Since our bio-inspired machine learning technology "Dynamic Boltzmann Machine (DyBM)" debuted in the fall of 2015, we received many comments on the music demo and human evolution image that we used to show how an artificial neural network learns about different topics in different formats. Many developers expressed interest in using the code to let DyBM learn other music or animation. This request for more "openness" made us wonder if we should dramatically change DyBM's bio-oriented design. It learns patterns like neurons: at each moment of a song or an image, DyBM adjusts its internal parameters. The more data fed into DyBM, the better it will master what it's trying to understand.

So we've talked a lot in this series about how computers fetch and display data, but how do they make decisions on this data? From spam filters and self-driving cars, to cutting edge medical diagnosis and real-time language translation, there has been an increasing need for our computers to learn from data and apply that knowledge to make predictions and decisions. This is the heart of machine learning which sits inside the more ambitious goal of artificial intelligence. We may be a long way from self-aware computers that think just like us, but with advancements in deep learning and artificial neural networks our computers are becoming more powerful than ever. Want to know more about Carrie Anne? https://about.me/carrieannephilbin

Alphabet's DeepMind has been making incredible strides in the field of artificial intelligence (AI). Their AI can create pictures based on sentences, play StarCraft, and explore strange environments. It has also developed memory and is imagining solutions to problems. AlphaGo, an AI, was created by DeepMind in order to conquer the oldest game in the world: Go; an incredibly popular game known for being even more complex than chess. What better game to test an AI on?

In this article, I clarify the various roles of the data scientist, and how data science compares and overlaps with related fields such as machine learning, deep learning, AI, statistics, IoT, operations research, and applied mathematics. As data science is a broad discipline, I start by describing the different types of data scientists that one may encounter in any business setting: you might even discover that you are a data scientist yourself, without knowing it. As in any scientific discipline, data scientists may borrow techniques from related disciplines, though we have developed our own arsenal, especially techniques and algorithms to handle very large unstructured data sets in automated ways, even without human interactions, to perform transactions in real-time or to make predictions. To get started and gain some historical perspective, you can read my article about 9 types of data scientists, published in 2014, or my article where I compare data science with 16 analytic disciplines, also published in 2014. I also wrote about the ABCD's of business processes optimization where D stands for data science, C for computer science, B for business science, and A for analytics science.

I hope that you would now be able to apply pre-trained models to your problem statements. Be sure that the pre-trained model you have selected has been trained on a similar data set as the one that you wish to use it on. There are various architectures people have tried on different types of data sets and I strongly encourage you to go through these architectures and apply them on your own problem statements. Please feel free to discuss your doubts and concerns in the comments section.

Reinforcement Learning (RL) can model complex behavior policies for goal-directed sequential decision making tasks. A hallmark of RL algorithms is Temporal Difference (TD) learning: value function for the current state is moved towards a bootstrapped target that is estimated using next state's value function. $\lambda$-returns generalize beyond 1-step returns and strike a balance between Monte Carlo and TD learning methods. While lambda-returns have been extensively studied in RL, they haven't been explored a lot in Deep RL. This paper's first contribution is an exhaustive benchmarking of lambda-returns. Although mathematically tractable, the use of exponentially decaying weighting of n-step returns based targets in lambda-returns is a rather ad-hoc design choice. Our second major contribution is that we propose a generalization of lambda-returns called Confidence-based Autodidactic Returns (CAR), wherein the RL agent learns the weighting of the n-step returns in an end-to-end manner. This allows the agent to learn to decide how much it wants to weigh the n-step returns based targets. In contrast, lambda-returns restrict RL agents to use an exponentially decaying weighting scheme. Autodidactic returns can be used for improving any RL algorithm which uses TD learning. We empirically demonstrate that using sophisticated weighted mixtures of multi-step returns (like CAR and lambda-returns) considerably outperforms the use of n-step returns. We perform our experiments on the Asynchronous Advantage Actor Critic (A3C) algorithm in the Atari 2600 domain.

We study the relationship between geometry and capacity measures for deep neural networks from an invariance viewpoint. We introduce a new notion of capacity --- the Fisher-Rao norm --- that possesses desirable invariance properties and is motivated by Information Geometry. We discover an analytical characterization of the new capacity measure, through which we establish norm-comparison inequalities and further show that the new measure serves as an umbrella for several existing norm-based complexity measures. We discuss upper bounds on the generalization error induced by the proposed measure. Extensive numerical experiments on CIFAR-10 support our theoretical findings. Our theoretical analysis rests on a key structural lemma about partial derivatives of multi-layer rectifier networks.

We present MILABOT: a deep reinforcement learning chatbot developed by the Montreal Institute for Learning Algorithms (MILA) for the Amazon Alexa Prize competition. MILABOT is capable of conversing with humans on popular small talk topics through both speech and text. The system consists of an ensemble of natural language generation and retrieval models, including template-based models, bag-of-words models, sequence-to-sequence neural network and latent variable neural network models. By applying reinforcement learning to crowdsourced data and real-world user interactions, the system has been trained to select an appropriate response from the models in its ensemble. The system has been evaluated through A/B testing with real-world users, where it performed significantly better than many competing systems. Due to its machine learning architecture, the system is likely to improve with additional data.

Google's DeepMind Artificial Intelligence AlphaGo Zero recently attained an important milestone--the Artificial Intelligence (AI) taught itself how to play the strategy game Go without any human interaction and was able to beat the world's best Go players. The ability to reach this level of performance with human input is a significant step forward in the maturation of AI. Over the past several years, AI has made significant progress in a wide variety of areas such as image and speech recognition, drug discovery, and algorithmic trading. In most of these cases, the AI relies on vast existing data sets and some degree of human engagement. A long-standing ambition of AI researchers has been to create algorithms that do not rely on already existing data sets nor the need for human input.

Google has recently open-sourced its framework for machine learning and neural networks called Tensorflow. In this session, we will teach you how to choose the right neural network for your problem and how to make it behave. Familiarity with differential equations is no longer required. Instead, a couple of lines ofTensorflow Python, and a bag of "tricks of the trade" will do the job. No previous Python knowledge required.