This paper addresses the high model complexity and overconfident frame labeling of state-of-the-art (SOTA) action segmenters. Their complexity is typically justified by the need to sequentially refine action segmentation through multiple stages of a deep architecture. However, this multistage refinement does not take into account uncertainty of frame labeling predicted...
The use of genetic algorithms to compose music and generate sounds is an area of interest in the artificial intelligence field. Music and instrument sounds have known rules and structures that can be followed which make them well-suited for genetic algorithms. However, genetic algorithms still struggle to generate sounds comparable...
Over 37,000 people die each year in automobile accidents, with many of these fatalities resulting from collisions with emergency vehicles. The rise of autonomous cars creates the need for an accurate and failsafe method of detecting and responding to emergency vehicles safely and on time. This thesis investigates the ability...
Autonomous robotic agents are on their way to becoming in-home personal assistants, construction assistants, and warehouse workers. The degree of autonomy of such systems is reflected by the manner in which we specify goals to them; the abstraction of low-level commands to high-level goals goes hand-in-hand with increased autonomy. In...
Learning to recognize objects is a fundamental and essential step in human perception and understanding of the world. Accordingly, research of object discovery across diverse modalities plays a pivotal role in the context of computer vision. This field not only contributes significantly to enhancing our understanding of visual information but...
While digital inclusivity researchers and software practitioners have been trying to address exclusion biases in Windows, Icons, Menus, and Pointers (WIMP) user interfaces (UIs) for a long time, little has been done to investigate if and how inclusive software design and its methods that have been devised for WIMP UIs...
Global Positioning Systems have allowed for precise timing of power system measurements over wide areas. This newly found capability has the potential to provide much greater insight into the operation of the power system and its response to contingencies, but few analytical techniques currently exist that provide enough robustness and...
Learning latent space representations of high-dimensional world states has been at the core of recent rapid growth in reinforcement learning(RL). At the same time, RL algo- rithms have suffered from ignored uncertainties in the predicted estimates of model-free or model-based methods. In our work, we investigate both of these aspects...
Papers proposing novel machine learning algorithms tend to present the algorithm or technique in question in the best possible light. The standard practice is generally for authors to emphasize their proposed algorithms' performance in the precise setting where it is maximally impressive, often by only fully evaluating their best known...
We explore the application of deep learning to the disparate fields of natural language processing and computational biology. Both the sentences uttered by humans as well as the RNA and protein sequences found within the cells of their bodies can be considered formal languages in computer science, as sets of...
Assessing AI systems is difficult. Humans rely on AI systems in increasing ways, both visible and invisible, meaning a variety of stakeholders need a variety of assessment tools (e.g., a professional auditor, a developer, and an end user all have different needs). We posit that it is possible to provide...
Manufacturing technology has continuously evolved and advanced over the past century; this has led to an increase in the production of consumer and industrial goods driven by simultaneous growth in population and wealth. Despite the resulting economic and labor growth, environmental impacts of manufacturing have increased dramatically due to the...
Top-performing approaches to embodied AI tasks like point-goal navigation often rely on training agents via reinforcement learning over tens of millions (or even billions) of experiential steps – learning neural agents that map directly from visual observations to actions. In this work, we question whether these extreme training durations are...
In this work, we study the problem of learning and improving policies for probabilistic planning problems. In the first part, we train neural network policies for probabilistic planning problems modeled as factored Markov decision problems. The objective is to train problem-specific neural networks via supervised learning to imitate the action...
Causal inference is an important analytical tool to bridge the gap between prediction and decision-making. However, learning a causal network solely from data is a challenging task. In this work, various techniques have been explored for a better and improved causal network learning from data. Firstly, the problem of learning...
Over the last two decades, satisfiability and satisfiability-modulo theory (SAT/SMT) solvers have grown powerful enough to be general purpose reasoning engines throughout software engineering and computer science. However, most practical use cases of SAT/SMT solvers require not just solving a single SAT/SMT problem, but solving sets of related SAT/SMT problems....
We propose an approach for understanding control policies represented as recurrent neural networks. Recent work has approached this problem by transforming such recurrent policy networks into finite-state machines (FSM) and then analyzing the equivalent minimized FSM. While this led to interesting insights, the minimization process can obscure a deeper understanding...
Advances in deep learning based image processing have led to their adoption for a wide range of applications, and in tow with these developments is a dramatic increase in the availability of high quality datasets. With this comes the need to accelerate and scale deep learning applications in order to...
We present a novel multi-objective optimization methodology built upon a multi-agent blackboard framework. This multi-agent blackboard system (MABS) synthesizes blackboard architectures, multi-agent environments, and optimization theory. The blackboard architecture creates the framework for initializing, storing, and solving a multi-objective optimization problem. Multiple agents allow for an optimization problem to be...
Information about named entities (real-world objects) is usually harvested from different sources and organized as a multiple relational directed graph in Knowledge Bases (KBs). KBs play essential roles in many NLP modules including question answering, fact-checking, search engines, etc. KBs are big but still incomplete: relational information among entities is...