Computer Engineering Commons^™

Utilizing Trajectory Optimization In The Training Of Neural Network Controllers, Nicholas Kimball

Master's Theses

Applying reinforcement learning to control systems enables the use of machine learning to develop elegant and efficient control laws. Coupled with the representational power of neural networks, reinforcement learning algorithms can learn complex policies that can be difficult to emulate using traditional control system design approaches. In this thesis, three different model-free reinforcement learning algorithms, including Monte Carlo Control, REINFORCE with baseline, and Guided Policy Search are compared in simulated, continuous action-space environments. The results show that the Guided Policy Search algorithm is able to learn a desired control policy much faster than the other algorithms. In the inverted pendulum …

Deep Neural Network Architectures For Modulation Classification Using Principal Component Analysis, Sharan Ramjee, Shengtai Ju, Diyu Yang, Aly El Gamal

The Summer Undergraduate Research Fellowship (SURF) Symposium

In this work, we investigate the application of Principal Component Analysis to the task of wireless signal modulation recognition using deep neural network architectures. Sampling signals at the Nyquist rate, which is often very high, requires a large amount of energy and space to collect and store the samples. Moreover, the time taken to train neural networks for the task of modulation classification is large due to the large number of samples. These problems can be drastically reduced using Principal Component Analysis, which is a technique that allows us to reduce the dimensionality or number of features of the samples …

Lionfish Detection System, Carmelo Furlan, Andrew Boniface

Computer Engineering

Deep neural networks have proven to be an effective method in classification of images. The ability to recognize objects has opened the door for many new systems which use image classification to solve challenging problems where conventional image classification would be inadequate. We trained a large, deep convolutional neural network to identify lionfish from other species that might be found in the same habitats. Google’s Inception framework served as a powerful platform for our fish recognition system. By using transfer learning, we were able to obtain exceptional results for the classification of different species of fish. The convolutional neural network …

Training Neural Networks To Pilot Autonomous Vehicles: Scaled Self-Driving Car, Jason Zisheng Chang

Senior Projects Spring 2018

This project explores the use of deep convolutional neural networks in autonomous cars. Successful implementation of autonomous vehicles has many societal benefits. One of the main benefits is its potential to significantly reduce traffic accidents. In the United States, the National Highway Traffic Safety Administration states that human error is at fault for 93% of automotive crashes. Robust driverless vehicles can prevent many of these collisions. The main challenge in developing autonomous vehicles today is how to create a system that is able to accurately perceive and process the world around it. In 2016, NVIDIA successfully trained a deep convolutional …

Neuron Clustering For Mitigating Catastrophic Forgetting In Supervised And Reinforcement Learning, Benjamin Frederick Goodrich

Doctoral Dissertations

Neural networks have had many great successes in recent years, particularly with the advent of deep learning and many novel training techniques. One issue that has affected neural networks and prevented them from performing well in more realistic online environments is that of catastrophic forgetting. Catastrophic forgetting affects supervised learning systems when input samples are temporally correlated or are non-stationary. However, most real-world problems are non-stationary in nature, resulting in prolonged periods of time separating inputs drawn from different regions of the input space.

Reinforcement learning represents a worst-case scenario when it comes to precipitating catastrophic forgetting in neural networks. …

Implementation Of A Neuromorphic Development Platform With Danna, Jason Yen-Shen Chan

Masters Theses

Neuromorphic computing is the use of artificial neural networks to solve complex problems. The specialized computing field has been growing in interest during the past few years. Specialized hardware that function as neural networks can be utilized to solve specific problems unsuited for traditional computing architectures such as pattern classification and image recognition. However, these hardware platforms have neural network structures that are static, being limited to only perform a specific application, and cannot be used for other tasks. In this paper, the feasibility of a development platform utilizing a dynamic artificial neural network for researchers is discussed.

An Implicit Surface Modeling Technique Based On A Modular Neural Network Architecture, Manuel Carcenac

Turkish Journal of Electrical Engineering and Computer Sciences

Independently from artificial intelligence applications, an artificial neural network can be viewed as a powerful tool for function reconstruction. Previous papers used this property to model an implicit surface out of some control points by reconstructing its underlying scalar field. Such an approach requests the neural network to memorize the control points, which has turned problematic for complex surfaces. In our paper, we show that this problem can be efficiently tackled by adapting the architecture of the neural network to the features compounding the surface: by learning first these features independently and then blending them gradually together, our modular architecture …