abstract: Robotic lower limb prostheses provide new opportunities to help transfemoral amputees regain mobility. However, their application is impeded by that the impedance control parameters need to be tuned and optimized manually by prosthetists for each individual user in different task environments. Reinforcement learning (RL) is capable of automatically learning from interacting with the environment. It becomes a natural candidate to replace human prosthetists to customize the control parameters. However, neither traditional RL approaches nor the popular deep RL approaches are readily suitable for learning with limited number of samples and samples with large variations. This dissertation aims to explore new RL based adaptive solutions that are data-efficient for controlling robotic prostheses. This dissertation begins by proposing a new flexible policy iteration (FPI) framework. To improve sample efficiency, FPI can utilize either on-policy or off-policy learning strategy, can learn from either online or offline data, and can even adopt exiting knowledge of an external critic. Approximate convergence to Bellman optimal solutions are guaranteed under mild conditions. Simulation studies validated that FPI was data efficient compared to several established RL methods. Furthermore, a simplified version of FPI was implemented to learn from offline data, and then the learned policy was successfully tested for tuning the control parameters online on a human subject. Next, the dissertation discusses RL control with information transfer (RL-IT), or knowledge-guided RL (KG-RL), which is motivated to benefit from transferring knowledge acquired from one subject to another. To explore its feasibility, knowledge was extracted from data measurements of able-bodied (AB) subjects, and transferred to guide Q-learning control for an amputee in OpenSim simulations. This result again demonstrated that data and time efficiency were improved using previous knowledge. While the present study is new and promising, there are still many open questions to be addressed in future research. To account for human adaption, the learning control objective function may be designed to incorporate human-prosthesis performance feedback such as symmetry, user comfort level and satisfaction, and user energy consumption. To make the RL based control parameter tuning practical in real life, it should be further developed and tested in different use environments, such as from level ground walking to stair ascending or descending, and from walking to running. / Dissertation/Thesis / Doctoral Dissertation Electrical Engineering 2020

Ces dernières années, l'apprentissage profond a complètement changé le domaine de vision par ordinateur. Plus rapide, donnant de meilleurs résultats, et nécessitant une expertise moindre pour être utilisé que les méthodes classiques de vision par ordinateur, l'apprentissage profond est devenu omniprésent dans tous les problèmes d'imagerie, y compris l'imagerie médicale.Au début de cette thèse, la construction de réseaux de neurones adaptés à des tâches spécifiques ne bénéficiait pas encore de suffisamment d'outils ni d'une compréhension approfondie. Afin de trouver automatiquement des réseaux de neurones adaptés à des tâches spécifiques, nous avons ainsi apporté des contributions à l’optimisation d’hyper-paramètres de réseaux de neurones. Cette thèse propose une comparaison de certaines méthodes d'optimisation, une amélioration en performance d'une de ces méthodes, l'optimisation bayésienne, et une nouvelle méthode d'optimisation d'hyper-paramètres basé sur la combinaison de deux méthodes existantes : l'optimisation bayésienne et hyperband.Une fois équipés de ces outils, nous les avons utilisés pour des problèmes d'imagerie médicale : la classification de champs de vue en IRM, et la segmentation du rein en échographie 3D pour deux groupes de patients. Cette dernière tâche a nécessité le développement d'une nouvelle méthode d'apprentissage par transfert reposant sur la modification du réseau de neurones source par l'ajout de nouvelles couches de transformations géométrique et d'intensité.En dernière partie, cette thèse revient vers les méthodes classiques de vision par ordinateur, et nous proposons un nouvel algorithme de segmentation qui combine les méthodes de déformations de modèles et l'apprentissage profond. Nous montrons comment utiliser un réseau de neurones pour prédire des transformations globales et locales sans accès aux vérités-terrains de ces transformations. Cette méthode est validé sur la tâche de la segmentation du rein en échographie 3D. / In the last few years, deep learning has changed irrevocably the field of computer vision. Faster, giving better results, and requiring a lower degree of expertise to use than traditional computer vision methods, deep learning has become ubiquitous in every imaging application. This includes medical imaging applications. At the beginning of this thesis, there was still a strong lack of tools and understanding of how to build efficient neural networks for specific tasks. Thus this thesis first focused on the topic of hyper-parameter optimization for deep neural networks, i.e. methods for automatically finding efficient neural networks on specific tasks. The thesis includes a comparison of different methods, a performance improvement of one of these methods, Bayesian optimization, and the proposal of a new method of hyper-parameter optimization by combining two existing methods: Bayesian optimization and Hyperband.From there, we used these methods for medical imaging applications such as the classification of field-of-view in MRI, and the segmentation of the kidney in 3D ultrasound images across two populations of patients. This last task required the development of a new transfer learning method based on the modification of the source network by adding new geometric and intensity transformation layers.Finally this thesis loops back to older computer vision methods, and we propose a new segmentation algorithm combining template deformation and deep learning. We show how to use a neural network to predict global and local transformations without requiring the ground-truth of these transformations. The method is validated on the task of kidney segmentation in 3D US images.

