Constructing a Quantitative Fusion Layer over the Semantic Level for Scalable Inference

2018 ◽  
pp. 41-53 ◽  
Author(s):  
Andras Gezsi ◽  
Bence Bruncsics ◽  
Gabor Guta ◽  
Peter Antal
Keyword(s):  
Semantic Level ◽  
Scalable Inference

scholarly journals CoCoX: Generating Conceptual and Counterfactual Explanations via Fault-Lines

2020 ◽  
Vol 34 (03) ◽  
pp. 2594-2601
Author(s):  
Arjun Akula ◽  
Shuai Wang ◽  
Song-Chun Zhu
Keyword(s):  
Neural Network ◽  
State Of The Art ◽  
Input Image ◽  
Classification Model ◽  
Learning Models ◽  
Fault Line ◽  
Semantic Level ◽  
Explainable Ai ◽  
Fault Lines ◽  
Classification Category

We present CoCoX (short for Conceptual and Counterfactual Explanations), a model for explaining decisions made by a deep convolutional neural network (CNN). In Cognitive Psychology, the factors (or semantic-level features) that humans zoom in on when they imagine an alternative to a model prediction are often referred to as fault-lines. Motivated by this, our CoCoX model explains decisions made by a CNN using fault-lines. Specifically, given an input image I for which a CNN classification model M predicts class cpred, our fault-line based explanation identifies the minimal semantic-level features (e.g., stripes on zebra, pointed ears of dog), referred to as explainable concepts, that need to be added to or deleted from I in order to alter the classification category of I by M to another specified class calt. We argue that, due to the conceptual and counterfactual nature of fault-lines, our CoCoX explanations are practical and more natural for both expert and non-expert users to understand the internal workings of complex deep learning models. Extensive quantitative and qualitative experiments verify our hypotheses, showing that CoCoX significantly outperforms the state-of-the-art explainable AI models. Our implementation is available at https://github.com/arjunakula/CoCoX

Hierarchical Concept-Driven Language Model

2021 ◽  
Vol 15 (6) ◽  
pp. 1-22
Author(s):  
Yashen Wang ◽  
Huanhuan Zhang ◽  
Zhirun Liu ◽  
Qiang Zhou
Keyword(s):  
Language Model ◽  
Generation Process ◽  
Data Generation ◽  
Modeling Framework ◽  
Long Distance ◽  
Short Text ◽  
Proposed Model ◽  
Scalable Inference ◽  
End To End ◽  
Hidden Layer

For guiding natural language generation, many semantic-driven methods have been proposed. While clearly improving the performance of the end-to-end training task, these existing semantic-driven methods still have clear limitations: for example, (i) they only utilize shallow semantic signals (e.g., from topic models) with only a single stochastic hidden layer in their data generation process, which suffer easily from noise (especially adapted for short-text etc.) and lack of interpretation; (ii) they ignore the sentence order and document context, as they treat each document as a bag of sentences, and fail to capture the long-distance dependencies and global semantic meaning of a document. To overcome these problems, we propose a novel semantic-driven language modeling framework, which is a method to learn a Hierarchical Language Model and a Recurrent Conceptualization-enhanced Gamma Belief Network, simultaneously. For scalable inference, we develop the auto-encoding Variational Recurrent Inference, allowing efficient end-to-end training and simultaneously capturing global semantics from a text corpus. Especially, this article introduces concept information derived from high-quality lexical knowledge graph Probase, which leverages strong interpretability and anti-nose capability for the proposed model. Moreover, the proposed model captures not only intra-sentence word dependencies, but also temporal transitions between sentences and inter-sentence concept dependence. Experiments conducted on several NLP tasks validate the superiority of the proposed approach, which could effectively infer meaningful hierarchical concept structure of document and hierarchical multi-scale structures of sequences, even compared with latest state-of-the-art Transformer-based models.

