Goncola, *Ref: Do you have a theory on why your model performance degrades with time* *"We are using streaming temporal data coming from an ECG sensor and a oximeter/PPG sensor alone and although we’re capable of tuning the algorithm to perform really well on the first few days the model degrades as longer periods of time go by."*
If the reason is new information that was not previously modeled, Then HTM may help as it is can learn continuously much beyond its initial training. However when you put a system in production use, you may want to somewhat control learning to have a predictable behavior [Although not everyone agree here] . You may want to test this out there are quite a few examples with EEG data .Here is a link to a hackathon video. https://www.youtube.com/watch?v=UEh48KOmkIA Chandan On Wed, Jan 14, 2015 at 3:55 PM, Gonçalo Sá <[email protected]> wrote: > Hi there guys, > > I was roaming through the net today and noticed Numenta for the first time. > > I’ve glanced over the whole idea of HTM and speed read the first chapters > of On Intelligence and found it so amazing I knew I had to dig deeper. > > To give you some background before I lay my question: > > I’m working on an inexpensive wearable blood pressure sensor with some > more colleagues as a personal project. It should be simple enough, the > catch is that we want to make it cuffless. > > Although a good amount of papers has already leaned on the subject no one > has made significant breakthrough when it comes to measuring SBP only > through a small array of biometric sensors (in our case very small, to make > the device inexpensive). > > We are using streaming temporal data coming from an ECG sensor and a > oximeter/PPG sensor alone and although we’re capable of tuning the > algorithm to perform really well on the first few days the model degrades > as longer periods of time go by. > > So, to sum it up: > > Would I benefit something in predictive accuracy of the algo if I were to > change to HTM? > > My readings so far have revealed fruitful and I’m getting pumped to try > this but would very much like to hear your opinions first! :) > > Gonçalo Sá >
