DML going Mainstream!
DYNAMICAL Machine Learning (DML) is a formal framework for “continuous learning” in IoT applications. The great news is that DML is “backward compatible”! Current machine learning (*static* ML) is Learn Once and Use for Ever” (or “LOUE”, as I like to call it); DML can be applied for LOUE cases also . . . with better results.
While the best way to capture a moving scene is by “video” (which is DML), the current static ML (or “LOUE”) takes still pictures! Clearly, anything that moves (or that is dynamic) is blurred. In certain applications (which are ubiquitous in industrial IoT), a “video frame” can be used instead to get better results.
DML’s true continuous learning applications will take some time to become fully evident; however, DML’s use for existing LOUE cases yielding better results will attract Data Scientists to DML and take it mainstream quickly; this will accelerate the identification of DML continuous learning “killer apps” in industrial IoT.
What does DML offer?
1. At a minimum, DML results will be superior to LOUE (due to the use of “video frame” rather than “still picture”) for ML applications where the sequence order of features has relevance to the classification task.
2. DML is the only principled solution today for Continuous Learning going beyond repeated “snap-shots” of *static* machine learning . . .
3. As machines age, DML adapts to *new* normal: Designed for long-term use.
4. Underlying system “states” provide a meta-model: Offers a more stable description leading to less False Positives.
5. System “states” as “Digital Twin” video: Leading to continuous “closed-loop” performance improvement.
DML is the essential enabler that will allow IoT to turn a manufactured object into a SERVICE!
For more, go to: DYNAMICAL Machine Learning . Here you can dig in deeper and deeper via connected links . . .
PG Madhavan, Ph.D. - “LEADER . . . of a life in pursuit of excellence . . . in IoT Data Science”
This makes so much sense for me both as a physicist and (aspiring) Data Scientist, PG Madhavan; I'm shocked that this hasn't been developed by now. The only constant is change especially in the type of interconnected and (as a consequence) complex systems that we've (un)intentionally built. There is no other way but DML if we want to capitalize on that dynamic behaviour using empirical data.