Recommender systems are widely used in many domains. In this work, the importance of a recommender system in an online learning platform is discussed. After explaining the concept of adding an intelligent agent to online education systems, some features of the Course Networking (CN) website are demonstrated. Finally, the relation between CN, the intelligent agent (Rumi), and the recommender system is presented. Along with the argument of three different approaches for building a community recommendation system. The result shows that the Neighboring Collaborative Filtering (NCF) outperforms both the transfer learning method and the Continuous bag-of-words approach. The NCF algorithm has a general format with two various implementations that can be used for other recommendations, such as course, skill, major, and book recommendations.

abstract: In recent years, Convolutional Neural Networks (CNNs) have been widely used in not only the computer vision community but also within the medical imaging community. Specifically, the use of pre-trained CNNs on large-scale datasets (e.g., ImageNet) via transfer learning for a variety of medical imaging applications, has become the de facto standard within both communities. However, to fit the current paradigm, 3D imaging tasks have to be reformulated and solved in 2D, losing rich 3D contextual information. Moreover, pre-trained models on natural images never see any biomedical images and do not have knowledge about anatomical structures present in medical images. To overcome the above limitations, this thesis proposes an image out-painting self-supervised proxy task to develop pre-trained models directly from medical images without utilizing systematic annotations. The idea is to randomly mask an image and train the model to predict the missing region. It is demonstrated that by predicting missing anatomical structures when seeing only parts of the image, the model will learn generic representation yielding better performance on various medical imaging applications via transfer learning. The extensive experiments demonstrate that the proposed proxy task outperforms training from scratch in six out of seven medical imaging applications covering 2D and 3D classification and segmentation. Moreover, image out-painting proxy task offers competitive performance to state-of-the-art models pre-trained on ImageNet and other self-supervised baselines such as in-painting. Owing to its outstanding performance, out-painting is utilized as one of the self-supervised proxy tasks to provide generic 3D pre-trained models for medical image analysis. / Dissertation/Thesis / Masters Thesis Computer Science 2020

No description available.

In this thesis, we proposed a deep learning based emotion recognition system in order to improve the successive classification rate. We first use transfer learning to extract visual features and use Mel frequency Cepstral Coefficients(MFCC) to extract audio features, and then apply the recurrent neural networks(RNN) with attention mechanism to process the sequential inputs. After that, the outputs of both channels are fused into a concatenate layer, which is processed using batch normalization, to reduce internal covariate shift. Finally, the classification result is obtained by the softmax layer. From our experiments, the video and audio subsystem achieve 78% and 77% respectively, and the feature fusion system with video and audio achieves 92% accuracy based on the RAVDESS dataset for eight emotion classes. Our proposed feature fusion system outperforms conventional methods in terms of classification prediction.

<p>Deep learning solutions to computer vision tasks have revolutionized many industries in recent years, but embedded systems have too many restrictions to take advantage of current state-of-the-art configurations. Typical embedded processor hardware configurations must meet very low power and memory constraints to maintain small and lightweight packaging, and the architectures of the current best deep learning models are too computationally intensive for these hardware configurations. Current research shows that convolutional neural networks (CNNs) can be deployed with a few architectural modifications on Field-Programmable Gate Arrays (FPGAs) resulting in minimal loss of accuracy, similar or decreased processing speeds, and lower power consumption when compared to general-purpose Central Processing Units (CPUs) and Graphics Processing Units (GPUs). This research contributes further to these findings with the FPGA implementation of a YOLOv4 object detection model that was developed with the use of transfer learning. The transfer-learned model uses the weights of a model pre-trained on the MS-COCO dataset as a starting point then fine-tunes only the output layers for detection on more specific objects of five classes. The model architecture was then modified slightly for compatibility with the FPGA hardware using techniques such as weight quantization and replacing unsupported activation layer types. The model was deployed on three different hardware setups (CPU, GPU, FPGA) for inference on a test set of images. It was found that the FPGA was able to achieve real-time inference speeds of 33.77 frames-per-second, a speedup of 7.74 frames-per-second when compared to GPU deployment. The model also consumed 96% less power than a GPU configuration with only approximately 4% average loss in accuracy across all 5 classes. The results are even more striking when compared to CPU deployment, with 131.7-times speedup in inference throughput. CPUs have long since been outperformed by GPUs for deep learning applications but are used in most embedded systems. These results further illustrate the advantages of FPGAs for deep learning inference on embedded systems even when transfer learning is used for an efficient end-to-end deployment process. This work advances current state-of-the-art with the implementation of a YOLOv4 object detection model developed with transfer learning for FPGA deployment.</p>

