Enhancement of Online Stochastic Gradient Descent using Backward Queried Images
Stochastic gradient descent (SGD) is one of the preferred online optimization algorithms. However, one of its major drawbacks is its predisposition to forgetting previous data when optimizing through a data stream, also known as catastrophic interference. In this project, we attempt to mitigate this drawback by proposing a new low-cost approach which incorporates backward queried images with SGD during online training. Under this new approach, we propose that for every new training sample through the data stream, the neural network is optimized using the corresponding backward queried image from the initial dataset. After compiling the accuracy of the proposed method and SGD under a data-stream of 50,000 training cases with 10,000 test cases and comparing our algorithm to SGD, we see substantial improvements in the performance of the neural network with two different MNIST datasets (Fashion and Kuzushiji), classifying the MNIST datasets at a high accuracy for the mean, minimum, lower quartile, median, and upper quartile, while maintaining lower standard deviation in performance, demonstrating that our proposed algorithm can be a potential alternative to online SGD.
What is the relationship between angular velocity and power efficiency of a twin blanded single rotor helicopter system, in hover?
A traditional helicopter requires 60 - 80% more power to hover than when in forward or lateral flight, making the manoeuvre extremely power inefficient. To maximise efficiency, industrially many properties of the helicopter and rotor have been changed and tested, for example: optimising blade shape, fuselage shape and changing weights of different helicopter components. This report in particular aims to find a relationship between power efficiency and angular velocity for a twin bladed hovering helicopter with a single rotor. The angular velocity of a blade measures the frequency of its revolution about a fixed point. A theoretical approach was first taken and then justified with empirical data. Firstly, a model for power efficiency was derived with William Froude’s momentum and blade element theory. The efficiency equations incorporated the thrust and power coefficients. Therefore, the research focused on determining values for these coefficients by manipulating equations, using industrial specifications and simulations from the XFOIL software. In order to validate the accuracy for such theoretically generated data, an experiment was conducted for a comparison. The theoretical and empirical data were concurrent, as they followed a similar trend and the empirical values overlapped within the theoretical error bars. The power efficiency for different angular velocities were then found by substituting values for the coefficients. The results demonstrated a positive relationship; where, as angular velocity increases, power efficiency increases too, then plateaus and repeats the same trend once again. The research raises many questions and could be extended by determining if a similar relationship exists for tri-copters and quadcopters.