The other way to deal with noise in data is Curve Fitting. In this algorithm, all the sensors are evenly distributed initially, as Binary Search. After the first set of data collected, we can do polynomial fitting to those data, and from the function we get, we can calculate the point where the gradient is highest. Then, we move the closest sensor to that point, and get new read. We add the new data to the previous data set, and do another polynomial fitting. From the new function, we can get a new point where the temperature gredient is highest. This point may not be the same as the one we get previously, then we move one sensor to there, and get new data. This procedure itereates until we always get the same point.
The first picture following is the output of the Polynomial Fitting algorithm, and the next picture is the temperature distribution alone the strand of the testbed gotten by even sampling the temperature. Those picture show that the point found by the Polynomial Fitting has the temperature gradient that is quite close to the the actual maximum gradient, and it takes much less steps than Back Track Binary Search. ( the points sampled are much less.)