Quality control is an important part of any production line. It can be done manually but is most efficient if automated. Inspecting qualitycan include many different processes but this thesisisfocusedon the visual inspection for cracks and scratches. The best way of doingthis at the time of writing is with the help of Artificial Intelligence (AI), more specifically Deep Learning (DL).However, these need a training datasetbeforehand to train on and for some smaller companies, this mightnotbean option. This study triesto find an alternative visual inspection method,that does notrelyon atrained deep learning modelfor when trainingdata is severely limited. Our method is to use edge detection algorithmsin combination with a template to find any edge that doesn’t belong. These include scratches, cracks, or misaligned stickers. These anomalies arethen highlighted in the original picture to show where the defect is. Since deep learningis stateof the art ofvisual inspection, it is expected to outperform template matching when sufficiently trained.To find where this occurs,the accuracy of template matching iscompared to the accuracy of adeep learning modelat different training levels. The deep learning modelisto be trained onimage augmenteddatasets of size: 6, 12, 24, 48, 84, 126, 180, 210, 315, and 423. Both template matching and the deep learning modelwas tested on the samebalanceddataset of size 216. Half of the dataset was images of scratched units,and the other half was of unscratched units. This gave a baseline of 50% where anything under would be worse thanjust guessing. Template matching achieved an accuracy of 88%, and the deep learning modelaccuracyrose from 51% to 100%as the training setincreased. This makes template matching have better accuracy then AI trained on dataset of 84imagesor smaller. But a deep learning modeltrained on 126 images doesstart to outperform template matching. Template matching did perform well where no data was available and training adeep learning modelis no option. But unlike a deep learning model, template matching would not need retraining to find other kinds of surface defects. Template matching could also be used to find for example, misplaced stickers. Due to the use of a template, any edge that doesnot match isdetected.  The ways to train deep learning modelis highly customizable to the users need. Due to resourceand knowledge restrictions, a deep dive into this subject was not conducted.For template matching, only Canny edge detection was used whenmeasuringaccuracy. Other edge detection methodssuch as, Sobel, and Prewitt was ruledoutearlier in this study.

L’apprentissage par transfert de réseaux profonds réduit considérablement les coûts en temps de calcul et en données du processus d’entraînement des réseaux et améliore largement les performances de la tâche cible par rapport à l’apprentissage à partir de zéro. Cependant, l’apprentissage par transfert d’un réseau profond peut provoquer un oubli des connaissances acquises lors de l’apprentissage de la tâche source. Puisque l’efficacité de l’apprentissage par transfert vient des connaissances acquises sur la tâche source, ces connaissances doivent être préservées pendant le transfert. Cette thèse résout ce problème d’oubli en proposant deux schémas de régularisation préservant les connaissances pendant l’apprentissage par transfert. Nous examinons d’abord plusieurs formes de régularisation des paramètres qui favorisent toutes explicitement la similarité de la solution finale avec le modèle initial, par exemple, L1, L2, et Group-Lasso. Nous proposons également les variantes qui utilisent l’information de Fisher comme métrique pour mesurer l’importance des paramètres. Nous validons ces approches de régularisation des paramètres sur différentes tâches de segmentation sémantique d’image ou de calcul de flot optique. Le second schéma de régularisation est basé sur la théorie du transport optimal qui permet d’estimer la dissimilarité entre deux distributions. Nous nous appuyons sur la théorie du transport optimal pour pénaliser les déviations des représentations de haut niveau entre la tâche source et la tâche cible, avec le même objectif de préserver les connaissances pendant l’apprentissage par transfert. Au prix d’une légère augmentation du temps de calcul pendant l’apprentissage, cette nouvelle approche de régularisation améliore les performances des tâches cibles et offre une plus grande précision dans les tâches de classification d’images par rapport aux approches de régularisation des paramètres. / Transfer learning with deep convolutional neural networks significantly reduces the computation and data overhead of the training process and boosts the performance on the target task, compared to training from scratch. However, transfer learning with a deep network may cause the model to forget the knowledge acquired when learning the source task, leading to the so-called catastrophic forgetting. Since the efficiency of transfer learning derives from the knowledge acquired on the source task, this knowledge should be preserved during transfer. This thesis solves this problem of forgetting by proposing two regularization schemes that preserve the knowledge during transfer. First we investigate several forms of parameter regularization, all of which explicitly promote the similarity of the final solution with the initial model, based on the L1, L2, and Group-Lasso penalties. We also propose the variants that use Fisher information as a metric for measuring the importance of parameters. We validate these parameter regularization approaches on various tasks. The second regularization scheme is based on the theory of optimal transport, which enables to estimate the dissimilarity between two distributions. We benefit from optimal transport to penalize the deviations of high-level representations between the source and target task, with the same objective of preserving knowledge during transfer learning. With a mild increase in computation time during training, this novel regularization approach improves the performance of the target tasks, and yields higher accuracy on image classification tasks compared to parameter regularization approaches.

