Indiana University-Purdue University Indianapolis (IUPUI) / This thesis aims to design and develop the spatial adaptation approach through spatial transformers to improve the accuracy of human keypoint recognition models. We have studied different model types and design choices to gain an accuracy increase over models without spatial transformers and analyzed how spatial transformers increase the accuracy of predictions. A neural network called Widenet has been leveraged as a specialized network for providing the parameters for the spatial transformer. Further, we have evaluated methods to reduce the model parameters, as well as the strategy to enhance the learning supervision for further improving the performance of the model. Our experiments and results have shown that the proposed deep learning framework can effectively detect the human key points, compared with the baseline methods. Also, we have reduced the model size without significantly impacting the performance, and the enhanced supervision has improved the performance. This study is expected to greatly advance the deep learning of human key points and movement dynamics.

In this thesis, the autonomization of extracting information from PDFs of Swedish film scriptsthrough various machine learning techniques and named entity recognition (NER) is explored.Furthermore, it is explored if labeled data needed for the NER tasks can be reduced to some degreewith the goal of saving time. The autonomization process is split into two subsystems, one forextracting larger chunks of text and one for extracting relevant information through named entitiesfrom some of the larger text-chunks using NER. The methods explored for accelerating the labelingtime for NER are active learning and self learning. For active learning, three methods are explored:Logprob and Word Entropy as uncertainty based active learning methods, and active learning byprocessing surprisal (ALPS) as a diversity based method. For self learning, Logprob and WordEntropy are used as they are uncertainty based sampling methods. The results find that ALPS isthe highest performing active learning method when it comes to saving time on labeling data forNER. For Self learning Word Entropy proved a successful method, whereas Logprob could notsufficiently be used for self learning. The entire script reading system is evaluated by competingagainst a human extracting information from a film script, where the human and system competeson time and accuracy. Accuracy is defined a custom F1-score based on the F1-score for NER.Overall the system performs magnitudes faster than human level, while still retaining fairly highaccuracy. The system for extracting named entities had quite low accuracy, which is hypothesisedto mainly be due to high data imbalance and too little diversity in the training data.Teknisk-naturvetenskapliga fakultetenUppsala universitet, Utgivningsort UppsalaHandledare: Björn Mosten Ämnesgranskare: Maria Andrína Fransisco Rodriguez

Recent advances in self-supervised learning have made it possible to reuse information-rich models that have been generally pre-trained on massive amounts of data for other downstream tasks. But the pre-training process can be drastically different from the fine-tuning training process, which can lead to inefficient learning. We address this disconnect in training dynamics by structuring the learning process like an open system in thermodynamics. Open systems can achieve a steady state when low-entropy inputs are converted to high-entropy outputs. We modify the the model and the learning process to mimic this behavior, and attend more to elements of the input sequence that exhibit greater changes in entropy. We call this architecture the Open System Neural Network (OSNN). We show the efficacy of the OSNN on multiple classification datasets with a variety of encoder-only Transformers. We find that the OSNN outperforms nearly all model specific baselines, and achieves a new state-of-the-art result on two classification datasets.

Humans are able to develop a solid knowledge of the world around them: they can leverage information coming from different sources (e.g., language, vision), focus on the most relevant information from the input they receive in a given life situation, and exploit what they have learned before without forgetting it. In the field of Artificial Intelligence and Computational Linguistics, replicating these human abilities in artificial models is a major challenge. Recently, models based on pre-training and on attention mechanisms, namely pre-trained multimodal Transformers, have been developed. They seem to perform tasks surprisingly well compared to other computational models in multiple contexts. They simulate a human-like cognition in that they supposedly rely on previously acquired knowledge (transfer learning) and focus on the most important information (attention mechanisms) of the input. Nevertheless, we still do not know whether these models can deal with multimodal tasks that require merging different types of information simultaneously to be solved, as humans would do. This thesis attempts to fill this crucial gap in our knowledge of multimodal models by investigating the ability of pre-trained Transformers to encode multimodal information; and the ability of attention-based models to remember how to deal with previously-solved tasks. With regards to pre-trained Transformers, we focused on their ability to rely on pre-training and on attention while dealing with tasks requiring to merge information coming from language and vision. More precisely, we investigate if pre-trained multimodal Transformers are able to understand the internal structure of a dialogue (e.g., organization of the turns); to effectively solve complex spatial questions requiring to process different spatial elements (e.g., regions of the image, proximity between elements, etc.); and to make predictions based on complementary multimodal cues (e.g., guessing the most plausible action by leveraging the content of a sentence and of an image). The results of this thesis indicate that pre-trained Transformers outperform other models. Indeed, they are able to some extent to integrate complementary multimodal information; they manage to pinpoint both the relevant turns in a dialogue and the most important regions in an image. These results suggest that pre-training and attention play a key role in pre-trained Transformers’ encoding. Nevertheless, their way of processing information cannot be considered as human-like. Indeed, when compared to humans, they struggle (as non-pre-trained models do) to understand negative answers, to merge spatial information in difficult questions, and to predict actions based on complementary linguistic and visual cues. With regards to attention-based models, we found out that these kinds of models tend to forget what they have learned in previously-solved tasks. However, training these models on easy tasks before more complex ones seems to mitigate this catastrophic forgetting phenomenon. These results indicate that, at least in this context, attention-based models (and, supposedly, pre-trained Transformers too) are sensitive to tasks’ order. A better control of this variable may therefore help multimodal models learn sequentially and continuously as humans do.

