### Table 1: A comparison between the Bayesian One-Shot learning algorithm and alternative approaches to object category recogni- tion. The error rate quoted for the Bayesian One-Shot model is for BH training images.

2003

"... In PAGE 6: ... In this case it achieves a recognition rate as high as BKBEB1, given only 1 training ex- ample. Table1 compares our algorithm with some pub- lished object categorization methods. Note that our algo- rithm has significantly faster learning speed due to much... In PAGE 8: ... Conclusions and future work We have demonstrated that given a single example (or just a few), we can learn a new object category. As Table1 shows, this is beyond the capability of existing algorithms. In order to explore this idea we have developed a Bayesian learn- ing framework based on representing object categories with probabilistic models.... ..."

Cited by 92

### Table 1: A comparison between the Bayesian One-Shot learning algorithm and alternative approaches to object category recogni- tion. The error rate quoted for the Bayesian One-Shot model is for BH training images.

2003

"... In PAGE 6: ... In this case it achieves a recognition rate as high as BKBEB1, given only 1 training ex- ample. Table1 compares our algorithm with some pub- lished object categorization methods. Note that our algo- rithm has significantly faster learning speed due to much... ..."

Cited by 92

### Table 3 List of alternative approaches

"... In PAGE 12: ... 6. Experimental setup and results To analyze the performance of the ten alternative approaches summarized in Table3 , a set of experiments were carried out on a group of random test problems. Our performance evaluation technique and the experimental setup is discussed next, followed by the results obtained from these experiments.... In PAGE 15: ... [18]). We evaluate each capacity case with each of the ten alternative approaches listed in Table3 . Hence,... ..."

### Table 1: Representative results for several prefectures obtained from the estimation of the two models

"... In PAGE 9: ... Thus, a multivariate spatial model was the result. Some of the results obtained are shown in Table1 . A detailed pairwise comparison of forecast errors showed that the Multivariate Hierarchical Bayesian approach outperforms the alternatives in 58 % of the cases for the primary sector, 41% for the secondary and 54% for the tertiary sector.... ..."

### Table 4. A comparison of two approaches to extending the Bayesian classifier.

"... In PAGE 22: ... This causes the Cartesian product of two discretized attributes to have 25 values, instead of 100, and leads to substantially more reliable probability estimates, given that the training set sizes are in the hundreds. The domains and training set sizes appear in the first two columns of Table4 . The remaining columns display the accuracy of the Bayesian classifier and extensions, averaged over 24 paired trials, and found by using an independent test set consisting of all examples not in the training set.... In PAGE 22: ... The remaining columns display the accuracy of the Bayesian classifier and extensions, averaged over 24 paired trials, and found by using an independent test set consisting of all examples not in the training set. In Table4 , Accuracy Once shows results for the backward stepwise joining algorithm of Pazzani (1996), forming at most one Cartesian product as determined by the highest accuracy using leave-one-out cross validation on the training set; Entropy Once is the same... ..."

### TABLE 1. Summary of Key Differences Between Frequentist and Bayesian Approaches

### TABLE III ALTERNATIVE APPROACH: SUMMARY OF OBTAINED RESULTS

### Table A. Comparing the second-generation wavelet approach with alternative approaches.

### Table 1. Bayesian Rank Lists for Synthetic Data

"... In PAGE 3: ... Class separability is set to be different between groups and so genes from the different groups have varying degree of importance to the process of classification. The result in Table1 , column Ran- king ahows that the Bayesian approach ranks the different groups successfully, a property which is important for guiding follow-up experiments and not available by filtering gene lists for shared gene names as, for example, used in (Hockley et al., 2006).... ..."