bias ~ Intercept the difference between the Bias expected value of a statistic and the corresponding true value (parameter) Shortcuts, Jumpers, ~ Main effects Bypass connections, direct linear feedthrough (direct connections from input to output) Functional links Interaction terms or transformations Second-order network Quadratic regression, Response-surface model Higher-order network Polynomial regression, Linear model with interaction terms Instar, Outstar iterative algorithms of doubtful convergence for approximating an arithmetic mean or centroid Delta rule, adaline rule, iterative algorithm of doubtful Widrow-Hoff rule, convergence for training a linear LMS (Least Mean Squares) rule perceptron by least squares, similar to stochastic approximation training by minimizing the LMS (Least Median of Squares) median of the squared errors Generalized delta rule iterative algorithm of doubtful convergence for training a nonlinear perceptron by least squares, similar to stochastic approximation Backpropagation Computation of derivatives for a multilayer perceptron and various algorithms such as the generalized delta rule based thereon Weight decay, Regularization > Shrinkage estimation, Ridge regression Jitter random noise added to the inputs to shrink the estimates Growing, Pruning, Brain Subset selection, Model selection, damage, Self-structuring, Pre-test estimation Ontogeny Optimal brain surgeon Wald test LMS (Least mean squares) OLS (Ordinary least squares) (see also "LMS rule" above) Relative entropy, Cross Kullback-Leibler divergence entropy Evidence framework Empirical Bayes estimation OLS (Orthogonal least squares) Forward stepwise regression Probabilistic neural network Kernel discriminant analysis General regression neural Kernel regression network Topologically distributed < (Generalized) Additive model encoding Adaptive vector quantization iterative algorithms of doubtful convergence for K-means cluster analysis Adaptive Resonance Theory 2a ~ Hartigan's leader algorithm Learning vector quantization a form of piecewise linear discriminant analysis using a preliminary cluster analysis Counterpropagation Regressogram based on k-means clusters Encoding, Autoassociation Dimensionality reduction (Independent and dependent variables are the same) Heteroassociation Regression, Discriminant analysis (Independent and dependent variables are different) Epoch Iteration Continuous training, Iteratively updating estimates one Incremental training, observation at a time via difference On-line training, equations, as in stochastic approximation Instantaneous training Batch training, Iteratively updating estimates after Off-line training each complete pass over the data as in most nonlinear regression algorithms ______________________________________________________________________ Further information on neural networks is available by anonymous ftp from ftp.sas.com (Internet gateway IP 192.35.83.8) in the directory /pub/sugi19/neural : README This document. neural1.ps Sarle, W.S. (1994), "Neural Networks and Statistical Models," Proceedings of the Nineteenth Annual SAS Users Group International Conference, Cary, NC: SAS Institute, pp 1538-1550. (Postscript file) neural2.ps Sarle, W.S. (1994), "Neural Network Implementation in SAS Software," Proceedings of the Nineteenth Annual SAS Users Group International Conference, Cary, NC: SAS Institute, pp 1551-1573. (Slightly revised version, postscript file) plots.ps Plots from the 2nd paper in high-resolution graphics. (Postscript file) macros.sas Macros from the 2nd paper. example.sas Examples using the macros with the XOR and sine data. example.bls Output from example.sas. example2.sas Examples using the macros with the motorcycle data. example2.bls Output from example2.sas. tnn2.sas The TNN system of macros for feedforward neural nets, alpha release, version 2. tnn2.doc Introductory documentation for TNN. tnn2.ref Reference guide to TNN macros and arguments tnn2ex.sas Examples using TNN with the XOR, iris, and sine data. tnn2ex.bls Output from tnn2ex.sas. tnn2exm.sas Examples using TNN with the motorcycle data. tnn2exm.bls Output from tnn1ex.sas. netiml.sas The NETIML system of IML modules and macros for multilayer perceptrons. netiml.ps Documentation for netiml.sas. netimlex.sas Examples using netiml.sas netimlex.bls Output from netimlex.sas. paint.sas Macro for setting colors and symbols in SAS/INSIGHT. jargon Translations of neural network and statistical jargon. kangaroos Nontechnical explanation of training methods and nonlinear optimization (plain ascii version of material from neural2.ps, plus related posts from the comp.ai.neural-nets newsgroup on Usenet). Please note that postscript files (those with a .ps extension) require a postscript printer or viewer in order for you to read them. -- Warren S. Sarle SAS Institute Inc. The opinions expressed here saswss@unx.sas.com SAS Campus Drive are mine and not necessarily (919) 677-8000 Cary, NC 27513, USA those of SAS Institute. >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Return-Path: Date: Tue, 27 Jun 1995 16:36:26 -0400 From: gps0@eureka (Gregory Piatetsky-Shapiro) Subject: Re: useful pointers about Montreal http://www.showboat.com/showboat/montreal.htm -- montreal home page http://www.cam.org/~delisle/Montreal.html Montreal at night Montreal weather http://www.droit.umontreal.ca/cgi-bin/weather and, of course, http://www-aig.jpl.nasa.gov/kdd95 -- KDD-95 home page >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ From ML-LIST From: Ronny Kohavi Date: Tue, 20 Jun 1995 20:01:47 -0700 Subject: MLC++ Utilities V1.2: Machine Learning Library in C++ MLC++ Utilities 1.2 MLC++ is a Machine Learning library of C++ classes being developed at Stanford. The utilities are compiled versions of programs for Sun and can be used without access to a C++ compiler. The utilities (and sources) are available freely. More information about the library can be obtained at URL http://robotics.stanford.edu:/users/ronnyk/mlc.html. Version 1.2, to be released 30 June 1995, includes the following new additions to 1.1: *. Discretization (binning, Holte, Entropy) [ML-95]. *. Better feature subset selection using the wrapper approach. *. Automatic tuning of C4.5 [paper to appear in ML-95]. *. Combining classifiers (bagging/ensemble). *. Utility