Global ETD Search
Search theses and dissertations gathered from participating repositories worldwide. Every result links back to the library that holds it. No account is needed.
Results
Showing 1 to 20 of 33 for “"Distribution shift"”.
-
Methods for Generalization Under Distribution Shift
… where test data closely resembles the training distribution. However, real-world applications often require systems capable of handling more challenging situations -- specifically, adapting to new tasks and extrapolating to data points outside the distribution of the training set. The current …
-
Robust and Fair Machine Learning under Distribution Shift
… this assumption is quite restrictive because the distribution shift can exist from the training data to the test data in many scenarios. In addition, the goal of traditional machine learning model is to maximize the prediction performance, e.g., accuracy, based on the historical training data, …
-
Dynamic pricing of airline ancillaries under distribution shift
Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2024-12-01
-
Adversarial Examples and Distribution Shift: A Representations Perspective
… methods to improve generalization to natural distribution shift and hypothesize that models trained with different notions of feature bias will learn fundamentally different representations. We find that combining such diverse representations can provide a more comprehensive representation of …
-
Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation
Distribution shift is a major source of failure for machine learning models. However, evaluating model reliability under distribution shift can be challenging, especially since it may be difficult to acquire counterfactual examples that exhibit a specified shift. In this work, we introduce the …
-
Enhancing Learning Algorithms via Sublinear-Time Methods
… long relied on the following two assumptions: 1. Distributional assumptions: data satisfies conditions such as Gaussianity or uniformity. 2. No distribution shift: data distribution does not change between training and deployment. While natural and often correct, these assumptions oftentimes do …
-
Realising data-centric UAV autonomy through learning-based prediction and feedback integration
… under non-linear, time-varying dynamics and distribution shift, where classical controllers depend on accurate models and extensive gain tuning. This dissertation develops an end-to-end pipeline — from raw flight logs to a deployable closed loop — that combines data-driven prediction, …
-
Guiding Deep Probabilistic Models
… deep neural networks to learn probability distributions in high-dimensional data spaces. Learning and inference in these models are complicated due to the difficulty of direct evaluation of the differences between the model distribution and the target. This thesis addresses this challenge …
-
Generative gradual domain adaptation with optimal transport
Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2025-05-01
-
A post-processing framework for group fairness
… This is an instance of the more general issue of distribution shift between training and test environments. In addition, privacy constraints, notably differential privacy, require injecting random noise to protect individual data, but it obscures the group statistics needed to enforce fairness. …
-
Methods for Enhancing Robustness and Generalization in Machine Learning
… for improving subgroup robustness and out of distribution generalization of machine learning models. First we introduce a formulation of Group DRO with soft group assignment. This formulation can be applied to data with noisy or uncertain group labels, or when only a small subset of the …
-
Understanding and Estimating the Adaptability of Domain-Invariant Representations
When the test distribution differs from the training distribution, machine learning models can perform poorly and wrongly overestimate their performance. In this work, we aim to better estimate the model’s performance under distribution shift, without supervision. To do so, we use a set of …
-
UNCERTAINTY QUANTIFICATION AND DECOMPOSITION THROUGH BAYESIAN DEEP LEARNING FOR BIG DATA SATELLITE REMOTE SENSING PROBLEMS
… collection/processing and virtual concept drift (distribution shift) detection. These methods apply wherever deterministic deep learning is currently being applied or where it might be applied in the future. The work presented in this dissertation has the potential to positively affect joint …
-
Probing, Improving, and Verifying Machine Learning Model Robustness
… models turn out to be brittle when faced with distribution shifts, making them hard to rely on in real-world deployment. This motivates developing methods that enable us to detect and alleviate such model brittleness, as well as to verify that our models indeed meet desired robustness …
-
Towards Out-of-distribution Problem for Reinforcement Learning
… from the curse of dimensionality and poor out-of-distribution generalization of current probabilistic models. Current machine learning models requires data points to be independently identically distributed which is often not satisfied in real-world applications. This mismatch damages the direct …
-
Reliable and efficient machine learning under distribution shifts
Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2026-12-01
-
Optimizing Video Streaming at Scale Across Devices, Networks, and Temporal Drift
… Additionally, AZEEM addresses temporal distribution shift—where the best-performing configurations change over time—by recommending a small, robust set of candidates rather than a single configuration. Evaluations using largescale real-world datasets show that AZEEM reduces exploration …
-
Ensembles of Quantifiers
… of cases belonging to each class (or class distribution) in a test set, using a training set that may have a substantially different distribution. At a first sight, the most intuitive way to quantification is to count the predictions of a classifier over a test set. This method has already …
-
Heterogeneous machine learning with decentralized data
… data, where multiple clients with distinct data distributions jointly train or adapt machine learning models under the coordination of a central server. Throughout the process, clients’ private data never leave their local devices. This paradigm underlies numerous real-world applications, such as …
-
SAFEGUARDING AI SYSTEMS AGAINST UNEXPECTED INPUTS
… inability to handle data outside the training distribution or knowledge. When facing unseen or otherwise challenging inputs, models often make incorrect decisions without warning users. This thesis improves the safety of machine learning systems by building three stages for handling …
Page 1 of 2