In an unsupervised learning system, the inputs to the learning system, data observations, are the only data elements that effect the model. Hence the model and its inputs are within a closed system. Analysis of the parameters of the model disclose a representations within the language of the data inputs.
For example, a model of data that consists of a 'mean' of the data inputs provides a single parameter that is in the same domain as the inputs. If the inputs are values within a certain range, the mean of the data inputs can be placed in the same domain. Or if we think about the graph that depicts the inputs then we can describe the domain as the axis of the graph.
However, in the discussion about justice and judges at Bar Ilan, I described how the judges create a model that enables the creation of a new language. Where does this new language emanate from?
The model maps the data to the new language. Here is the magic. When I place a new data point into the model, the result of the mapping is a description in a new linguistic space.
For example, after calculating the mean, I take an observation and map it through the model, the output is greater or smaller than the mean. This new linguistic space is a direct result of the model and its data inputs, the closed system. Yet, somehow the closed system enabled a new linguistic space to be created.
This happens because at some point, the model is employed. The statement that a data observation is above the mean is a judgmental statement. Yes, the model (and the closed system) facilitated this new description, yet the closed system alone is unable to make a judgement.
Yet it is a very weak judgement. No one is saying that above the mean is good or bad. The beauty is that the model itself begs the question to be asked, is the data above or below the mean. This is because the compression of the data, the amount of information stored in the model is so large that the model begs for the question (the single interpretation) to be asked.
It is the negative space created by the efficient model that requires the question. Like a vacuum it pulls into itself the question, or pulls the question out of us, the observer.
magic.
For example, a model of data that consists of a 'mean' of the data inputs provides a single parameter that is in the same domain as the inputs. If the inputs are values within a certain range, the mean of the data inputs can be placed in the same domain. Or if we think about the graph that depicts the inputs then we can describe the domain as the axis of the graph.
However, in the discussion about justice and judges at Bar Ilan, I described how the judges create a model that enables the creation of a new language. Where does this new language emanate from?
The model maps the data to the new language. Here is the magic. When I place a new data point into the model, the result of the mapping is a description in a new linguistic space.
For example, after calculating the mean, I take an observation and map it through the model, the output is greater or smaller than the mean. This new linguistic space is a direct result of the model and its data inputs, the closed system. Yet, somehow the closed system enabled a new linguistic space to be created.
This happens because at some point, the model is employed. The statement that a data observation is above the mean is a judgmental statement. Yes, the model (and the closed system) facilitated this new description, yet the closed system alone is unable to make a judgement.
Yet it is a very weak judgement. No one is saying that above the mean is good or bad. The beauty is that the model itself begs the question to be asked, is the data above or below the mean. This is because the compression of the data, the amount of information stored in the model is so large that the model begs for the question (the single interpretation) to be asked.
It is the negative space created by the efficient model that requires the question. Like a vacuum it pulls into itself the question, or pulls the question out of us, the observer.
magic.
No comments:
Post a Comment