to display trees generated by C4.5 using dot. *. Interface to Aha IB series. *. Entropy-based decision graphs [EODG, IJCAI-95]. Ronny Kohavi (ronnyk@CS.Stanford.EDU, http://robotics.stanford.edu/~ronnyk) >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Return-Path: To: kdd@gte.com, comp.ai@super-pan.ICS.UCI.EDU Subject: Postdoc at UC Irvine Date: Thu, 29 Jun 1995 08:53:08 -0700 From: Michael Pazzani Content-Type: text Content-Length: 1166 I would like to hire a PostDoctoral researcher to do research on Machine learning algorithms and applications of machine learning to biomedical problems and intelligent agents. This is a 3 year position starting in the Fall of 1995. To apply, send a resume to pazzani@ics.uci.edu. Application screening will begin immediately upon receipt. Maximum consideration will be given to applications received by July 1, 1995. UC Irvine is located in Southern California, three miles from the Pacific Ocean adjacent to Newport Beach, and approximately forty miles south of Los Angeles. The campus is situated in the heart of a national center of high-technology enterprise. Both the campus and the enterprise area are growing rapidly and offer exciting professional and cultural opportunities. The University of California is an Affirmative Action/Equal Opportunity Employer, committed to excellence through diversity. Michael Pazzani Associate Professor Department of Information and Computer Science University of California Irvine, CA 92717-3425 phone (714) 824-5888 fax (714) 824-4056 e-mail pazzani@ics.uci.edu http://www.ics.uci.edu/dir/faculty/AI/pazzani >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Return-Path: From: Roth Tucker Date: Fri, 30 Jun 1995 11:14:43 -0400 To: kdd%eureka@gte.com (KDD Nuggets Moderator) Subject: Renascence Partners Renascence Partners Limited is a team of business consulting professionals who bring the practical science of decision making to bear on those strategic, tactical and procedural issues that are under-adressed using the simplistic tools of yesterday. Like many decision science firms, we believe that data alone in not enough. Neither is good old seat-of-the-pants intuition. A seamless melding of the two requires a system that can agressively seek data and combine it with an accurate representation of relevent human knowledge to reach an inclusive, complete solution. Our tools include proprietary software engines using data mining, fuzzy sets, neural networks, cognitive process capture and knowledge representaion. Recent examples of work include: Building a model to forecast physician prescribing behavior, which takes seasonal illness and physician attitudes into account Creating a system to select future customer "parters" which relies on both hard data (such as financial figures) and more fuzzy measures (such as "clarity of management vision") Above all, we are dedicated to bringing the most powerful tools to bear on the tough probelms faced by our clients, but doing so in a non-threatening, inclusive process.. Roth Tucker Managing Director, Systems and Analytics Our address is http://www.rpl.com/ Please address inquiries to: Rtucker@rpl.com >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Return-Path: To: kdd@gte.com Subject: Job available in database mining Reply-To: dixon%phmms0.mms@sb.com Date: Fri, 30 Jun 1995 15:55:14 -0400 From: Scott Dixon Content-Type: text Content-Length: 2879 We have several positions open. Note in particular the jobs in database mining and in protein structure (which could include pattern recognition/machine learning approaches in biopolymer sequence databases): OPPORTUNITIES IN COMPUTATIONAL CHEMISTRY/MOLECULAR MODELING SmithKline Beecham, a worldwide leader in pharmaceutical research has a number of openings in our Physical & Structural Chemistry Department. Database Mining The appropriate individual will have a Ph.D. in chemistry or biochemistry or computer science (or related field) with extensive experience in pattern recognition, machine learning or chemometrics. Excellent communication and computer skills, including the ability to develop new algorithms, are required. Experience in chemical or biological database analysis is desirable. Job Code #H00L. Protein Modeling This individual will be responsible for the development and application of methods to assign structure and function to genome sequence information. Qualifications required include a Ph.D. in chemistry, biophysics or bioinformatics with extensive experience in biopolymer sequence analysis and protein modeling. Other requirements include excellent computer skills and the ability to develop computer programs and new algorithms. A knowledge of DNA sequencing methods and molecular biology is desirable. Refer to Job Code #H0118. Scientific Programmer As a member of an established group, the selected candidate will assist in the development of state-of-the-art software. Qualifications include BS/MS in computer science or electrical engineering (with experience or course work in chemistry) or BS/MS in chemistry with demonstrated abilities in scientific computer programming. Excellent programming skills and knowledge of UNIX, C and FORTRAN are needed. Experience with Silicon Graphics workstations and computational chemistry software are desirable. Refer to Job Code #H0117. Combinatorial Chemistry/ Molecular Diversity The selected individual will join with other team members to develop and apply methods for the design of combinatorial chemical libraries and the analysis of diversity in chemical databases. The necessary qualifications include a Ph.D. in chemistry or a related field and extensive experience in computational chemistry. Excellent computer skills and communication skills are also necessary. Experience in pattern recognition or chemical diversity methods is desirable. Job Code #H00M. Located in our state-of-the-art research facility in suburban Philadelphia, SmithKline Beecham offers an excellent compensation/benefits/relocation package. Interested candidates should send resume with salary requirements indicating desired Job Code, to: SmithKline Beecham Pharmaceuticals, Job Code ____ P.O. Box 2645 Bala Cynwyd, PA 19004. We are an Equal Opportunity Employer, M/F/D/V. >~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~