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 29 for “"Low-data regimes"”.
-
Video-based Parkinson's disease detection in low data regimes
Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2025-08-01
-
Learning in Low Data Regimes for Image and Video Understanding
… their increased representational power has allowed for great progress in core areas of computer vision, and in their applications to our day-to-day life. Unfortunately the performance of these systems rests on the "big data" assumption, where large quantities of annotated data are freely and …
-
Learning to Segment with Deep Models in Low-Data Regimes
… to achieve accurate image understanding in low-data regimes. Traditional few-shot semantic segmentation methods often struggle to generalize effectively, primarily due to the limited availability of labeled support examples. This scarcity makes it difficult to capture the full variability of …
-
Energy-Based Models for Medical Computer-Vision Applications in Low-Data Regimes.
… in domains like life sciences, where data is abundant but often lacks extensive annotations due to the reliance on expert ratings. Supervised learning, especially in the life science domain, frequently leads to mediocre results, as models may focus on specific features like texture …
-
Improving Sparse Reward Classifiers for Model-Based Reinforcement Learning in Low-Data Regimes
… the performance and robustness in sparse-reward, low-data regimes. Additionally, I propose modifications to SimPLe [1] a model-based algorithm, and show that they increase its performance for this setup. Finally, I further improve these results by leveraging the model-based setup, in particular I …
-
Addressing low-data regimes in Australian wildlife recognition from camera trap images using modern computer vision
… adoption of motion-sensitive camera traps as a low-disturbance and scalable alternative for wildlife monitoring. While camera traps generate valuable ecological data, they also produce large volumes of imagery that are time-consuming to process manually. In addition, camera trap images are …
-
New Methods For Domain Adaptation And Low Data Deep Learning
Real-world data coming from settings like hospital collections for detecting disease experience multiple sources of distributional shifts. These issues affect the performance of diagnostic methods, reducing the quality of service provided and leading to health or economic harm. Deep learning has …
-
Spectroscopy Pre-trained Transformer (SpecPT): A Universal Spectroscopic Analysis and Redshift Measurement Framework
… physical processes driving galaxy evolution. As datasets grow in scale and complexity, traditional analysis methods face increasing limitations, motivating the development of scalable, data-driven alternatives. This thesis presents SpecPT, a transformer-based deep learning framework for …
-
Data augmentation as an engine: a multidimensional few-shot analysis
… study presents a multidimensional analysis of data augmentation (DA) in few-shot learning (FSL), investigating the dynamics among augmentation strategies, model architectures, and label granularity. We evaluate four FSL paradigms (Linear Probe, Prototypical Networks, SetFit, and LoRA) across …
-
Causality, identifiability, and representation learning for machine learning with non-i.i.d. data
… Their capabilities keep improving as compute, data and model size scale, captured by empirical scaling laws. However, these capabilities are typically evaluated on held-out test sets assumed independent and identically distributed (i.i.d.) with the training data. Hence, while scaling laws …
-
Benchmarking Reinforcement Learning and Off Policy Evaluation for Medical Decision Making
… (RL) methods due to implementation risks, low data availability, short treatment episodes, sparse re[1]wards, partial observations, and heterogeneous treatment effects (HTE). Despite significant interest in developing Dynamic Treatment Regimes (DTRs) for longitudinal patient care scenarios, …
-
Robust and efficient deep learning under data constraints
… everyday applications, the foundational role of data has become more critical than ever. However, the growing demand for high-quality data, coupled with its limited availability and the rising computational costs associated with large-scale datasets, presents substantial challenges to the …
-
Data-Efficient Learning for Robust Vision and Multimodal Models in Open-World Scenarios
… models are typically trained on massive datasets, requiring substantial computational and annotation resources. However, in many real-world applications, collecting and curating such large-scale data is neither scalable nor practical. Moreover, simply increasing the amount of training …
-
Intelligent Fault Diagnosis and Health State Recognition of Rotating Machinery Under Variable Working Conditions Using Deep Transfer Learning
… learning methods often assume that vibration data from a source domain and a target domain share similar feature distributions, an assumption that rarely holds under variable working conditions in industrial settings. Transfer learning mitigates distribution discrepancies, yet important …
-
Methods for Language Model Alignment: Inference-Time Sampling and Preference Fine-Tuning
… LLMs generate fluent text, they often fail to follow instructions, meet users’ objectives, or avoid harmful content. This thesis investigates methods for aligning language models with human preferences and offers both theoretical insights and practical techniques to make the alignment process more …
-
On Using Simulated Annealing in Training Deep Neural Networks
… for a neural network to perform well on new data. This issue is especially prevalent in low-data regimes, or when training for too many epochs. Iterative learning methods have been devised to improve the generalization performance of neural networks when trained for a prolonged duration. …
-
Data-efficient machine learning for predicting compressive strength of fibre-reinforced concrete with waste rubber and recycled aggregate using continual and transfer learning
… machine learning (ML) have been considered as data-driven alternatives. Numerous studies have explored the application of various ML methods in the prediction of concrete’s mechanical properties. However, these methods have some drawbacks. One major limitation is that the ML models are …
-
Nowe metody i zbiory danych do inteligentnego przetwarzania dokumentów
… aims to contribute innovative solutions and datasets to the Intelligent Document Processing (IDP) domain. The focus is set on two key areas within IDP: Span Identification (SI) and Document Understanding (DU). Significant emphasis is placed on addressing the challenges posed by low-data …
-
Learning Robust and General Representations of the Brain Using Large Heterogeneous Multi-Study Neuroimaging Data
… and they often require task-specific labeled datasets that are too small for reliable training. This dissertation addresses these two barriers by learning robust and general representations of brain structure from large, heterogeneous, multi-study MRI data. Using tens of thousands of …
-
Molecular representation learning for extreme low data in vivo toxicity modeling
… to toxicology is severely limited by the "data scarcity bottleneck." Unlike the abundant bioactivity data available for target screening, high-quality toxicity labels from animal models and clinical trials are expensive, ethically constrained, and rare. This thesis addresses the critical …
Page 1 of 2