Báo cáo khoa học: "Learning Condensed Feature Representations from Large Unsupervised Data Sets for Supervised Learning" docx

... novel approach for ef-fectively utilizing unsupervised data in addi-tion to supervised data for supervised learn-ing. We use unsupervised data to gener-ate informative condensed feature represen-tations’ ... original features into the condensed features. For this purpose, we deﬁne the feature po-tency, which is evaluated by employing an existing supervised model with unsupervised data sets. Fig-ure ... number of condensed features to bemade; the number of condensed features becomes large if δ is large. Obviously, the upper bound ofthe number of condensed features is the number oforiginal features.To...

Báo cáo khoa học: "Fast, Space-Efﬁcient, non-Heuristic, Polynomial Kernel Computation for NLP Applications" docx

... only the explicitly speciﬁed features, butalso all available sets of size d of features. For d = 2, this means considering all feature pairs,while for d = 3 all feature triplets. In practice, ... al.,2006). The training data is split into several datasetsaccording to an application speciﬁc heuristic. A sep-arate classiﬁer is then trained for each dataset. For example, it might be known ... well. For completeness, ourtoolkit provides code for d = 3 as well as 2.so much space: every rare feature adds many possi-ble feature pairs.We propose a combined method. We ﬁrst splitcommon from...

Báo cáo khoa học: "Using Non-lexical Features to Identify Effective Indexing Terms for Biomedical Illustrations" docx

... features for the extracted terms and usethem to train a classiﬁer that selects the terms thatare useful for indexing the associated images. supervised learning scenario with data obtained from ... identiﬁer as a feature because some fre-quently mapped concepts are consistentlyineffective for indexing the images in ourtraining and evaluation data. For exam-ple, the CUI for “Original,” ... evaluation data. Notice that the associated F1-score for this experiment is only marginally bet-ter than that of the unseen data. This implies thatit is possible to use training data from particularsubdomains...

Tài liệu Báo cáo khoa học: "Applications of GPC Rules and Character Structures in Games for Learning Chinese Characters" doc

... characters, e.g., “鈿鉀鍾” for “鋰”, “裸袖嘿” for “裡”, “湮湩渭＂ for “浬”, “狎猥狠狙” for “狸” , and “黑墨＂ for “里”. We employed similar techniques to recommend characters for In-lists and Out-lists. The database that ... Jhuyin information. In Fig-ure 2, the Jhuyin information is now added beside the sample Chinese words, i.e., “裡面” (li3 mian4). 4 An Open Authoring Tool for the Games Our game platform has ... a player could choose the class level before the game begins. Hence, for example, it is possible for a lower class player to play the games designed for middle or even upper class levels to...

Tài liệu Báo cáo khoa học: "A Novel Feature-based Approach to Chinese Entity Relation Extraction" ppt

... include feature- based and kernel-based classification. Feature- based approaches transform the context of two entities into a liner vector of carefully selected linguistic features, varying from ... entity related information with context information. 3.1 Classification Features The classification is based on the following four types of features. z Entity Positional Structure Features We ... performance, while over-inclusion of complex features might hurt the performance. Previous approaches mainly focused on English relations. Most of them were evaluated on the ACE 2004 data...

Tài liệu Báo cáo khoa học: "Optimizing Typed Feature Structure Grammar Parsing through Non-Statistical Indexing" doc

... the support for eas-ily adding information extracted from further staticanalysis of the grammar rules, while maintainingthe same indexing strategy. Flexibility is one of thereasons for the successful ... used as formal representatives of richgrammatical categories. In this paper, the formal-ism from (Carpenter, 1992) will be used. A TFSGis deﬁned relative to a ﬁxed set of types and set offeatures, ... q speciﬁes the node reachable from q by feature f when one exists. The path valuepartial function δ π q speciﬁes the node reachable from q by following a path of features π when oneexists. TFSs...

Tài liệu Báo cáo khoa học: "HORN EXTENDED FEATURE STRUCTURES: FAST UNIFICATION WITH NEGATION AND LIMITED DISJUNCTION" ppt

... only if it does not satisfy ~. For any set • of feature terms, Mod(&) denotes the set of all feature structures for which each E r~ is true. For a formal definition of satisfaction, ... Horn feature clauses A feature literal is ei- ther an atomic feature term (e.g., (~: a), (~ ~ /~), or _L) or its negation. A feature clause is a finite disjunction £lvt~v vl,n of feature ... using llorn feature logic is positive disjunction; i.e., formu- las of tim form ~rlva2, with both a.l and aa feature terms. Of course, this has nothing in particular to do with feature- term...

Báo cáo khoa học: "Semisupervised condensed nearest neighbor for part-of-speech tagging" pot

... and obtain a condensed set of labeled data points. To this set of labeled data points we add a large number of unlabeled data points labeled by aNN classiﬁer T on the original data set. We use ... labeled and unlabeled data. It ﬁnds a condensed set of labeled and unlabeled data points, typically smaller than what is obtainedusing condensed nearest neighbor on the la-beled data only, and improves ... strength are methods for removing data points close to decision boundaries, CNN ide-49Figure 3: Unlabeled data may help ﬁnd better representa-tives in condensed training sets. ally only removes...

Báo cáo khoa học: "Efﬁcient, Feature-based, Conditional Random Field Parsing" potx

... models tend to work better for small datasetsand discriminative models tend to work better for larger datasets (Ng and Jordan, 2002). Additionally,they made no use of features, one of the primarybeneﬁts ... usuallyprovide much smaller performance gains than thegains possible from good feature engineering. For example, in (Lafferty et al., 2001), when switching from a generatively trained hidden ... potentials take an exponential form. Wehave a feature function, represented by f(r,s), whichreturns a vector with the value for each feature. Wedenote the value of feature fiby fi(r,s) and...

Báo cáo khoa học: "A Progressive Feature Selection Algorithm for Ultra Large Feature Spaces" doc

... uniform distribution over all values of y and an empty feature set. For each candidate feature in a predefined feature space, it computes the likelihood gain achieved by includ-ing the feature ... correspond to the selected features. Then, the SGC algorithm is performed on each of the feature groups and new features are se-lected from each of them. In other words, the feature space splitting ... number of features for each feature sub-space with a certain distribution. We use a simple method for merge() in the experiments reported here, i.e., adding together the features from a set...

Xem thêm