Machine Learning

Lung Nodule Classification using Convolutional Neural Networks

In this project, we applied deep learning to analyze medical imagery and used CNN to classify lung cancers. We evaluated different preprocessing methods on the dataset and used different neural networks (including ResNet-50 and Inception-v3) to extract features, and then used XGBoost to classify the data using different hyperparameters. In addition, we attempted to train a simple three-layer 3D CNN on the original dataset and tested for its performance. Results show that the best prediction was made by using ResNet and XGBoost with minimum preprocessing with a log loss 0.573 on the validation set. The naive 3D CNN implemented also showed performance that matches that of the pre-trained ResNet, showing its potential in solving the presented problem. PDF

A Comparison of different CNNs for the Tiny ImageNet Visual Recognition

This current project uses Convolutional Naural Networks (CNNs) trained on GPUs for classifying images in the tiny ImageNet dataset. Results show that while using pre­trained VGG model and MatConvNet in matlab offers a quick method to predict classes, the accuracy is not high enough. On the other hand, using CaffeNet and Googlenet takes a longer time, but both result in a much higher accuracy for the validation dataset. PDF

A Comparison of Classical Supervised Machine Learning Algorithms

This paper uses several classic machine learning algorithms on datasets obtained from UCI dataset, and the result is com- pared with Caruana and Niculescu-Mizils research paper. Re- sult shows that without calibrating the model, bagged deci- sion tree has the best performance, which is consistent with the result in Caruana and Niculescu-Mizils paper. PDF

Signal-Verses-Background Classification of Higgs Boson Using Machine Learning Methods

The study of experimental particle physics requires distinguishing exotic particles generated by particle collision with the background noises. The non-linear function in hyperspace gives major obstacles in classification when using data analysis and machine learning methods. In this paper, several different classification methods are used on the benchmark datasets of Higgs bosons and supersymmetry particles and are evaluated based on their performance on the dataset. The result shows significant improvement in accuracy when using deep learning neural network. PDF


Droplet and Pilot Wave Theory

The quantum-like behavior of the walking droplet raises the interest in research. For this project, we examined the behavior of the walking droplet within the boundary and the single slit experiment using the walking droplet. Basically we are trying to replicate previous works. Results show that though the data is not as good as expected, we are still able to see the diffraction patterns for the velocity in random walking, as well as the diffraction though single slit. PDF

Brownian Motor

Brownian motion is a continuous-time stochastic process with stationary independent increments. In biology, Brownian motion is common due to thermal fluctuations. However, in macroscopic view motors are often unidirectional, and scientists propose that non- equilibrium fluctuations can bias the Brownian motion of particles. Many protein-based molecular motors can be Brownian motors. In this paper I will review several relevant researches simply one and two state models regarding one direction, more theoretical models for potentials and forces, and possible cooperation of several motors. PDF


Compliments and Initial Meetings

The syntactic structure of compliments as linguistic events has been studied in depth. Less studied is the compliments role in social interactions. In that vein, the Pilot study aims to observe the distributed cognitive effect of compliments in initial meetings. Do observations support the role of compliments in fostering solidarity (H1), as Face Threatening Acts, or are effects attributed to perceived attractiveness (HA)? The use of a compliment was correlated with a significant increase in duration spent smiling and gazing towards one’s conversational partner. There was no apparent correlation between compliments and body orientation, gesture, or perceived likability. Those who gave a compliment rated recipients as less attractive on average on a scale of 1-10. PDF PPT