This is the fifth post (post1, post2, post 3, post 4) in the series that I am writing based on the book First contact with DEEP LEARNING, Practical introduction with Keras. In it I will present an intuitive vision of the main components of the learning process of a neural network and put into practice some of the concepts presented here with an interactive tool called TensorFlow Playground. Remember that a neural network is made up of neurons connected to each other; at the same time, each connection of our neural network is associated with a weight that dictates the importance of this relationship in the neuron when multiplied by the input value. Each neuron has an activation function that defines the output of the neuron. The activation function is used to introduce non-linearity in the modeling capabilities of the network. We have several options for activation functions that we will present in this post. Training our neural network, that is, learning the values of our parameters (weights wij and bj biases) is the most genuine part of Deep Learning and we can see this learning process in a neural network as an iterative process of "going and return" by the layers of neurons. The "going" is a forwardpropagation of the information and the "return" is a backpropagation of the information. The first phase forwardpropagation occurs when the network is exposed to the training data and these cross the entire neural network for their predictions (labels) to be calculated.

How can you empower yourself? Because the skills required for jobs in the AI economy are changing so rapidly, we need to ensure that our systems for preparing, educating, training, and retraining the current and future workforce also evolve. Not only will the new AI economy require new technical skills, but there is a growing recognition that most workers will need to learn new skills throughout their working lives. At Microsoft, we believe everyone should have the resources to learn more and become an expert in AI, and this is why we are collaborating with organisations such as AI Singapore to empower as many persons as we can. In addition to that, we are making a lot of the AI learning assets that Microsoft uses free to use for individuals ranging from students to developers.

In the previous course Introduction to Deep Learning, we saw how to use Neural Networks to model a dataset of many examples. The good news is that the basic architecture of Neural Networks is quite generic whatever the application: a stacking of several perceptrons to compose complex hierarchical models and their optimization using gradient descent and backpropagation. Inspite of this, you have probably heard about Multilayer Perceptrons (MLPs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), LSTM, Auto-Encoders, etc. These deep learning models are different from each other. Each model is known to be particulary performant in some specific tasks, even though, fundamentally, they all share the same basic architecture.

Lungs cancer is one of the most common cancer among men and second most in women. In U.S alone lung cancer is responsible for more then 29 percent of cancer deaths. The general symptoms of lung cancer are coughing in blood, pain in chest and shortness in breath. For the prevention of this cancer and to detect it as early as possible, a team of researches from New York has developed an artificial intelligence tool that analyzes patients' lung tissue. It also specify the cancer types, and even identify the altered genes driving abnormal cell growth.

All the biggest labs leading AI research will have you believe that their fancy game-playing software bots will one day be applicable to the real world. The skills from playing Go, Poker or Dota 2 will be transferable to algorithms designing new drugs, controlling robots, teaching computers how to negotiate – you name it. One startup, Kindred.AI, decided to put some of those claims to the test, particularly with respect to robotics and machine-learning software. "We wanted to know the readiness of the state-of-the-art RL algorithms on real robotic applications," Rupam Mahmood, lead of AI at Kindred, told The Register. Reinforcement learning (RL) is a popular AI method that teaches agents how to perform a specific task by rewarding them every time they get closer to the stated goal.

Microsoft wants the Australian government to close loopholes in proposed data sharing-and-release legislation that it believes could be used to shut down or limit access to data without explanation. The software giant used a submission [pdf] to a Prime Minister & Cabinet-led consultation to outline concerns that data could be too easily withheld or not offered in the first place, despite assertions that "much of the Australian government's data is not personal or sensitive". Microsoft suggested that Australian laws should, in part, mimic the EU's reuse of public sector information directive, which requires agencies to explain why they deny access to data. "We note that the proposed process for sharing data does not appear to require Commonwealth data custodians to provide an explanation either when denying a data access request, or if they decide not to provide open access to data in the first instance," Microsoft said. "[We] suggest that the bill require data custodians to provide such an explanation.

As a tech geek and writer for an IT company, I love having discussions about things like Blockchain and AI -- and their impact on our current and future lives -- with my layman best friend (who works in Public Security). He's still waiting for the day I show up at his house in a fully operating Iron Man-like flight suit. I still believe that one day I will. A while ago, we had a very interesting discussion about AI and intelligent systems in general, which reminded me how important it is to regularly talk to people outside of your field of work, to gain new insights. I realize that my friend was intuitively meshing his definition of'Intelligence' with a vague notion of Free Will, which I find to be a strong and useful intuition. The discussion also led me to want to have a better understanding of the current developments in applied Artificial Intelligence in the world, so as to be able to make a better informed prediction of its future.

We will use the Google AIY Vision Kit to show a real world example of how to setup AI and discuss competing frameworks (CNTK, Tensorflow, Pytorch, etc.) The objective of this session will be to help attendees understand the field of AI, tools of the trade and how to get started as a developer. Please feel free to bring your own Google AIY Vision kit but this is not a requirement since most of the material will be showcasing the technology and not a hands-on lab.

The state-of-the-art performance of deep learning algorithms has led to a considerable increase in the utilization of machine learning in security-sensitive and critical applications. However, it has recently been shown that a small and carefully crafted perturbation in the input space can completely fool a deep model. In this study, we explore the extent to which face recognition systems are vulnerable to geometrically-perturbed adversarial faces. We propose a fast landmark manipulation method for generating adversarial faces, which is approximately 200 times faster than the previous geometric attacks and obtains 99.86% success rate on the state-of-the-art face recognition models. To further force the generated samples to be natural, we introduce a second attack constrained on the semantic structure of the face which has the half speed of the first attack with the success rate of 99.96%. Both attacks are extremely robust against the state-of-the-art defense methods with the success rate of equal or greater than 53.59%. Code is available at https://github.com/alldbi/FLM.

Gradient-based meta-learning techniques are both widely applicable and proficient at solving challenging few-shot learning and fast adaptation problems. However, they have practical difficulties when operating on high-dimensional parameter spaces in extreme low-data regimes. We show that it is possible to bypass these limitations by learning a data-dependent latent generative representation of model parameters, and performing gradient-based meta-learning in this lowdimensional latent space. Our evaluation shows that LEO can achieve state-of-the-art performance on the competitive miniImageNet and tieredImageNet few-shot classification tasks. Further analysis indicates LEO is able to capture uncertainty in the data, and can perform adaptation more effectively by optimizing in latent space. Humans have a remarkable ability to quickly grasp new concepts from a very small number of examples or a limited amount of experience, leveraging prior knowledge and context. Just as humans can efficiently learn new tasks, it is desirable for learning algorithms to quickly adapt to and incorporate new and unseen information. Few-shot learning tasks challenge models to learn a new concept or behaviour with very few examples or limited experience (Fei-Fei et al., 2006; Lake et al., 2011).