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.

No / Nowadays, there is no argument that deep learning algorithms provide impressive results in many applications of medical image analysis. However, data scarcity problem and its consequences are challenges in implementation of deep learning for the digital histopathology domain. Deep transfer learning is one of the possible solutions for these challenges. The method of off-the-shelf features extraction from pre-trained convolutional neural networks (CNNs) is one of the common deep transfer learning approaches. The architecture of deep CNNs has a significant role in the choice of the optimal learning transferable features to adopt for classifying the cancerous histopathological image. In this study, we have investigated three pre-trained CNNs on ImageNet dataset; ResNet-50, DenseNet-201 and ShuffleNet models for classifying the Breast Cancer Histopathology (BACH) Challenge 2018 dataset. The extracted deep features from these three models were utilised to train two machine learning classifiers; namely, the K-Nearest Neighbour (KNN) and Support Vector Machine (SVM) to classify the breast cancer grades. Four grades of breast cancer were presented in the BACH challenge dataset; these grades namely normal tissue, benign tumour, in-situ carcinoma and invasive carcinoma. The performance of the target classifiers was evaluated. Our experimental results showed that the extracted off-the-shelf features from DenseNet-201 model provide the best predictive accuracy using both SVM and KNN classifiers. They yielded the image-wise classification accuracy of 93.75% and 88.75% for SVM and KNN classifiers, respectively. These results indicate the high robustness of our proposed framework.

Molecular property prediction has a vast range of applications in the chemical industry. A powerful molecular property prediction model can promote experiments and production processes. The idea behind this degree program lies in the use of transfer learning to predict molecular properties. The project is divided into two parts. The first part is to build and pre-train the model. The model, which is constructed with pure attention-based Transformer Layer, is pre-trained through a Masked Edge Recovery task with large-scale unlabeled data. Then, the performance of this pre- trained model is tested with different molecular property prediction tasks and finally verifies the effectiveness of transfer learning.The results show that after self-supervised pre-training, this model shows its excellent generalization capability. It is possible to be fine-tuned with a short period and performs well in downstream tasks. And the effectiveness of transfer learning is reflected in the experiment as well. The pre-trained model not only shortens the task- specific training time but also obtains better performance and avoids overfitting due to too little training data for molecular property prediction. / Prediktion av molekylers egenskaper har en stor mängd tillämpningar inom kemiindustrin. Kraftfulla metoder för att predicera molekylära egenskaper kan främja vetenskapliga experiment och produktionsprocesser. Ansatsen i detta arbete är att använda överförd inlärning (eng. transfer learning) för att predicera egenskaper hos molekyler. Projektet är indelat i två delar. Den första delen fokuserar på att utveckla och förträna en modell. Modellen består av Transformer-lager med attention- mekanismer och förtränas genom att återställa maskerade kanter i molekylgrafer från storskaliga mängder icke-annoterad data. Efteråt utvärderas prestandan hos den förtränade modellen i en mängd olika uppgifter baserade på prediktion av molekylegenskaper vilket bekräftar fördelen med överförd inlärning.Resultaten visar att modellen efter självövervakad förträning besitter utmärkt förmåga till att generalisera. Den kan finjusteras med liten tidskostnad och presterar väl i specialiserade uppgifter. Effektiviteten hos överförd inlärning visas också i experimenten. Den förtränade modellen förkortar inte bara tiden för uppgifts-specifik inlärning utan uppnår även bättre prestanda och undviker att övertränas på grund otillräckliga mängder data i uppgifter för prediktion av molekylegenskaper.

No description available.

No description available.

No description available.

No description available.

