Intelligent Agent = True Machine Learning
Automated machine learning platforms currently on the market do little more than take known model structures and attempt to fit the data. Their machine learning advantage is in speeding up the experimentation process to identify the champion model. This results in a lack of insights, actionability, and a black-box model that cannot explain its predictions. Compellon’s Intelligent Agent is true machine learning and is revolutionizing the industry.
Compellon’s Intelligent Agent actually learns, understands, and adapts to the environment in which it operates. Built on decades of rigorous research and practical implementation of well proven mathematical approaches, our Agent avoids many of the pitfalls found in conventional analytical methods. Compellon’s Intelligent Agent provides the ability to simplify the process, remove bias, discover hidden relationships, provide strategic and tactical actions, and identify when the environment changes. Let’s take a deeper dive into the Agent to see how it differs from conventional machine learning approaches.
Intelligent Agent Differentiators:
Runs independent of human input into the analytical process, so you can focus on the solution, and not the process of solving your problem.
Has no preconceived idea about the distribution of data or a model structure that best fits the data. It finds the important relationships within the data and builds a custom model. This removes the human bias that is often injected into the analytical process.
Not limited to the types of math it can use to describe the data. Boolean, linear, and non-linear math can be used within the same model structure, giving you a model that best fits the data rather than data fitted to a predefined structure. The results in a reliable and interpretable “clear-box” model.
Aware of the environment it operates in and uses this information to diagnose weaknesses in the model structure or identify changes in the environment. The Agent can also optimize model structures toward particular user objectives. These result in better initial model quality, extended model life and enables unique actionable capabilities.
Once weaknesses have been identified, the Agent can execute a set of improvement strategies in order to achieve targeted quality thresholds. This allows deployment in real-time, dynamic environments.
Compellon’s Intelligent Agent Key Benefits:
Find groups that behave together to best predict future
Advice on what to change and why
Clear-box (As opposed to black-box)
Same process regardless of use case
Answers in seconds/minutes/hours vs days/weeks/months
What makes Compellon’s breakthrough user experience possible is the industry-unique artificial intelligence technology under the hood. Compellon’s approach to AI is to extend (rather than to imitate) human expertise. The engine acts as an autonomous intelligent agent that performs the analytics based only on the user-provided goal and the evidence in the data. This empowers users to focus solely on the business challenges and the actions to achieve their objective, without extensive skills and assumptions, which are typically required to perform advanced analytics.
Compellon’s approach incorporates and builds upon elements of mainstream machine learning practice, while differing in several substantial ways:
Compellon autonomously derives a unique model structure for each individual data set. This is in stark contrast to mainstream statistical and machine learning approaches that always start by assuming a type of structure (such as a decision tree, a logistic regression, or SVM) or use trial and error to choose from multiple structures.
Self-Aware & Adaptive
The Compellon engine stores the image of its own structure and has the ability to change it. By using this characteristically-AI ability, it modifies its own behavior to avoid past errors and to account for changes in environment. This is in contrast to the mainstream machine learning approaches that require human intervention to adapt models, often by using techniques that add complexity (such as “bagging” and “boosting”) to correct errors rather than recognize and remove their reasons.
The Compellon engine consistently uses concepts of information theory (e.g., Shannon entropy and many other measures based on it) for every phase of analysis, including variable selection, discovery of relations, determining the model structure, tuning the model, forming prescriptive output, and providing adaptive capabilities when conditions are changing. Due to this approach, the engine avoids many analytical guesses and arbitrary decisions that would otherwise be necessary.
These capabilities today power the breakthrough user experience that transforms knowledge discovery and predictive modeling with the unique combination of speed, ease, and powerful insight into why and what to do. They also position Compellon with the technology to uniquely address the growing need for autonomous operation to deal with the explosion of data across a variety of use cases.