<pre><pre><p> </p><p>Recent progress in machine learning has been mainly due to the availability of large amounts of annotated data used for training complex models with deep architectures. Annotating this training data becomes burdensome and creates a major bottleneck in maintaining machine-learning databases. Moreover, these trained models fail to generalize to new categories or new varieties of the same categories. This is because new categories or new varieties have data distribution different from the training data distribution. To tackle these problems, this thesis proposes to develop a family of transfer-learning techniques that can deal with different training (source) and testing (target) distributions with the assumption that the availability of annotated data is limited in the testing domain. This is done by using the auxiliary data-abundant source domain from which useful knowledge is transferred that can be applied to data-scarce target domain. This transferable knowledge serves as a prior that biases target-domain predictions and prevents the target-domain model from overfitting. Specifically, we explore structural priors that encode relational knowledge between different data entities, which provides more informative bias than traditional priors. The choice of the structural prior depends on the information availability and the similarity between the two domains. Depending on the domain similarity and the information availability, we divide the transfer learning problem into four major categories and propose different structural priors to solve each of these sub-problems.</p><p> </p><p>This thesis first focuses on the unsupervised-domain-adaptation problem, where we propose to minimize domain discrepancy by transforming labeled source-domain data to be close to unlabeled target-domain data. For this problem, the categories remain the same across the two domains and hence we assume that the structural relationship between the source-domain samples is carried over to the target domain. Thus, graph or hyper-graph is constructed as the structural prior from both domains and a graph/hyper-graph matching formulation is used to transform samples in the source domain to be closer to samples in the target domain. An efficient optimization scheme is then proposed to tackle the time and memory inefficiencies associated with the matching problem. The few-shot learning problem is studied next, where we propose to transfer knowledge from source-domain categories containing abundantly labeled data to novel categories in the target domain that contains only few labeled data. The knowledge transfer biases the novel category predictions and prevents the model from overfitting. The knowledge is encoded using a neural-network-based prior that transforms a data sample to its corresponding class prototype. This neural network is trained from the source-domain data and applied to the target-domain data, where it transforms the few-shot samples to the novel-class prototypes for better recognition performance. The few-shot learning problem is then extended to the situation, where we do not have access to the source-domain data but only have access to the source-domain class prototypes. In this limited information setting, parametric neural-network-based priors would overfit to the source-class prototypes and hence we seek a non-parametric-based prior using manifolds. A piecewise linear manifold is used as a structural prior to fit the source-domain-class prototypes. This structure is extended to the target domain, where the novel-class prototypes are found by projecting the few-shot samples onto the manifold. Finally, the zero-shot learning problem is addressed, which is an extreme case of the few-shot learning problem where we do not have any labeled data in the target domain. However, we have high-level information for both the source and target domain categories in the form of semantic descriptors. We learn the relation between the sample space and the semantic space, using a regularized neural network so that classification of the novel categories can be carried out in a common representation space. This same neural network is then used in the target domain to relate the two spaces. In case we want to generate data for the novel categories in the target domain, we can use a constrained generative adversarial network instead of a traditional neural network. Thus, we use structural priors like graphs, neural networks and manifolds to relate various data entities like samples, prototypes and semantics for these different transfer learning sub-problems. We explore additional post-processing steps like pseudo-labeling, domain adaptation and calibration and enforce algorithmic and architectural constraints to further improve recognition performance. Experimental results on standard transfer learning image recognition datasets produced competitive results with respect to previous work. Further experimentation and analyses of these methods provided better understanding of machine learning as well.</p><p> </p></pre></pre>