skip to main content

Title: Representing Mixtures of Word Embeddings with Mixtures of Topic Embeddings
Award ID(s):
Author(s) / Creator(s):
; ; ; ; ; ;
Date Published:
Journal Name:
International Conference on Learning Representations
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The Raman spectral behavior of N2, CO2, and CH4in ternary N2–CO2–CH4mixtures was studied from 22°C to 200°C and 10 to 500 bars. The peak position of N2in all mixtures is located at lower wavenumbers compared with pure N2at the same pressure (P)–temperature (T) (PT) conditions. The Fermi diad splitting in CO2is greater in the pure system than in the mixtures, and the Fermi diad splitting increases in the mixtures as CO2concentration increases at constantPandT. The peak position of CH4in the mixtures is shifted to higher wavenumbers compared with pure CH4at the samePTconditions. However, the relationship between peak position and CH4mole fraction is more complicated compared with the trends observed with N2and CO2. The relative order of the peak position isotherms of CH4and N2in the mixtures in pressure–peak position space mimics trends in the molar volume of the mixtures in pressure–molar volume space. Relationships between the direction of peak shift of individual components in the mixtures, the relative molar volumes of the mixtures, and the attraction and repulsion forces between molecules are developed. Additionally, the relationship between the peak position of N2in ternary N2–CO2–CH4mixtures with pressure is extended to other N2‐bearing systems to assess similarities in the Raman spectral behavior of N2in various systems.

    more » « less
  2. Node embedding is the task of extracting concise and informative representations of certain entities that are connected in a network. Various real-world networks include information about both node connectivity and certain node attributes, in the form of features or time-series data. Modern representation learning techniques employ both the connectivity and attribute information of the nodes to produce embeddings in an unsupervised manner. In this context, deriving embeddings that preserve the geometry of the network and the attribute vectors would be highly desirable, as they would reflect both the topological neighborhood structure and proximity in feature space. While this is fairly straightforward to maintain when only observing the connectivity or attribute information of the network, preserving the geometry of both types of information is challenging. A novel tensor factorization approach for node embedding in attributed networks is proposed in this paper, that preserves the distances of both the connections and the attributes. Furthermore, an effective and lightweight algorithm is developed to tackle the learning task and judicious experiments with multiple state-of-the-art baselines suggest that the proposed algorithm offers significant performance improvements in downstream tasks. 
    more » « less