Announcements

Author: Matthew Keeler

Title: DESIGN NOVELTY EVALUATION THROUGH ORDINAL EMBEDDING: COMPARISON OF NOVELTY AND TRIPLET ERRORS

Date/time: April 19th at 12:00pm

Location: 2164 DeWALT Seminar Room, Glenn L. Martin Hall.

Committee members:
Professor Mark D. Fuge, Chair/Advisor
Professor Shapour Azarm
Professor Nikhil Chopra

Abstract:
A practical and well-studied method for computing the novelty of a design is to construct an ordinal embedding via a collection of pairwise comparisons between items (called triplets), and use distances within that embedding to compute which designs are farthest from the center. Unfortunately, ordinal embedding methods can require a large number of triplets before their primary error measure—the triplet violation error—converges. But if our goal is accurate novelty estimation, is it really necessary to fully minimize all triplet violations? Can we extract useful information regarding the novelty of all or some items using fewer triplets than classical convergence rates might imply? This thesis addresses this question by studying the relationship between triplet violation error and novelty score error when using ordinal embeddings.

We find that estimating the novelty of a set of items via ordinal embedding can require significantly fewer human provided triplets than is needed to converge the triplet error, and that this effect is modulated by the type of triplet sampling method (random versus uncertainty sampling). Having learned this, we propose the use of a custom metric we call the ‘Expected Model Change’ (EMC) which we use to observe when novelty information in the embedding has stopped updating under newly labeled triplets, so that conservative bounding functions need not be used. Moreover, to avoid the dangers of low accuracy in selecting the dimension of the ordinal embedding, we propose use of the Expected Model Change for tuning the embedding dimension to an appropriate value. In this framework, we explore the convergence properties of ordinal embeddings reconstructed from triplets taken from a variety of synthetic and real-world design spaces.

Author: Cashen Diniz

Title: Denoising the Design Space: Diffusion Models for Accelerated Airfoil Shape Optimization

Date/time: April 16th at 9:30am

Location: 2164 DeWALT Seminar Room, Glenn L. Martin Hall.

Committee members:
Professor Mark D. Fuge, Chair/Advisor
Professor James Baeder
Professor Michael Otte

Abstract:

Generative models offer the possibility to accelerate and potentially substitute parts of the often expensive traditional design optimization process. We present Aero-DDM, a novel application of a latent denoising diffusion model (DDM) capable of generating airfoil geometries conditioned on flow parameters and an area constraint. Additionally, we create a novel, diverse dataset of optimized airfoil designs that better reflects a realistic design space than has been done in previous work. Aero-DDM is applied to this dataset, and key metrics are assessed both statistically and with an open-source computational fluid dynamics (CFD) solver to determine the performance of the generated designs. We compare our approach to an optimal transport GAN, and demonstrate that our model can generate designs with superior performance statistically, in aerodynamic benchmarks, and in warm-start scenarios. We also extend our diffusion model approach, and demonstrate that the number of steps required for inference can be reduced by as much as ~86%, compared to an optimized version of the baseline inference process, without meaningful degradation in design quality, simply by using the initial design to start the denoising process.

Author: Anthony Fabbri

Title of dissertation: Second Wave Mechanics

Date/time: April 16th, 2024 at 2:00pm

Location: 2162 DeWALT Conference Room, Glenn L. Martin Hall.

Committee members:

• Dr. Jeffrey Herrmann, Advisor and Chair
• Dr. Cole Miller
• Dr. Yi Xu

Abstract:

The COVID-19 pandemic experienced very well-documented “waves” of the virus’ progression, which can be analyzed to predict future wave
behavior. Common wave shape patterns can be identified in real-time
data, to predict the pattern of mutations that have recently occurred,
but have not become popularly known as yet, to predict the remaining
future outcome of the wave. By only considering the patterns in the
data that could possibly have acted in tandem to generate the observed
results, many false patterns can be eliminated, and, therefore, hidden
variables can be estimated to a very high degree of probability.
Similar mathematical relationships can reveal hidden variables in
other underlying differential equations.

