On-line learning

Freund Y, Schapire RE. A decision-theoretic generalization of on-line learning and an application to boosting. J Comput Syst Sci 1997 55 119-139. 

The robustness to sensor drift of the method under study was evaluated using a simple synthetic drift model. A gain for each of the 60 sensors was initiated to 1 after which the gain factor was subject for over 100 random-walk steps taken from a Gaussian distribution with = 0.01. In the on-line learning condition while testing drift robustness, the last unsupervised vector quantization step was run continuously. 

Fig. 2.4. Drift robustness of SVM, new method and new on-line learning method. Solid and dash-dotted lines represent performance on training and test sets respectively. Diamond, cross and circle refers to SVM, new method, and new on-line method respectively. Error bars are given only for performance of new on-line method on test data. At step 75 a complete recalibration is performed.

Drift tolerance was tested according to the description above using this data set and results are shown in Figure 2.4. As can be seen, the new algorithm with on-line learning has a much superior drift tolerance under these conditions. 

In the test of robustness to sensor drift it was shown that when the unsupervised part of the algorithm was allowed to run in on-line training mode drift robustness much superior to SVM and the new algorithm with no on-line learning was demonstrated. This is a promising result, but further characterization of this property is required. Additional evaluation is currently ongoing on a real chemosensor dataset. 

Freund, Y., and R. E. Schapire. 1997. A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences 55(1) 119-139. Eriedman, J., T. Hastie, and R. Tibshirani. In Press. The Elements of Statistical Learning Prediction, Inference and Data Mining. 

Five of the seven pre-lab modules are simple documents with images and links (written in MS Word with embedded hyperlinks and saved as web pages). The pre-lab quizzes were written using the University s On-Line Learning Software IMS BlackBoard). The movies were compiled by 2 students for an undergraduate final year project, and were originally intended to be shown at the start of the laboratory sessions. 

Students are seldom exposed to analyses/equations for cases not described in the book. This situation should be specifically addressed by including representative equations as exercises in derivation under on-line learning modules. 

Flotzinger, D., Kalcher, J., PfurtscheUer, G. Suitabibty of Learning Vector Quantization for on-line learning. In A case study of EEG classification Proc WCNN 93 World Congress on Neural Networks, vol. I, pp. 224-227. Lawrence Erlbaum, Hillsdale, NJ... 

Overall, use of learning systems has two main advantages. First, the construction cost may be lower, certainly if the system is not construeted from scratch but uses existing software tools and components. Second, a learning system is able to adapt to changing conditions, though not always on-line. 

CAS/STlS[Interna.tiona.1. CAS/STN offers stmcture searchable files such as Registry, Beilstein, MARPAT, CASREACT, and Gmelin a variety of learning files, eg, LRegistry, LBeHstein, LMARPAT, and LCASREACT and software products such as STN Express for on-line stmcture and substmcture searching. Chemical Abstracts Service, a division of the American Chemical Society, has pubHshed Chemical Abstracts since 1907 and joindy operates STN International with EIZ Kadsmhe and the Japan Information Center of Science and Technology. 

The space-frequency localization of wavelets has lead other researchers as well (Pati, 1992 Zhang and Benveniste, 1992) in considering their use in a NN scheme. In their schemes, however, the determination of the network involves the solution of complicated optimization problem where not only the coefficients but also the wavelet scales and positions in the input space are unknown. Such an approach evidently defies the on-line character of the learning problem and renders any structural adaptation procedure impractical. In that case, those networks suffer from all the deficiencies of NNs for which the network structure is a static decision. 

Koulouris, A., and Stepanopoulos, G., On-line empirical learning of process dynamics with Wave-Nets, submitted to Comput. Chem. Eng. (1995). 

MATLAB is a formidable mathematics analysis package. We provide an introduction to the most basic commands that are needed for our use. We make no attempts to be comprehensive. We do not want a big, intimidating manual. The beauty of MATLAB is that we only need to know a tiny bit to get going and be productive. Once we get started, we can pick up new skills quickly with MATLAB s excellent on-line help features. We can only learn with hands-on work these notes are written as a "walk-through" tutorial—you are expected to enter the commands into MATLAB as you read along. 

It is also salutary to note figure 2, which reminds us that agreement and correctness are not always linked. [This figure is from the on-line dBase of particle properties http //pdg.lbl.gov.] Systematic errors always exist, and may be much larger in amplitude than expected. In general, deducing from uncertain data that a model is acceptable is not useful scientific progress. One learns from the failure of models, not from their successes. 

Rapid (near real time) Multicomponent capability High precision Can be on-line (no sampling) Can be remote (with fiber optics) Can multiplex multiple sample points Little or no sample preparation Nondestructive Indirect (secondary method) Multicomponent capability (interferences) Large investment (time, money) Steep learning curve Typically not a trace level technology... 

Solstice, the Center for Renewable Energy and Sustainable Technologies, is an on-line resource for sustainable-energy information. Be sure to explore the related website www.crest.org where you will learn about the US. Renewable Energy Policy Project. 

Refs. [i] The free dictionary by Farlex (An on-line, internet dictionary) http //encyclopedia.thefreedictionary.com/resistor, [ii] Gates ED (2000) Introduction to electronic, 4th edn. Thomson Delmar Learning, New... 

---