Adversarial Machine Learning

This lecture overviews Adversarial MachinShoShould you require access to the resource, please contact the author directly.uld you require access to the resource, please contact the author directly.e Learning that has many applications in DNN robustness and in privacy protection. It covers the following topics in detail: Adversarial Examples, Attack Methods, Adversarial Face De-Identification, Adversarial Defenses.

Should you require access to the resource, please contact the author directly.

Deep Semantic Image Segmentation

Semantic image segmentation is a very important computer vision task with several applications in autonomous systems perception, robotic vision and medical imaging. Recent semantic image segmentation methods rely on deep neural networks and aim to assign a specific class label to each pixel of the input image. This lecture overviews the topic and addresses some of the semantic image segmentation challenges, notably: Deep semantic Image Segmentation architectures. Skip connections. U-nets. BiSeNet. Semantic image segmentation performance, computational complexity and generalization.

Should you require access to the resource, please contact the author directly.

Few Shot Object Recognition

This lecture overviews Few Shot Object Recognition that has many applications in image classification, when few training data are available.  It covers the following topics in detail: Few-shot Image Learning definitions and methods. Applications in sports video athlete recognition.

Should you require access to the resource, please contact the author directly.

Special topics in Object Detection

This lecture overviews Special topics in Object Detection that has many applications in embedded computing and drone vision. It covers the following topics in detail: Embedded object detection, Small object detection, Person detection from aerial views.

Should you require access to the resource, please contact the author directly.

Deep Object Detection

Recently, Convolutional Neural Networks (CNNs) have been used for object/target (e.g., face, person, car, pedestrian, road sign) detection with great results. However, using such CNN models on embedded processors for real-time processing is prohibited by HW constraints. In that sense, various architectures and settings will be examined in order to facilitate and accelerate the use of embedded CNN-based object detectors with limited computational capabilities. The following target detection topics will be presented: Object detection as search and classification task. Detection as classification and regression task. Modern architectures for target detection (e.g., RCNN, Faster-RCNN, YOLO v4, SSD, RetinaNet, RBFNet,  CornerNet, CenterNet, DETR), Lightweight architectures, Data augmentation and Deployment are presented in detail. Evaluation and benchmarking measures are detailed.

Should you require access to the resource, please contact the author directly.

Attention and Transformers Networks

In this lecture, the limitations of Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) in effectively processing sequences are emphasized. However, a breakthrough solution known as Transformers is introduced, which addresses these limitations comprehensively. The architecture of Transformers is meticulously described, with a particular emphasis on its fundamental building blocks. These include positional encoding, which captures the sequential information of input data, as well as multi-headed self or cross attention mechanisms that enable the model to capture dependencies between different elements of the sequence. Additionally, the lecture covers important concepts such as residual connections, which aid in the smooth propagation of information through the network, and layer normalization, which ensures stable training and efficient learning. The lecture also delves into the causal self-attention mechanism employed in decoding, enabling the model to generate output sequences in an autoregressive manner. Lastly, a brief mention is made regarding the optimization algorithms used to train Transformers effectively. Overall, this lecture provides a comprehensive understanding of Transformers and its key components, highlighting its ability to overcome the limitations of traditional RNNs and CNNs in sequence processing tasks.