Author: Matthew Weiner

Title of dissertation: A Framework For Remaining Useful Life Prediction and Optimization For Complex Engineering Systems

Date/time: April 15th, 2024 at 10:00am

Zoom: https://umd.zoom.us/j/6365088363

Location: 2164 DeWALT Conference Room, Glenn L. Martin Hall.

Committee members:

• Dr. Shapour Azarm, Advisor and Chair
• Dr. Katrina Groth, Co-Advisor
• Dr. Yunfei Zhao

Abstract:

Author: Yekanth Ram Chalumuri

Title of dissertation: TOWARDS AUTOMATION OF HEMORRHAGE DIAGNOSTICS AND THERAPEUTICS

Date/time:  April 10th, 2024 at 12pm

Location:  ERF 1207 (IREAP Building)

Committee members:

Dr. Jin-Oh Hahn, Advisor and Chair

Dr. Yang Tao, Dean’s Representative

Dr. Hosam Fathy

Dr. Mark Fuge

Dr. Yancy Diaz-Mercado

Dr. Andrew Reisner

Abstract:

This thesis primarily focuses on developing algorithms and methodologies to achieve the broader aim of advancing the care in hemorrhage diagnostics and therapeutics, especially in low resource settings. The first goal of this thesis is to develop algorithms that can detect hemorrhage using non-invasive physiological signals. Our results showed that our machine learning algorithm can successfully classify various types of hypovolemia. We also developed a physics-based approach to detect and estimate internal hemorrhage even when the body is being compensated by fluid resuscitation using continuous non-invasive hemoglobin measurements. We evaluated the model in silico and determined the maximum acceptable measurement noise that can make these algorithms effective.

The second objective is to advance the regulatory aspects of physiological closed-loop control systems in maintaining blood pressure at a desired value during hemorrhage. Physiological closed-loop control systems offer an exciting opportunity to treat hemorrhage in low resource settings but often face regulatory challenges due to safety concerns. A physics-based model with rigorous validation can improve regulatory aspects of such systems but current validation techniques are very naïve. We developed a physics-based model that can predict blood pressure, heart rate, cardiac output, and hematocrit during hemorrhage and resuscitation and validated using a validation framework that uses sampled digital twins for validation along with more rigorous validation metrics. Overall, the goal of this research is to improve the care and outcomes of patients with hemorrhagic shock by utilizing machine learning, inference, and modeling approaches for early detection and treatment.

Author: Sidney Molnar

Title of dissertation: Metareasoning Strategies to Correct Navigation Failures of Autonomous Ground Robots

Date/time: April 8th, 2024 at 10:00am

Zoom: https://umd.zoom.us/j/9892794999?omn=94673990602

Location: 2164 DeWALT Conference Room, Glenn L. Martin Hall.

Committee members:

• Dr. Jeffrey Herrmann, Advisor and Chair
• Dr. Michael Otte
• Dr. Shapour Azarm

Abstract: Due to the complexity of autonomous systems, theoretically perfect path planning algorithms sometimes fail because of unexpected behaviors that occur when these systems interact with various sub-processes like perception, mapping, and goal planning. These failures can prevent the success of a mission, especially in complex and unexplored environments. Metareasoning, or “thinking about thinking,” is one approach to mitigate these planning failures. This project introduces a novel metareasoning approach that employs various methods of monitoring and control to identify and address path planning irregularities that lead to failures. All methods were integrated into the ARL ground autonomy stack, which utilizes both global and local path planning ROS nodes. The monitoring methods proposed include listening to messages from the planning algorithms, assessing the environmental context of the robot, expected progress methods that evaluate the robot’s progress based on its movement capabilities from a milestone checkpoint, and fixed radius methods that use parameters selected based on mission objectives to assess progress from a milestone checkpoint. The control policies introduced are metric-based sequential policies which select new planner combinations based on benchmark robot performance metrics, context-based pairs policies that assess the effects of switching between two planner combinations, and a restart policy that relaunches a new instance of the same planner combination. The study evaluated which combinations of monitoring and control policies improved or degraded navigation performance by assessing how close the robot could get to the final mission goal. Additionally, this thesis suggests areas for further research to determine the conditions under which metareasoning can improve navigation.