The self-organized nest construction behaviors of European paper wasps (Polistes dominula) show potential for adoption in artificial intelligence and robotic systems where centralized control proves challenging. However, P. dominula nest construction mechanisms are not fully understood. This research investigated how nest structures stimulate P. dominula worker action at different stages of nest construction. A novel stochastic site selection model, weighted by simple rules for cell age, height, and wall count, was implemented in a three-dimensional, step-by-step nest construction simulation. The simulation was built on top of a hexagonal coordinate system to improve precision and performance. Real and idealized nest data were used to evaluate simulated nests via two parameters: outer wall counts and compactness numbers. Structures generated with age-based rules were not significantly different from real nest structures along both parameters.

Modern day musicians rely on a variety of instruments for musical expression. Tones produced from electronic instruments have become almost as commonplace as those produced by traditional ones as evidenced by the plethora of artists who can be found composing and performing with nothing more than a personal computer. This desire to embrace technical innovation as a means to augment performance art has created a budding field in computer science that explores the creation and manipulation of sound for artistic purposes. One facet of this new frontier concerns timbral creation, or the development of new sounds with unique characteristics that can be wielded by the musician as a virtual instrument. This thesis presents Timcat, a software system that can be used to create novel timbres from prerecorded audio. Various techniques for timbral feature extraction from short audio clips, or grains, are evaluated for use in timbral feature spaces. Clustering is performed on feature vectors in these spaces and groupings are recombined using concatenative synthesis techniques in order to form new instrument patches. The results reveal that interesting timbres can be created using features extracted by both newly developed and existing signal analysis techniques, many common in other fields though not often applied to music audio signals. Several of the features employed also show high accuracy for instrument separation in randomly mixed tracks. Survey results demonstrate positive feedback concerning the timbres created by Timcat from electronic music composers, musicians, and music lovers alike.

Self-driving cars rely on their sense of sight to function effectively in chaotic and uncontrolled environments. Thanks to recent developments in computer vision, specifically convolutional neural networks, autonomous vehicles have developed the ability to see at or above human-level capabilities, which in turn has allowed for rapid advances in self-driving cars. Unfortunately, much like humans being confused by simple optical illusions, convolutional neural networks are susceptible to simple adversarial inputs. As there is no overlap between the optical illusions that fool humans and the adversarial examples that threaten convolutional neural networks, little is understood as to why these adversarial examples dupe such advanced models and what effective mitigation techniques might exist to resolve these issues. This thesis focuses on these adversarial images. By extending existing work, this thesis is able to offer a unique perspective on adversarial examples. Furthermore, these extensions are used to develop a novel attack that can generate physically robust adversarial examples. These physically robust instances provide a unique challenge as they transcend both individual models and the digital domain, thereby posing a significant threat to the efficacy of convolutional neural networks and their dependent applications.

Depth detection is a very common computer vision problem. It shows up primarily in robotics, automation, or 3D visualization domains, as it is essential for converting images to point clouds. One of the poster child applications is self driving cars. Currently, the best methods for depth detection are either very expensive, like LIDAR, or require precise calibration, like stereo cameras. These costs have given rise to attempts to detect depth from a monocular camera (a single camera). While this is possible, it is harder than LIDAR or stereo methods since depth can't be measured from monocular images, it has to be inferred. A good example is covering one eye: you still have some idea how far away things are, but it's not exact. Neural networks are a natural fit for this. Here, we build on previous neural network methods by applying a recent state of the art model to panoramic images in addition to pinhole ones and performing a comparative evaluation. First, we create a simulated depth detection dataset that lends itself to panoramic comparisons and contains pre-made cylindrical and spherical panoramas. We then modify monodepth2 to support cylindrical and cubemap panoramas, incorporating current best practices for depth detection on those panorama types, and evaluate its performance for each type of image using our dataset. We also consider the resources used in training and other qualitative factors.

A brain tumor is detected and classified by biopsy that is conducted after the brain surgery. Advancement in technology and machine learning techniques could help radiologists in the diagnosis of tumors without any invasive measures. We utilized a deep learning-based approach to detect and classify the tumor into Meningioma, Glioma, Pituitary tumors. We used registration and segmentation-based skull stripping mechanism to remove the skull from the MRI images and the grab cut method to verify whether the skull stripped MRI masks retained the features of the tumor for accurate classification. In this research, we proposed a transfer learning based approach in conjunction with discriminative learning rates to perform the classification of brain tumors. The data set used is a 3064 T MRI images dataset that contains T1 flair MRI images. We achieved a classification accuracy of 98.83%, 96.26%, and 95.18% for training, validation, and test sets and an F1 score of 0.96 on the T1 Flair MRI dataset.

The fitness device market is young and rapidly growing. More people than ever before take count of how many steps they walk, how many calories they burn, their heart rate over time, and even their quality of sleep. New, and as of yet, unreleased fitness devices have promised the next evolution of functionality with exercise technique analysis. These next generation of fitness devices have wrist and armband style form factors, which may not be optimal for barbell exercises such as back squat, bench press, and overhead press where a sensor on one arm may not provide the most relevant data about a lift. Barbell path analysis is a well-known visual tool to help diagnose weightlifting technique deficiencies, but requires a camera pointed at the athlete that is integrated with motion-tracking software. This camera set up is not available at most gyms, so this motivates the use of a small, unobtrusive sensor to obtain data about an athlete's weightlifting technique. Researchers have shown that an accelerometer attached to a barbell while the athlete is lifting yields just as accurate acceleration information as a camera. The LIFT (Leveraging Information For Training) automated weight lifting coach attempts to implement a simple, unobtrusive system for analyzing and providing feedback on barbell weight lifting technique.

