Browsing by Subject "Neural Network"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Development and Assessment of Interatomic Neural Network Potentials for Reactive Chemistry and Molecular Dynamics Simulations
    (2024-05) Hu, Huakun
    Large, condensed phased, and extended systems remain a challenge for theoretical studies due to the compromise between accuracy and computational cost in their calculations. Machine learning methods are on the rise to solve this trade off by training on large datasets of highly accurate calculations that are traditionally hard to obtain. The development of interatomic machine learning potentials has resulted in the ability to model high-quality potential energy surfaces with near ab initio level of accuracy at low computational cost. However, just like other machine learning applications, such methods face challenges when it comes to quality training data and transferability, specifically to systems of chemical space beyond its training. In this thesis, we present the continuous exploration of utilizing machine learning methods to build and achieve accurate and efficient potential energy surface for bond dissociation and reactive chemistry, and explore sampling techniques that can allow neural network potentials (NNPs) designed to model potential energy surfaces, such as ANI and NequIP, to accurately predict bond dissociation energy and model reactive chemistry, and to obtain transferability beyond its training data across chemical space. Chapter 2 of this work details the start of this endeavor, starting with training NNPs to accurately predict single C-C bond dissociation at the DFT level and then to the CASPT2 level. Chapter 3 of this work continues the exploration to examine the ability of the NNPs to perform molecular dynamics simulations and evaluate their accuracy of high energy and reactive chemical space. In Chapter 4, the transferability of NNPs is extensively tested with alternative systems beyond the initial benchmark research. Finally, Chapter 5 summarizes the overall findings and discuss potential future directions.
  • Thumbnail Image
    Item
    Effects of Macrostructure on Synchrony in SONET Model Neuron Networks
    (2017-04) Kirkeide, Marina; Nykamp, Duane; Baker, Brittany
    The SONET model randomly generates neural networks with microstructure using four second order statistics. In order to make the SONET model more realistic, another parameter was added to the model that would control the macrostructure. The new parameter affects the probability that two neurons are connected based on the distance between the neurons; the closer two neurons are to each other, the more likely they are to be connected. To test the new parameter, L, against the existing parameters, alpha-chain, alpha-converge, alpha-diverge, and alpha-reciprocal, 400 neuron networks were generated with random values of each variable, and a 2000 millisecond simulation was run on each network using the Brian2 neuron simulating software. The synchrony of each network was then measured monotonically. Before the new parameter was added, it was known that the rate of the chain motif, alpha-chain, had the greatest effect on the synchrony of a network. The testing with macrostructure showed alpha-chain was still the most important factor for predicting synchrony, though L, the intensity of the macrostructure, did somewhat affect the synchrony. When the macrostructure of a network was more prominent, had a smaller L value, the network tended to be more synchronous.
  • Thumbnail Image
    Item
    Multi-Modal Brain Tumor Segmentation Model to solve Mutual Inhibition between Modes
    (2023-12-19) Vashishtha, Shridhar
    Medical image segmentation has become a key research area in the machine learning community with brain tumor segmentation as one of the most challenging problems in the field. Brain tumor segmentation using machine learning models can help in diagnosing, treating, and monitoring of brain tumors which would significantly improve the medical care of patients. The aim of this research is to develop a network that could solve the problem of mutual inhibition in multi-modal image segmentation for brain tumors. Specifically, multi-modal image segmentation represents the true day-to-day scenario of brain tumor imaging which will be automated using machine learning networks. Contribution to the multi-modal brain tumor segmentation problem will allow for the fast detection and classification of brain tumors which will lead to improved medical care to patients.
  • Thumbnail Image
    Item
    On the Effectiveness of Neural Networks Classifying the MNIST Dataset
    (2017-03) Blum, Carter W
    Convolutional Neural Networks (CNNs) are the primary driver of the explosion of computer vision. Initially popularized by AlexNet's performance in the ImageNet Competition in 2012, convolutional neural networks have since far-surpassed the traditional `hand-wired' models that were previously used in computer vision. They have been a focus of major investment and research from major institutes such as Google and OpenAI. This project is part 1 of a 2 part project researching potential optimizations of CNNs in the areas of convergence, processing speed, over fitting and accuracy. The first semester of the project implemented several optimizations from literature and combined them with CNNs to analyze their effectiveness. It also lays the groundwork for the second semester of research, which will be focused on combining recurrency from Recurrent Neural Networks (particularly Long Short-Term Memory (LSTM) networks.
  • Thumbnail Image
    Item
    Sharing the Load - Offloading Processing and Improving Emotion Classification for the SoftBank Robot Pepper""
    (2021-04) Savela, Shawn
    Pepper is a humanoid robot created by SoftBank Robotics that was designed and built with the purpose of being used for robot-human interaction. There is an application interface that allows development of custom interactive programs as well as a number of built-in applications that can be extended and used when creating other custom programs for the robot. Among the pre-installed applications are applications that will classify a person's emotion and mood using data from several data points including facial characteristics and vocal pitch and tone. Due to the Covid-19 pandemic many people have been wearing face masks in both public and private areas. Detecting emotions based on facial recognition and voice tone analysis may not be as accurate when a person is wearing a mask. An alternative method that can be used to classify emotion is to analyze the actual words that are spoken by a person. However, this feature is not currently available on Pepper. In this study we describe a software solution that will allow Pepper to perform sentiment classification based on spoken words using a neural network. We will describe the testing procedure that was used to interview participants by Pepper and compare the F1 score of each classification method with each other. Pepper was able to be programmed to use a neural network for emotion classification. A total of 32 participants were interviewed, with the NLP spoken-word analysis classification achieving an averaged F1 score of .2860 as compared to the built-in software average F1 scores of .2362 from the mood application, .1986 from the vocal tone and pitch application, and .0811 from the facial characteristics application.