Recap From Day 047
In day 047, we ended the day with learning that specifically, for classification, we can either have the algorithm choose the best single class label for the motion, once the motion is completed, or we could get a posterior probability distribution over class labels. Again after the motion is completed. And for regression, we can only compute a single number from the motion. Again, of course, after the motion has finished.
Today, we’ll continue from where we left off in day 047
Working with time
Motivating other modeling approaches continued
The first algorithm we’ll discuss in the coming days, dynamic time warping, allows us to work a bit more easily with actions that unfold over time. It doesn’t rely on us providing excellent segmentation information and it can gracefully handle changes in speed.
Furthermore, there are all sorts of other interesting things we might want to know about a gesture over time. For example, looking at the actions we’ve described so far with the Wiimote, before an action is completed, what category of action is it likely to be? Or we might want to know where, over the course of one action, are we at a particular point in time? Are we at the beginning of a circle, 10%, 50% of the way through? or looking at our action so far, when is the action likely to finish?
If we are trying to build a computer system that allows us to conduct it, for instance, giving it a gesture to tell it to play a sound, maybe we want it to begin playing exactly as we finish a particular gesture, rather than taking some time to do some computation first.
Or we might be interested in knowing more about how a gesture is performed. How quickly did we draw a circle in the air? How big was it? Did we move through the circle smoothly or did we slow down halfway? or even switched directions and move backwards? Different types of algorithms have been designed to be able to answer these types of questions.
Each approach opens up very different possibilities for the types of interactive systems we might build. And it also comes with different trade-offs, such as the number of examples it needs to learn, the ease with which we can change a model using control parameters or changes to the date, and the number of input dimensions it can reasonable handle.
In the coming days, we’ll select a few relatively general purpose modeling algorithms to focus on. Furthermore, building useful algorithms to model gestures or sounds that change over time, is an active area of research.
It’s awesome to know that you made it to the end of day 048. I hope you found this informative. Thank you for taking time out of your schedule and allowing me to be your guide on this journey. And until next time, be legendary.
Reference