Convolutional neural networks (CNNs) have dominated the computer vision field since the early 2010s, when deep learning largely replaced previous approaches like hand-crafted feature engineering and hierarchical image parsing. Meanwhile transformer architectures have attained preeminence in natural language processing, and have even begun to supplant CNNs as the state of the art for some computer vision tasks. This study proposes a novel transformer-based architecture, the attentional parsing network, that reconciles the deep learning and hierarchical image parsing approaches to computer vision. We recast unsupervised image representation as a sequence-to-sequence translation problem where image patches are mapped to successive layers of latent variables; and we enforce symmetry and sparsity constraints to encourage these mappings take the form of a parse tree. We measure the quality of learned representations by passing them to a classifier and find high accuracy (> 90%) for even small models. We also demonstrate controllable image generation: first by “back translating” from latent variables to pixels, and then by selecting subsets of those variables with attention masks. Finally we discuss our design choices and compare them with alternatives, suggesting best practices and possible areas of improvement.

In the United States alone, common diseases spread among plants account for billions of dollars lost in crop yield each year. This issue is exacerbated in countries with less infrastructure to defend against crop epidemics, and can lead to famine and forced migration. Farmers can seek the help of plant pathology experts to defend against diseases and detect crop irregularities early on. However, access to experts can be difficult, and even those trained in the field may miss symptoms before it is too late. To assist in early disease detection, a number of papers have been released on the potential for machine learning image classifiers to identify healthy plants from infected ones using convolutional neural networks. While these papers are promising, they often fail to implement a set of standardized practices in their model implementation or make use of realistic data sets. This thesis outlines a set of best practices to use when creating a convolutional neural network for plant disease detection. These principles were selected through a combination of related work analysis and generalized best practices on machine learning. A selection of 11 research articles that discuss their own plant disease image classifiers are analyzed on the grounds of these principles to assess their validity. Then, to demonstrate these principles in practice, we trained six models that each follow our set of guidelines to distinguish healthy strawberry plant images from diseased ones. While the focus of our paper centers on the need to use these practices to create field-realistic models, we achieved the best results on our strawberry image classifier using a VGG16 model architecture. We hope that this work will inspire a set of standardized practices to follow when developing a plant disease image classifier, and allow for more accurate model comparisons in the future.

Due to automation, the world is changing at a rapid pace. Autonomous agents have become more common over the last several years and, as a result, have created a need for improved software to back them up. The most important aspect of this greater software is path prediction, as robots need to be able to decide where to move in the future. In order to accomplish this, a robot must know how to avoid humans, putting frame prediction at the core of many modern day solutions. A popular way to solve this complex problem of frame prediction is Auto Encoder LSTMs. Though there are many implementations of this, at its core, it is a neural network comprised of a series of time sensitive processing blocks that shrink and then grow the data’s dimensions to make a prediction. The idea of using Auto Encoder styled networks to do frame prediction has also been adapted by others to make Temporal Encoders. These neural networks work much like traditional Auto Encoders, in which the data is reduced then expanded back up. These networks attempt to tease out a series of frames, including a predictive frame of the future. The problem with many of these networks is that they take an immense amount of computation power, and time to get them performing at an acceptable level. This thesis presents possible ways of pre-processing input frames to these networks in order to gain performance, in the best case seeing a 360x improvement in accuracy compared to the original models. This thesis also extends the work done with Temporal Encoders to create more precise prediction models, which showed consistent improvements of at least 50% for some metrics. All of the generated models were compared using a simulated data set collected from recordings of ground level viewpoints from Cities: Skylines. These predicted frames were then analyzed using a common perceptual distance metric, that is, Minkowski distance, as well as a custom metric that tracked distinct areas in frames. All of the following was run on a constrained system in order to see the effects of the changes as they pertain to systems with limited hardware access.

Communication Accommodation Theory (CAT) states that individuals adapt to each other’s communicative behaviors. This adaptation is called “convergence.” In this work we explore the convergence of writing styles of users of the online music distribution plat- form SoundCloud.com. In order to evaluate our system we created a corpus of over 38,000 comments retrieved from SoundCloud in April 2014. The corpus represents comments from 8 distinct musical genres: Classical, Electronic, Hip Hop, Jazz, Country, Metal, Folk, and World. Our corpus contains: short comments, frequent misspellings, little sentence struc- ture, hashtags, emoticons, and URLs. We adapt techniques used by researchers analyzing other short web-text corpora in order to deal with these problems. We use a supervised machine learning approach to classify the genre of comments in our corpus. We examine the effects of different feature sets and supervised machine learning algorithms on classification accuracy. In total we ran 180 experiments in which we varied: number of genres, feature set composition, and machine learning algorithm. In experiments with all 8 genres we achieve up to 40% accuracy using either a Naive Bayes classifier or C4.5 based classifier with a feature set consisting of 1262 token unigrams and bigrams. This represents a 3 time improvement over chance levels.