2022 CSSE Senior Thesis Poster Presentations
- Meghna Allamudi
- Jonas Buehrle
- Can we cross the line? Security Analysis of a multitenant SaaS cluster hosted in Kubernetes
- poster
- Zeming Chen and Qiyue Gao
- CURRICULUM: Diagnosing Neural Language Models through
Broad-Coverage Linguistic Phenomena
- poster
- Max Dierschke
- Automation of Machine-Learning Pipelines for Next Best Offer Systems
- poster
- Nils Engleder
- Faster and Accessible Human Robot Collaboration Research using XR Technology
- poster
- Carlos Feng
- Cullen LaKemper, Cehong Wang
- Bio-Inspired AI through Developmental Neural Networks
- Abstract. The adult human
brain differs extremely from that of an infant's. At birth, an
individual's brain is much smaller and ready to absorb information.
It develops over the individual's lifetime by tweaking the arrangement
and connectivity of neurons. This type of developmental process is
found all across biology and yet remains critically under-utilized
within the field of artificial intelligence. There are models that use
development to generate the structure of a network, but the process
tends to be halted before the network is used. We propose a new
process to develop neural network structure during its lifetime and
explore the consequences of this approach.
- poster
- Christian Meinzen
- Analysis of Genetic Algorithms for Feature Selection
- poster
- Megan Merz
- Discourse on ASR Measurement: Introducing the ARPOCA Assessment Tool
- Abstract. This project aims to address several problems with
current ASR models. First is the method of measuring performance of
ASR models. ASR is typically measured using WER (word error
rate). However, WER does not provide much feedback to developers; it
is merely a measure of accuracy. Second is the accent gap, which is a
phenomenon in which a model performs better with some accents than
others. Models typically perform worse for nonnative speakers of a
language. Third is that models can be very costly to develop and are
often very resource intensive to create. ARPOCA is new tool for
assessing ASR models using a variety of metrics. ARPOCA stands for
Assessment of ASR Using Phonemes, Originality, Cost, and Accent
Performance. ARPOCA includes a phoneme recognizer, originality
measurement, and cost-based scoring. ARPOCA is designed to be used in
peer reviews or conferences as a tool to give feedback and suggestions
for improvement.
- poster
- Steven Payne
- Why in the World is Learning Continuation Passing Style (CPS) So Hard?
- poster
- Reed Phillips
- Convergence of 𝐴𝐶𝑂ℝ on the Hyperplane Model
- Abstract. The continuous optimization algorithm ACOR (Ant Colony
Optimization over R) is analyzed from a theoretical perspective. It is
shown that ACOR can achieve an exponentially increasing progress rate
on the hyperplane model in any number of dimensions with the
appropriate parameters. The effect on the progress rate of varying
certain parameters is examined.
- poster
- Neelie Shah
- Trust in Human-Robot Collaboration
- poster
- Andrea Wynn, Duncan McKee, Bohdan Vakhitov
- DDoS Detection Using Machine Learning
- poster
- Shixin Wu