Expressivism and Crossed Disagreements

2019 ◽  
Vol 86 ◽  
pp. 111-132
Author(s):  
Javier Osorio ◽  
Neftali Villanueva
Keyword(s):  
Semantic Theory ◽  
Semantic Level ◽  
Semantic Account ◽  
Evaluative Language ◽  
Truth Conditional ◽  
Pragmatic Strategies

AbstractThe purpose of this paper is to explore the connection between expressivism and disagreement. More in particular, the aim is to defend that one of the desiderata that can be derived from the study of disagreement, the explanation of ‘crossed disagreements’, can only be accommodated within a semantic theory that respects, at the meta-semantic level, certain expressivistic restrictions. We will compare contemporary dynamic expressivism with three different varieties of contextualist strategies to accommodate the specificities of evaluative language –indexical contextualism – truth-conditional pragmatics –, pragmatic strategies using implicatures, and presuppositional accounts. Our conclusion will be that certain assumptions of expressivism are necessary in order to provide a semantic account of evaluative uses of language that can allow us to detect and prevent crossed disagreements.

scholarly journals Fast and scalable inference of multi-sample cancer lineages

2015 ◽  
Vol 16 (1) ◽  
Author(s):  
Victoria Popic ◽  
Raheleh Salari ◽  
Iman Hajirasouliha ◽  
Dorna Kashef-Haghighi ◽  
Robert B West ◽  
...  
Keyword(s):  
Scalable Inference

scholarly journals Robust and scalable inference of population history from hundreds of unphased whole genomes

2016 ◽  
Vol 49 (2) ◽  
pp. 303-309 ◽  
Author(s):  
Jonathan Terhorst ◽  
John A Kamm ◽  
Yun S Song
Keyword(s):  
Population History ◽  
Whole Genomes ◽  
Scalable Inference

ABOUT PROJECTIONS OFTHE MENTAL STRUCTURE INTO LANGUAGE AND SPEECH (on the example ofthe concept-proposition)

2021 ◽  
pp. 70-79
Author(s):  
N.A. Iliukhina ◽  
Keyword(s):  
Propositional Logic ◽  
Structural Complexity ◽  
Focus Of Attention ◽  
Word Formation ◽  
Semantic Structure ◽  
Semantic Level ◽  
Linguistic Activity ◽  
Structuring Knowledge ◽  
Main Components ◽  
Linguistic Units

The author examines, using the example of a concept-proposition, multiple manifestations of the connection between a mental unit and units of a language system, argues that the principle of structuring knowledge, characteristic of a proposition, is found in the semantic structure of a sentence, a word-formation nest, in the logic of lexical metonymy, in the logic of transferring definitions, as well as in the ability of a noun to represent knowledge of a propositional nature in speech, concludes that there is a deep commonality of mental and linguistic activity. The semantic structure of the sentence and the structure of the syntactic proposition are isomorphic to the structure of the mental proposition. The vectors of the transfer of the definition from one term of the sentence to another, as well as to the designation of the entire situation, often have propositional logic and are closed by the framework of one sentence. Propositional logic is observed in many word-building nests, especially consistent with the structure of nests organized by a polyactant verb. Among the verbal derivatives in the nest, the percentage of lexemes that name the components of the corresponding situation is significant, in some cases - all the main components of the situation. At the lexico-semantic level, the projection of a proposition on the phenomenon of metonymy is described. Among the models of transference, a variety is highlighted, called propositional metonymy. It includes transfers of the name, the vectors of which (shift of the focus of attention) reflect the structure of the proposition. Another manifestation of the connection between a proposition and linguistic units, considered in the article, is the facts of using a single noun in speech of any lexical and grammatical semantics to represent all situations of any structural complexity. The realized perspective of the research (from the mental unit to linguistic units and processes that reveal with it a certain isomorphism in the logic of categorization and conceptualization of knowledge) allows us to reveal an important line of interaction between the mental and linguistic levels.

Attribute theorems as a prtotypic type of relative clauses in contemporary German

2021 ◽  
Vol 3 (138) ◽  
pp. 23-32
Author(s):  
UDAY HATTIM MAHMOD
Keyword(s):  
Relative Clause ◽  
Relative Clauses ◽  
German Language ◽  
Semantic Level

In this project, we try to do justice to the grammatical phenomenon of the current German language, which we want to explain in this research. This study deals with the topic of attribute theorems as a prtotypic type of relative clauses in contemporary German, not only from the grammatical or syntactic level, but also from the semantic level. The presented work thus covers the most important rules of the relative clause as an attribute with regard to: a) Construction and formation of the attribute theorem as a prototypical type of relative clauses in German. b) Meaning and use of attribute theorem as a prototypical type of relative clauses in German.

Sign in / Sign up

Export Citation Format

Share Document