Abstracts of papers in Neural Processing Letters are now available on and by FTP from /pub/neural-nets/NPL on ftp.dice.ucl.ac.be. [esann@dice.ucl.ac.be, connectionists, 3/1/95.]

Warren L. Kovach maintains a WWW page for shareware and public-domain statistical and data analysis software, including his own DOS/Windows programs MVSP and Oriana. . [, c.i.www.announce, 2/13/95. Chris Matheus.]

"NNs are used almost exclusively for prediction rather than explanation or hypothesis testing. The results are often difficult to interpret due to the nonlinearities and high dimensionalities. If you want interpretable results, you might prefer projection pursuit regression or additive models. However, there are NN analogs of projection pursuit regression and additive models, as well as principal components and maximum redundancy analysis, so your options in NNs are quite broad. ... You might use an MLP, rather than _linear_ regression, if you have reason to expect nonlinear relationships. You might use a flexible nonlinear model such as an MLP, rather than a specific parametric nonlinear model, if you have no prior knowledge from which to construct a specific parametric model. You might use an MLP, rather than kernel or k-nearest-neighbor regression, if you have many predictors, some of which are likely to be irrelevant. You might use an MLP, rather than projection pursuit regression or LOESS, if you want a simple formula for rapidly computing predicted values. ... One minor problem is that NNs are usually ill-conditioned or outright singular, but one can deal with this by looking at various diagnostics for ill-conditioning and taking various types of remedial action." -- Warren S. Sarle (saswss@unx.sas.com), sci.stat.consult, 11/21/94.

"The other unique element in neural nets is the idea of simultaneously selecting all basis elements using backpropagation. My first impression of this methods was that it was bound to fail by winding up in poor local minima. This does not seem to happen and the why is mysterious." -- Breiman, reviewing B. Cheng and D. Titterington, "Neural networks: a review from a statistical perspective," Statistical Science (1 994). [William Verkooijen (william@kub.nl), comp.ai.neural-nets, 11/21/94.]

Students of quantitative uncertainty may want the newly updated 20-page "Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results, TN 1297." Free from NIST Calibration Program, A104 Bldg. 411, Gaithersburg, MD 20899-0001; (301) 975-2002, (301) 926-2884 Fax. ($17.50 from NTIS, #PB 95-143087.) [Linda Joy , NIST UPDATE, 2/21/95.]

Neural-network conferences, workshops, and calls for papers are listed on the IDIAP neural network home page, , and by FTP of /html/NN-events/{conferences,workshops,other}.txt.Z on ftp.idiap.ch. [Georg Thimm , connectionists, 2/23/95. Chuck Morefield.]