Emblaze: Illuminating Machine Learning Representations through Interactive Comparison of Embedding Spaces

@article{Sivaraman2022EmblazeIM,
  title={Emblaze: Illuminating Machine Learning Representations through Interactive Comparison of Embedding Spaces},
  author={Venkatesh Sivaraman and Yiwei Wu and Adam Perer},
  journal={27th International Conference on Intelligent User Interfaces},
  year={2022}
}
Modern machine learning techniques commonly rely on complex, high-dimensional embedding representations to capture underlying structure in the data and improve performance. In order to characterize model flaws and choose a desirable representation, model builders often need to compare across multiple embedding spaces, a challenging analytical task supported by few existing tools. We first interviewed nine embedding experts in a variety of fields to characterize the diverse challenges they face… 

Figures and Tables from this paper

ConceptExplainer: Interactive Explanation for Deep Neural Networks from a Concept Perspective
TLDR
C ONCEPT E XPLAINER is a visual analytics system that enables non-expert users to interactively probe and explore the concept space to explain model behavior at the instance/class/global level.

References

SHOWING 1-10 OF 52 REFERENCES
embComp: Visual Interactive Comparison of Vector Embeddings
TLDR
This paper introduces embComp, a novel approach for comparing twoembeddings that capture the similarity between objects, such as word and document embeddings, and surveys scenarios where comparing these embedding spaces is useful, and assesses it by applying it in several use cases.
EmbeddingVis: A Visual Analytics Approach to Comparative Network Embedding Inspection
TLDR
An exploratory visual analytics system that supports comparison of what and how node metrics are preserved across different embedding models and investigation of relationships between node metrics and selected embedding vectors is designed.
Latent Space Cartography: Visual Analysis of Vector Space Embeddings
TLDR
An integrated visual analysis system is presented, enabling users to discover, define, and verify meaningful relationships among data points, encoded within latent space dimensions, and demonstrate how users of this system can compare latent space variants in image generation.
Visual Interaction with Dimensionality Reduction: A Structured Literature Analysis
TLDR
This work systematically studied the visual analytics and visualization literature to investigate how analysts interact with automatic DR techniques, and proposes a “human in the loop” process model that provides a general lens for the evaluation of visual interactive DR systems.
Visual Exploration of Semantic Relationships in Neural Word Embeddings
TLDR
New embedding techniques for visualizing semantic and syntactic analogies, and the corresponding tests to determine whether the resulting views capture salient structures are introduced, to address a number of domain-specific tasks difficult to solve with existing tools.
LAMVI-2: A Visual Tool for Comparing and Tuning Word Embedding Models
TLDR
This work introduces LAMVI-2, a visual analytics system to support a developer in comparing hyperparameter settings and outcomes, focusing on word-embedding models ("deep learning for text") and integrates views to compare both high-level statistics as well as internal model behaviors.
Parallel embeddings: a visualization technique for contrasting learned representations
TLDR
With this tool, users were more accurate and learned strategies for the tasks more quickly, and users' analytical process in the comparison condition was positively influenced by using the tool beforehand.
Exploring dimension-reduced embeddings with Sleepwalk
TLDR
This work presents Sleepwalk, a simple but powerful tool that allows the user to interactively explore an embedding, using colour to depict original or any other distances from all points to the cell under the mouse cursor, and shows how this approach not only highlights distortions, but also reveals otherwise hidden characteristics of the data.
Multidimensional Projection for Visual Analytics: Linking Techniques with Distortions, Tasks, and Layout Enrichment
TLDR
This survey provides detailed analysis and taxonomies as to the organization of MDP techniques according to their main properties and traits, discussing the impact of such properties for visual perception and other human factors and providing future research axes to fill discovered gaps in this domain.
Visual Analytics in Deep Learning: An Interrogative Survey for the Next Frontiers
TLDR
A survey of the role of visual analytics in deep learning research is presented, which highlights its short yet impactful history and thoroughly summarizes the state-of-the-art using a human-centered interrogative framework, focusing on the Five W's and How.
...
...