Tools
Sorted by:
Try your query at:
 |
 |
 |
 |
 |
 |
Results 1 - 10
of
37,210
Table 3: The four communities in this study Information about the community Type of user-generated content
"... In PAGE 18: ...As shown in Table3 there are some differences between the communities; for example, in Biip and HamarUngdom users cannot contribute videos. There is also a difference in the age of the communities, HamarUngdom is particularly old (founded in 2002).... ..."
Table 1: Ten Topic Results At present we are about to run evaluation experiments of the whole system using a larger set of user topics than the sample topics described in this paper. We have build object detectors for Simpsons characters and run these against the test database and are evaluating its accuracy and we will use this as the basis for automatic indexing of the content. In the user experiments we will use real users to perform interactive searching using object- based relevance feedback, with, and without, ostensive relevance feedback on user generated queries.
"... In PAGE 6: ... Future work will include performance evaluation of ostensive relevance feedback by taking inputs from real user searches, applying ostensive shot weighting and comparing a non-ostensive weighting scheme and comparing performance results. Looking at the results for each topic in this baseline as presented in Table1 , there are three main sections. The first is the recall and average precision for the first 100 ranked results, the second is the recall and precision for the first 1000 ranked results while the final section is the recall and precision over all documents (4868 shots).... ..."
Table 1. Outcomes of binary classification
"... In PAGE 3: ... 2 The Experimental Setup The WEKA Machine Learning environment [10] was used as the platform for perform- ing all experiments described in this paper. Classification performance was measured by the standard metrics, which may be described in terms of possible outcomes of binary classification summarized in Table1 . Accuracy is the ratio of correctly classified exam-... ..."
Table 1. Annotation symbols. The system uses an optimizer generator (GENesis) that allows users to generate a wide variety of optimizers from the specification of optimizations in GOSpeL. Since VOSpeL is a visual form of GOSpeL, it can also be accepted by the GENesis tool with the aid of a translator
1992
Cited by 15
Table 2. Deep Learning Differences by Discipline
2005
"... In PAGE 12: ... In fact, many seniors in every area use deep learning approaches at least some of the time. For the deep learning scale ( Table2 ), seniors in the social sciences have the highest average score even after controlling for student characteristics (effect size with controls = 0.26, p lt; 0.... ..."
Table 1: Contemporary learning strategies supporting deep approaches to learning
2002
"... In PAGE 2: ...mphasis on student-centred instruction (p. 45). Many writers have attempted to conceptualise the attributes and nature of learning settings for higher education that promote deep learning through an emphasis on learning processes. Table1 provides a summary and synthesis of the descriptions of a number of researchers and writers who have explored these conditions. A Framework Describing Learning Approaches A number of consistent elements appear to emerge from the literature which describes the conditions under which students can be encouraged to seek understanding and comprehension as distinct from surface level learning in instances where generic skills development is being sought.... ..."
Cited by 1
Table 5: Classification accuracies for the multiview learning compared with the performance of binary SVM processing feature sets separately and in concatenation.
2005
Cited by 2
Table 1. Experiment results in learning drifting concepts Windowing Approach Ensemble Approach CA Approach
"... In PAGE 8: ... Three measures of the performance are calculated for each experiment: 1) convergence time, ct, which is the average time the system remains unstable during updating to a concept change, measured by the number of instances; 2) classification accuracy, acc, which is the percentage of the classifier apos;s correct predictions, and 3) re-learning requirement, , as the total number of times the classifier is re-learned from scratch. l N Results are shown in Table1 . It is observed that almost in every case the CA approach outperforms the windowing and ensemble approaches: it achieves higher classification accuracy; spends less time on updating classifiers upon concept changes; and require less frequently building classifiers from scratch.... ..."
Table 1: A comparison of different reductions from multiclass to binary classification
in Abstract
"... In PAGE 1: ... Given that there are many good binary learning algorithms and many multiclass classifica- tion problems, a common approach has been to create meta-algorithms which use binary classifiers to make multiclass predictions. Table1 summarizes the characteristics of some known reductions from multiclass to binary classification. Here e is the average binary error rate, and the Error column gives an upper bound on the error rate of the multiclass classifier as a function of e.... ..."
Results 1 - 10
of
37,210
