# Practical Applications of Deep Learning *Talk by* Daniel Micallef --- ## What is Deep Learning  --- ### Deep Learning vs Machine Learning - Artficial Neural Networks and Deep Learning - Manual vs Automatic extraction of features - DL performance keeps improving with scaling amounts of data --- ## Problems with Deep Learning - Needs massive amounts of data - Collection of Data - Processing of Data - GPUs are Expensive - Training Time  --- ## What NOT to do with Deep Learning - DL is not good for everything! Vincent Warmerdam explains: https://www.youtube.com/watch?v=68ABAU_V8qI --- ## Algorithms and their Applications *NB: Bare with me!* --- Multilayer Perceptron Nerual Network (with Backpropogation) Good for: - Image verification and reconstruction - Speech recognition - Machine translation - Data classification Note: A feed-forward supervised learning algorithm with up to two hidden layers to generate a set of outputs from a given set of inputs. As the name suggests, it is composed of more than one perceptron. Backpropogation: foundation of neural network training. The supervised learning algorithm computes a gradient descent with the weights updated backward — from output toward input — or backpropagation --- Convolutional Neural Networks  --- CNNs are good for: - Image classification and Object Detection - Video Recognition - Natural Language processing tasks - Pattern recognition - Recommendation engines --- Recurrent Neural Network (RNN) - Recognizes sequential attributes  --- RNN normally good for: - Sentiment classification - Image captioning - Speech recognition - Natural Language processing - Machine Translation - Prediction (e.g. in Search) Note: Designed to recognize a data set's *sequential* attribute and use patterns to predict the next likely scenario How it works: an RNN the hidden layer preserves sequential information from previous steps. This means the output from an earlier step is fed as the input to a current step, --- Long Short-Term Memory (LSTM) A variant of the RNN. Good for: - Captioning of images and videos - Language translation and modeling - Sentiment Analysis - Stock market predictions --- Generative Adversarial Network (GAN) - Unsupervised Learning - Discovers and learns regularities and patterns Good for: - Health diagnostics - Mail spam detection - Natural Language Processing - Speech Processing Note: Network which automatically discovers and learns regularities and patterns in input data GAN compromised of 2 adversial nets --- Deep Belief Network - An unsupervised probabilistic deep-learning algorithm - Generative Learning Model DBNs are useful for: - Image and face recognition - Video-sequence recognition - Motion-capture data - Classifying high-resolution satellite image data --- ## Real life Applications of DL *What did the big guys do with DL?* --- 1. Search: Google Photos https://photos.google.com/ --- 2. Website Auto Translation: Google Translate, Facebook comment translations   --- 3. Automatic Reply: Gmail, Linkedln Chat, Facebook Messenger  --- 5. Spam messages - Gmail blocking 100m spam messages every day --- 6. Self-driving cars: Tesla Motors and Nissan  --- 7. Speech Recognition: Amazon echo, Google Assistant, Cortana, Siri --- ## What about us What input can we give to society using Deep learning? Note: What can we do with DL? Brainstorm any ideas. --- ## Demo Note: Brain MRIs. Tumour vs Healthy MRI scans --- ## Frameworks - Tensorflow and Keras - Pytorch --- Resources: [1] https://www.simplilearn.com/deep-learning-algorithms-article [2] https://www.kaggle.com/navoneel/brain-mri-images-for-brain-tumor-detection