Measuring semantic similarity between two sentences is an ongoing research field with big leaps being taken every year. This thesis looks at using modern methods of semantic similarity measurement for an ad-hoc information retrieval (IR) system. The main challenge tackled was answering the question "What happens when you don’t have situation-specific data?". Using encoder-based transformer architectures pioneered by Devlin et al., which excel at fine-tuning to situationally specific domains, this thesis shows just how well the presented methodology can work and makes recommendations for future attempts at similar domain-specific tasks. It also shows an example of how a web application can be created to make use of these fast-learning architectures.

No description available.

Understanding the information encoded in the human genome and the influence of each part of the DNA sequence is a fundamental problem of our society that can be key to unveil the mechanism of common diseases. With the latest technological developments in the genomics field, many research institutes have the tools to collect massive amounts of genomic data. Nevertheless, there is a lack of tools that can be used to process and analyse these datasets in a biologically reliable and efficient manner. Many deep learning solutions have been proposed to solve current genomic tasks, but most of the times the main research interest is in the underlying biological mechanisms rather than high scores of the predictive metrics themselves. Recently, state-of-the-art in deep learning has shifted towards large transformer models, which use an attention mechanism that can be leveraged for interpretability. The main drawbacks of these large models is that they require a lot of memory space and have high inference time, which may make their use unfeasible in practical applications. In this work, we test the appropriateness of knowledge distillation to obtain more usable and equally performing models that genomic researchers can easily fine-tune to solve their scientific problems. DNABERT, a transformer model pre-trained on DNA data, is distilled following two strategies: DistilBERT and MiniLM. Four student models with different sizes are obtained and fine-tuned for promoter identification. They are evaluated in three key aspects: classification performance, usability and biological relevance of the predictions. The latter is assessed by visually inspecting the attention maps of TATA-promoter predictions, which are expected to have a peak of attention at the well-known TATA motif present in these sequences. Results show that is indeed possible to obtain significantly smaller models that are equally performant in the promoter identification task without any major differences between the two techniques tested. The smallest distilled model experiences less than 1% decrease in all performance metrics evaluated (accuracy, F1 score and Matthews Correlation Coefficient) and an increase in the inference speed by 7.3x, while only having 15% of the parameters of DNABERT. The attention maps for the student models show that they successfully learn to mimic the general understanding of the DNA that DNABERT possesses.

Detecting neutrinos, especially ultra-high-energy (UHE) neutrinos, is inherently challenging. Highly sensitive detection devices are required to effectively capture these rare particles, which often results in significant noise in the data. This project focuses on enhancing the detection sensitivity of UHE neutrinos interacting with glacier ice by employing deep learning and transformer models. These models are trained on simulated data that mimics the radio signals produced by neutrino interactions in ice. The developed models have demonstrated improved performance compared to current hardware implementations, offering a promising advancement in neutrino detection technology.

Comparative anatomical studies of diverse plant species are vital for the understanding of changes in gene functions such as those involved in solute transport and hormone signaling in plant roots. The state-of-the-art method for confocal image analysis called PlantSeg utilized U-Net for cell wall segmentation. U-Net is a neural network model that requires training with a large amount of manually labeled confocal images and lacks generalizability. In this research, we test a foundation model called the Segment Anything Model (SAM) to evaluate its zero-shot learning capability and whether prompt engineering can reduce the effort and time consumed in dataset annotation, facilitating a semi-automated training process. Our proposed method improved the detection rate of cells and reduced the error rate as compared to state-of-the-art segmentation tools. We also estimated the IoU scores between the proposed method and PlantSeg to reveal the trade-off between accuracy and detection rate for different quality of data. By addressing the challenges specific to confocal images, our approach offers a robust solution for studying plant structure. Our findings demonstrated the efficiency of SAM in confocal image segmentation, showcasing its adaptability and performance as compared to existing tools. Overall, our research highlights the potential of foundation models like SAM in specialized domains and underscores the importance of tailored approaches for achieving accurate semantic segmentation in confocal imaging. / Master of Science / Studying different plant species' anatomy is crucial for understanding how genes work, especially those related to moving substances and signaling in plant roots. Scientists often use advanced techniques like confocal microscopy to examine plant tissues in detail. Traditional techniques like PlantSeg in automatically segmenting plant cells require a lot of computational resources and manual effort in preparing the dataset and training the model. In this study, we develop a novel technique using Segment-Anything-Model that could learn to identify cells without needing as much training data. We found that SAM performed better than other methods, detecting cells more accurately and making fewer mistakes. By comparing SAM with PlantSeg, we could see how well they worked with different types of images. Our results show that SAM is a reliable option for studying plant structures using confocal imaging. This research highlights the importance of using tailored approaches like SAM to get accurate results from complex images, offering a promising solution for plant scientists.

The design of a high-frequency transformer at levels above 1 kVA is limited by the winding and core materials which are available. This res~arch presents methods for the design and modeling of a 10 kVA transformer operating at a frequency of 20 kHz using readily available materials. A special winding technique is employed to increase both energy density and transformation efficiency by reducing leakage inductance and eddy current losses in the windings. The procedures for calculating the equivalent circuit parameters applicable to this design are outlined, and the calculated values compared with the measured quantities. A thermal analysis of the design is also explored using the equivalent circuit model as a basis for the calculation. Some of the calculations are specific to this particular design, whereas others are quite generic, however the overall concepts employed in the design and analysis of this device have widespread application within the area of high-frequency, high-power transformer design. / Master of Science