Nowadays, damage detection strategies built on the application of Artificial Neural Network tools to define models that mimic the dynamic behavior of structural systems are viral. However, a fundamental issue in developing these strategies for damage assessment is given by the unbalanced nature of the available databases for civil, mechanical, or aerospace applications, which commonly do not contain sufficient information from all the different classes that need to be identified. Unfortunately, when the aim is to classify between the healthy and damaged conditions in a structure or a generic dynamic system, it is extremely rare to have sufficient data for the unhealthy state since the system has already failed. At the same time, it is common to have plenty of data coming from the system under operational conditions. Consequently, the learning task, carried on with deep learning approaches, becomes case-dependent and tends to be specialized for a particular case and a very limited number of damage scenarios. This doctoral research presents a framework for damage classification in dynamic systems intended to overcome the limitations imposed by unbalanced datasets. In this methodology, the model's classification ability is enriched by using lower-level features derived through an improved extraction strategy that learns from a rich audio dataset how to characterize vibration traits starting from human voice recordings. This knowledge is then transferred to a target domain with much less data points, such as a structural system where the same discrimination approach is employed to classify and differentiate different health conditions. The goal is to enrich the model's ability to discriminate between classes on the audio records, presenting multiple different categories with more information to learn. The proposed methodology is validated both numerically and experimentally.

Data is a critical component in a supervised machine learning system. Many successful applications of learning systems on various tasks are based on a large amount of labeled data. For example, deep convolutional neural networks have surpassed human performance on ImageNet classification, which consists of millions of labeled images. However, one challenge in conventional supervised learning systems is their generalization ability. Once a model is trained on a specific dataset, it can only perform the task on those \emph{seen} classes and cannot be used for novel \emph{unseen} classes. In order to make the model work on new classes, one has to collect and label new data and then re-train the model. However, collecting data and labeling them is labor-intensive and costly, in some cases, it is even impossible. Also, there is an enormous amount of different tasks in the real world. It is not applicable to create a dataset for each of them. These problems raise the need for Transfer Learning, which is aimed at using data from the \emph{source} domain to improve the performance of a model on the \emph{target} domain, and these two domains have different data or different tasks. One specific case of transfer learning is Zero-Shot Learning. It deals with the situation where \emph{source} domain and \emph{target} domain have the same data distribution but do not have the same set of classes. For example, a model is given animal images of `cat' and `dog' for training and will be tested on classifying 'tiger' and 'wolf' images, which it has never seen. Different from conventional supervised learning, Zero-Shot Learning does not require training data in the \emph{target} domain to perform classification. This property gives ZSL the potential to be broadly applied in various applications where a system is expected to tackle unexpected situations. In this dissertation, we develop algorithms that can help a model effectively transfer visual and semantic knowledge learned from \emph{source} task to \emph{target} task. More specifically, first we develop a model that learns a uniform visual representation of semantic attributes, which help alleviate the domain shift problem in Zero-Shot Learning. Second, we develop an ensemble network architecture with a progressive training scheme, which transfers \emph{source} domain knowledge to the \emph{target} domain in an end-to-end manner. Lastly, we move a step beyond ZSL and explore Label-less Classification, which transfers knowledge from pre-trained object detectors into scene classification tasks. Our label-less classification takes advantage of word embeddings trained from unorganized online text, thus eliminating the need for expert-defined semantic attributes for each class. Through comprehensive experiments, we show that the proposed methods can effectively transfer visual and semantic knowledge between tasks, and achieve state-of-the-art performances on standard datasets. / Computer and Information Science

The thesis explores the application of Convolutional Neural Networks (CNNs) in the fashion industry, focusing on fashion style classification and image retrieval. Employing transfer learning, the study investigates the effectiveness of fine-tuning pre-trained CNN models to adapt them for a specific fashion recognition task by initially performing an extensive hyperparameter optimisation, utilising the Optuna framework.  The impact of dataset size on model performance was examined by comparing the accuracy of models trained on datasets containing 2000 and 8000 images. Results indicate that larger datasets significantly improve model performance, particularly for more complex models like EfficientNetV2S, which showed the best overall performance with an accuracy of 85.38% on the larger dataset after fine-tuning. The best-performing and fine-tuned model was subsequently used for image retrieval as features were extracted from the last convolutional layer. These features were used in a cosine similarity measure to rank images by their similarity to a query image. This technique achieved a mean average precision (mAP) of 0.4525, indicating that CNNs hold promise for enhancing fashion retrieval systems, although further improvements and validations are necessary. Overall, this research highlights the versatility of CNNs in interpreting and categorizing complex visual data. The importance of well-prepared, targeted data and refined model training strategies is highlighted to enhance the accuracy and applicability of AI in diverse fields.