Here is a brief summary of topics discussed, thus far, in classes. Slides and coding scripts would be distributed after each lecture via the Google groups mailing-list. Refer to the home page on how to subscribe.

27/02/2020 -- Introduction to Advanced Machine Learning. The first class introduces Machine Learning, its applications and main research areas, with particular reference to Computer Vision, as well as Deep Learning. In this first lecture we introduce case studies of 3D reconstruction and object recognition from Computer Vision.

28/02/2020 -- Basic concepts in Computer Vision and Digital Image Filtering. Topics include: basic terminology of Computer Vision, linear filtering with emphasis on Gaussian filtering, multi-cale image representation with emphasis on Gaussian pyramids, edge detection, image 1st and 2nd derivatives, notes on Hough transform and parameterized curves.

12/03/2020 -- Object recognition with color. Topics presented include: challenges in object recognition, image representation with histograms of colors, distance measures, nearest neighbor recognition, performance evaluation.

13/03/2020 -- Image classification: Topics include: data-driven approaches to image classification, K-nearest neighbor classification, linear classification, loss functions, regularization, soft-max classification.

19/03/2020 -- Neural Networks and backpropagation: Discussed topics include: optimization with stochastic gradient descent, image feature embedding, the computational graph, backpropagation.

20/03/2020 -- Neural networks, backpropagation and introduction to Pytorch. Discussed topics include: backpropagation for matrices and tensors, neural networks as features extractors, end-to-end training, hardware for deep learning, introduction to Pytorch.

26/03/2020 -- Pytorch. Discussed topics include: tensors, autograd, nn, optim, new modules, dataloaders, static and dynamic graphs, caffe2 and ONNX.

27/03/2020 -- Convolutional neural networks. Discussed topics include: applications, a bit of history, convolutional layers, pooling, stride, importance of depth, invariances to translation, scale and rotation.

02/04/2020 -- ConvNet visualization and training. Discussed topics include: visualization of ConvNet feature map activations, training of DNNs, with specific reference to activation functions and data preprocessing.

03/04/2020 -- Training Neural Networks. Discussed topics include: weight initialization, batch normalization, neural network optimizers, learning rate decay, model ensembles and dropout.

16/04/2020 -- More on Regularization. In the lecture, there has been further discussion on the regularization techniques of DropOut and BatchNormalization, as well as the introduction of others, including variants of data augmentation, DropConnect, Fractional Max Pool, Stochastic Depth, CutOut and Mixup.

17/04/2020 -- More on training DNN and semantic segmentation. Discussed topics include: hyper-parameter tuning, transfer learning, semantic segmentation, upconvolutions and skip connections.

23/04/2020 -- Semantic segmentation part 2. In the lecture, there has been further discussion on semantic segmentation, upsampling techniques including max unpooling and transpose convolutions, convolution and transpose convolution as matrix multiplication, benchmark metrics and data collection for semantic segmentation.

24/04/2020 -- Object detection. In the lecture there has been discussion on object detection including details on classification and bounding box regression losses, selective search, RCNN, Fast RCNN, RoI Pool and RoI Align, Region Proposal Network, Faster RCNN, two-stage and one-stage detectors.

30/04/2020 -- Object detection part 2. In the lecture there has been discussion on instance segmentation with Mask R-CNN, the COCO benchmark, image captioning and object relations, 3D object detection and representation, visual object search with local representations.

07/05/2020 -- Object search. In the lecture there has been discussion on object search by using local representations with geometric verification, visual words, global representations, VLAD, Fisher-Vector, deep representations, R-MAC descriptor, triplet loss, introduction to person search and re-identification.

08/05/2020 -- Person Re-identification and Search. In the lecture there has been discussion on re-identification techniques including feature embedding and verification models, CMC and mAP metrics, best practices for re-identification with DNNs, person search, online instance matching (OIM) loss.

14/05/2020 -- Person and Semantic Search. In the lecture there has been discussion on person search with OIM, query-guided person search with QEEPS including modelling and performance analysis, Squeeze-and-excitation network, semantic search with a triplet Siamese network and loss.

15/05/2020 -- Semantic Search and Human Pose Estimation (w/ detection). In the lecture there has been discussion on semantic search, visual embeddings, the visual genome dataset, the PCK metric for pose estimation, DeepPose, regression and heatmap-based techniques, pictorial structures and flexible mixture of parts, autocontext and inference machines, convolution pose machines, stacked hourglass networks.

21/05/2020 -- Human Pose Estimation w/o detection. In the lecture there has been discussion on top-down and bottom-up approaches to the joint detection and human pose estimation, including Mask R-CNN and Part Affinity Fields, as well as 3D pose estimation of people and objects.

22/05/2020 -- Sequence modelling and forecasting. In the lecture there has been discussion on applications of sequence modelling and forecasting, and the models for those based on temporal convolution networks and recurrent neural networks.

28/05/2020 -- Sequence modelling and forecasting part 2. In the lecture there has been discussion on RNNs, LSTMs and attention, and their applications to text modelling, image captioning, visual question answering, as well as multi-agent sequence modelling including head pose and the scene context for people trajectory forecasting.

29/05/2020 -- Sequence modelling and forecasting part 3. In the lecture there has been discussion on Transformer Networks, self-attention, positional encoding, attention with relative positions, BERT and its variance, as well as applications, including language models, the Image Transformer and the Music Transformer.

04/06/2020 -- Multi-task and Meta-learning. In the lecture there has been discussion on the motivation, definitions, notations, assumptions and datasets for both technical fields, multi-task approaches to condition on the task and to optimize for the objective, optimization-based and non-parametric meta-learning techniques, including MAML, Matching Networks and Prototypical Networks.