Robotics Couse Assignment – August 2, 2021
Machine Learning代做 1.Consider the problem of estimating the value, in Ohms, of a resistor that is part of the control system of a self-driving car···
Machine Learning代做
1.Consider the problem of estimating the value, in Ohms, of a resistor that is part of the control system of a self-driving car. You take the following measurements with two different multimeters that have different levels of accuracy. The rated value of the resistor is 1,000 Ohms. You wish to calculate a better estimate of the ‘true’ resistance.
The values in the table are in Ohms; the standard deviation of each measurement is also indicated. Using these data, and following the method of weighted least squares (weighting each measurement by the inverse of the variance of the uncertainty), what is the best estimate of the resistance?
2.Neurons in artificial neural networks include a nonlinear activation function: the inputs to the neuron are multiplied by the respective weights and summed before being passed through this nonlinear function. Why, briefly, is it important for this function to be nonlinear? (Hint:consider a linear activation function with linear inputs.)
3.Until about ten years ago, neural networks typically used sigmoid or hyperbolic tangent activation functions. What might be one problem with these functions for very large or very small inputs? (Hint: consider the derivatives of these functions far away from zero.)
4.Explain, briefly, why the U-Net neural network architecture is so useful for sematic segmentation tasks. What does the ‘bottleneck’ do?