A 3D Analytic Framework for Visual Data Discovery
A presented 3D Visual Analytic Framework enables Visual Data Discovery, and allows all discovery actions to performed inside 3D Unified Visual Data Representation Space.
By Edward Rotenberg, Nov 2014.
The final word in decision making belongs to humans. To simplify understanding of machine algorithms outputs or arbitrary data sets, we need to take advantage of the most advanced of all the human cognitive systems – the visual cortex. Our version of the Visual Data Discovery Framework attempts to do just that.
To visually present different types of data, we employ a Unified Visual Data Representation Space (UVDRS), where a data set is visually represented as a set of objects and relations between them. This is high level of abstraction and it let us represent almost any data set. We discover data “insight” by finding meaningful relations in data sets together with maximizing amount of information about data objects inside the same screen space.
The UVDRS design was prompted by searching for potential linkages between even the most unthinkable relations. Initial studies of transformative discoveries such as Nobel Prize winning discoveries are particularly promising.
To further simplify visual recognition of the relative strength of relations, we moved our Discovery Framework to a 3D space where we can:
There is one other important application - it is very difficult to detect, explain or predict System Level Events composed of multiple simultaneous local events and involving multiple relations amongst multiple distributed domain objects. The financial “bubbles” are examples of such events. We used our framework to visualize the stock market crash of October 1987.
The following are screen shots from our demo located on http://tradingpipeline.com/ under Products menu item.
After choosing data set organization users can search for valuable relations in a data set by:
BIO: Edward Rotenberg, Ph.D., is a principal at Rational Solutions Group specializing in analytic data visualization and data-mining. His work experience includes AT&T Bell Labs and Microsoft.
He holds a PhD degree in EE and an MS in theoretical cybernetics. Dr. Rotenberg is the author of more than twenty patents in analytic visualization, probabilistic modeling systems, analytical models and methods in process control, adaptive and training systems.
Related:
The final word in decision making belongs to humans. To simplify understanding of machine algorithms outputs or arbitrary data sets, we need to take advantage of the most advanced of all the human cognitive systems – the visual cortex. Our version of the Visual Data Discovery Framework attempts to do just that.
To visually present different types of data, we employ a Unified Visual Data Representation Space (UVDRS), where a data set is visually represented as a set of objects and relations between them. This is high level of abstraction and it let us represent almost any data set. We discover data “insight” by finding meaningful relations in data sets together with maximizing amount of information about data objects inside the same screen space.
The UVDRS design was prompted by searching for potential linkages between even the most unthinkable relations. Initial studies of transformative discoveries such as Nobel Prize winning discoveries are particularly promising.
To further simplify visual recognition of the relative strength of relations, we moved our Discovery Framework to a 3D space where we can:
- Visually represent the relative strength of relations
- Simultaneously explore bigger number of relations and objects
- Use the unique 3D space rotation capabilities in order to:
- overview whole relational volume, concentrating on objects and relation details
- to “literally” see different sides of 3D objects in order to collect a multifaceted object’s information
- In case of complex objects representing data clusters or hierarchies we can
- zoom-in inside cluster’s UVDRS or
- traverse UVDRS hierarchies to find strong or in range relational dependencies
- For objects with multiple attributes
- we can explore relations independently per each attribute’s space or
- use weighted multi-attribute spaces.
There is one other important application - it is very difficult to detect, explain or predict System Level Events composed of multiple simultaneous local events and involving multiple relations amongst multiple distributed domain objects. The financial “bubbles” are examples of such events. We used our framework to visualize the stock market crash of October 1987.
The following are screen shots from our demo located on http://tradingpipeline.com/ under Products menu item.
- 3D spheres represent objects and 3D cylindrical links between them represent relations. Diameter of links proportional to strength of relations between objects and link’s color reflect relation’s behavior.
- Strength of relations calculated as correlation or similarity values between each pair of objects.
- a preexisting classification,
- an output of cluster analysis or
- a classification created on the spot from multiple available options.
After choosing data set organization users can search for valuable relations in a data set by:
- filtering out weak relations or selecting relations inside a given range;
- looking for relations in different object’s attributes spaces;
- employing different types of relations;
- zooming inside clusters and
- traversing classification’s hierarchies and
- selecting and researching any relation separately.
Fig.1: Data Set representation in Unified Visual Data Representation Space.
Fig.2: Data Set after filtration: shown only significant correlation between 0.9 and 1 (max)
Fig.3: The single mode: shown relations strength of one object to all other objects.
He holds a PhD degree in EE and an MS in theoretical cybernetics. Dr. Rotenberg is the author of more than twenty patents in analytic visualization, probabilistic modeling systems, analytical models and methods in process control, adaptive and training systems.
Related: