f(AI)nder View
October 2018
f(AI)nder View - a monthly digest in the field of
Artificial Intelligence, Machine Learning, Deep Learning
exploring the various levels and faces of AI: from basic to sophisticated ones,
from algorithms and technologies to business applications

AI ecosystem
AI industry analytics, trends, foresights, insights, challenges, regulations, etc.
Blockchain, quantum computing, augmented analytics and artificial intelligence will drive disruption and new business models.
"The future will be characterized by smart devices delivering increasingly insightful digital services everywhere. We call this the intelligent digital mesh"
.

Read more
Research on China's artificial intelligence industry in the new ranking from China Money Network, the leading source of intelligence and data on China's venture capital and technology sector.
The 14 unicorns, all included in the China AI Top 50 ranking, are worth a combined US$40.5 billion.

Read more
At the heart of Japanese 'Society 5.0' future model, is the concept that innovation and the wellbeing of its population need to work in tandem.
Society 5.0 will target Japan's economic and social challenges, such as: its ageing population, labour shortages and weak growth.

Read more
Enterprise companies comprise a $3.4 trillion market worldwide of which an increasingly larger share is being allocated to artificial intelligence technologies. "Enterprise" technology companies create tools for workplace roles and functions that a large number of businesses use.
Read more
"Building a rocket is hard. If AI is a rocket, then we will all have tickets on board some day".
DeepMind Safety Research discusses three areas of technical AI safety:
specification, robustness, and assurance.

Read more
AI and blockchain are philosophically opposed in many ways.
Could blockchain help create better AI?
Are there ways AI can help blockchain?
A view at the topic from the perspective of a VC investor.

Read more
Applications
As artificial intelligence matures so does its potential in the creative industries — one of which happens to be music production.
AI experts generally agree that music composition software won't replace humans, but it is already changing the process of music creation and will have an even greater impact in the near future
.

Read more
Christie's becomes the first auction house to offer a work of art created by an algorithm. Portrait of Edmond Belamy was created by GAN (Generative Adversarial Network).
"The algorithm is composed of two parts. On one side is the Generator, on the other the Discriminator. We fed the system with a data set of 15,000 portraits painted between the 14th century to the 20th".

Read more
Generating realistic images based on descriptions takes years of graphic design training. In machine learning this is a generative task, which is also much more challenging than discriminative tasks. Generative models (with some control) can be extremely useful in many cases:
Content creation, Content-aware smart editing, Data augmentation
.

Read more
How to estimate the age of your brain with MRI data and Artificial Intelligence. How to create state-of-the-art algorithms, from entry-level linear regression to advanced deep neural networks that automatically rediscover known features of brain aging, such as cortical atrophy and leukoariosis.
A 30 to 60 minute crash course to understand AI in practice and its pitfalls.

Read more
Alzheimer's is the most common form of dementia. With an aging global population, the prevalence of Alzheimer's disease is rapidly increasing, creating a heavy burden on public healthcare systems.
Artificial Intelligence algorithm can accurately predict whether a person's cognitive decline will lead to Alzheimer's disease in the next five years or not
.

Read more
Since the medical field of radiology mostly relies on extracting useful information from images, it is a very natural application area for deep learning.
It is presented a survey of the research
in deep learning applied to radiology,
and opportunities and challenges for incorporating deep learning in the radiology practice of the future.

Read more
How Artificial Intelligence creates $1 trillion of change in the front, middle and back office of the financial services industry.
In US alone, 2.5 million financial services employees are exposed to AI technologies. Potential cost exposure of $490 billion in front office (distribution), $350 billion in middle office, $200 billion in back office (manufacturing)
.

Read more
AI, IoT and 5G collectively are transforming the energy sector in fundamental ways.
By 2020, the industrial IoT is expected to comprise more than a trillion sensors.
A McKinsey study projects that AI innovations could save oil and gas companies as much as $50 billion in production costs annually.

Read more
The most popular applications of AI in agriculture appear to fall into three major categories: Agricultural Robots, Crop and Soil Monitoring, Predictive Analytics.
The amount of data that can potentially be captured by technologies such as drones, and satellites on a daily basis will give agricultural business a new ability to predict changes and identify opportunities.

Read more
Introductions
Cheatsheets from Stanford's CS 229 Machine Learning course: Supervised Learning (linear models, generative learning, support vector machines and kernel methods), Unsupervised Learning (clustering methods and dimensionality reduction), Deep Learning (neural networks, backpropagation and reinforcement learning), ML tips and tricks.
Read more
The post tries to condense ~15 years' worth of work into eight milestones that are the most relevant today, including: Neural language models, Multi-task learning, Word embeddings, Neural networks for NLP, Sequence-to-sequence models, Attention, Memory-based networks, Pretrained language models, Adversarial learning, Reinforcement learning.
Read more
A guide to the ML Engineering Loop,
where ML Engineers iteratively
1. Analyze 2. Select an approach
3. Implement
4. Measure
to rapidly and efficiently discover the best models and adapt to the unknown.
The ML Engineering Loop will help you make methodical progress toward a better model despite the inherent uncertainty of the task.

Read more
Many real world problems can be formulated as the interaction between an agent and an environment. The agent interacts with the environment performing actions trying to achieve a certain goal. Every action might cause a change in the environment state. We need to introduce some kind of feedback, going from the environment to the agent. In Reinforcement Learning this takes the form of a scalar signal, called reward.
Read more
A significant number of the photos shared on Facebook and Instagram contain text in various forms. It might be overlaid on an image in a meme, or inlaid in a photo of a storefront, street sign, or restaurant menu. Facebook built and deployed a large-scale machine learning system named Rosetta, that extracts text from more than a billion public Facebook and Instagram images and video frames (in a wide variety of languages), daily and in real time, and inputs it into a text recognition model.
Read more
Recurrent Neural Networks suffer from short-term memory. During back propagation, recurrent neural networks suffer from the vanishing gradient problem. LSTM 's and GRU's were created as the solution to short-term memory. They have internal mechanisms called gates that can regulate the flow of information.
LSTM's and GRU's can be found in speech recognition, speech synthesis, and text generation.
Read more
An embedding is a mapping of a discrete (categorical) variable to a vector of continuous numbers. Neural network embeddings are useful because they can reduce the dimensionality of categorical variables and meaningfully represent categories in the transformed space.
What neural network embeddings are,
why we want to use them, and
how they are learned ?

Read more
Ensemble models in machine learning combine the decisions from multiple models to improve the overall performance network. Basic and Advanced (Stacking, Blending, Bagging, Boosting) Ensemble Techniques, Algorithms based on Bagging (Bagging meta-estimator, Random Forest) and Boosting (AdaBoost, Gradient Boosting, XGBoost, Light GBM, CatBoost).
Read more
Mathematical Optimization is the branch of mathematics that aims to solve the problem of finding the elements that maximize or minimize a given real-valued function. Introduction to optimization alrorithms: gradient descent, Newton's method, quasi-Newton methods, Stochastic Gradient Method.
Read more
Toolbox
A new library of useful building blocks for writing reinforcement learning (RL) agents in TensorFlow.
The TRFL library includes functions to implement both classical RL algorithms as well as more cutting-edge techniques.
The library was created by the Research Engineering team at DeepMind
.

Read more
For building graph networks in Tensorflow and Sonnet. The library will work with both the CPU and GPU version of TensorFlow. The library includes demos which show how to create, manipulate, and train graph networks to reason about graph-structured data, on a shortest path-finding task, a sorting task, and a physical prediction task.
Read more
Google researchers present a deep bidirectional Transformer model that redefines the state of the art for 11 natural language processing tasks, even surpassing human performance in the challenging area of question answering. BERT is a huge model, with 24 Transformer blocks, 1024 hidden layers, and 340M parameters.
Read more
Fast.ai is releasing v1 of a new free open source library for deep learning, called fastai. The library sits on top of PyTorch v1, and provides a single consistent API to the most important deep learning applications and data types for vision, text, tabular data, time series, and collaborative filtering.
Read more
Imitation learning provides an appealing framework for autonomous control: in many tasks, demonstrations of preferred behavior can be readily obtained from human experts. In contrast, classical model-based reinforcement learning (MBRL) offers considerably more flexibility: a model learned from data can be reused at test-time to achieve a wide variety of goals.
Read more
A tool for neural network visualizations and art. Networks are differentiable with respect to their inputs: we can slightly tweak the image to better fit the desired properties, and then iteratively apply such tweaks in gradient descent. Differentiable image parameterizations invite us to ask "what kind of image generation process can we backpropagate through?"
Read more
AI hardware
The core advantage of the TPU is its MXU — a systolic array matrix multiplication unit.
A GPU is a vector machine. The data of a neural network is arranged in a matrix.
So, we'll build a matrix machine.
The way to achieve that matrix performance is through a piece of architecture called a systolic array.

Read more
Why building your own Deep Learning Computer is 10x cheaper than AWS? How can I make my computer even cheaper? Cloud GPU machines are expensive at $3 / hour and you have to pay even when you're not using the machine. Building an expandable Deep Learning Computer
with 1 top-end GPU only costs $3k.
You'll break even in just a few months.

Read more
Canadian startup D-Wave Systems Inc. launched a real-time online quantum computing environment called Leap.
Leap is the latest addition to the quantum cloud—services that virtualize quantum computing for almost anyone with a computer and a broadband connection to use.

Read more
Most AI chips for use in low-power or battery-operated IoT devices have a neural network that has been trained by a more powerful computer to do a particular job. Startup Eta Compute showed off the first commercial low-power (at 100-microwatt scale) AI chip capable of learning on its own using a type of machine learning called spiking neural networks.
Read more
The Internet giants' race to create custom AI chips has so far produced Google's powerful TPUs and Microsoft's FPGAs. Earlier this year, Alibaba and Baidu announced their respective development plans for AI chips Ali-NPU and Kunlun. Huawei jumped on the bandwagon with the unveiling of Ascend 910 and Ascend 310, two 7nm-based AI chip IPs that run on the cloud for training and inferencing.
Read more
What is the latency or energy cost for an inference made by a Deep Neural Network (DNN)? Is it possible to predict this latency or energy consumption before a model is trained? If yes, how can machine learners take advantage of these models to design the hardware-optimal DNN for deployment? One cannot optimize what isn't properly modeled.
Read more
AI competitions
Find ships on satellite images as quickly as possible. Shipping traffic is growing fast. More ships increase the chances of infractions at sea. Airbus offers comprehensive maritime monitoring services by building a meaningful solution.
November 7, 2018 - Entry deadline.
You must accept the competition rules before this date in order to compete.

Read more
The objective of this contest is to promote and encourage innovation in an open environment amongst best-in-class Universities and students and to support young innovators in taking their solution forward. The contest is organized by Atos International SAS.
The 3 finalists will be invited in Paris for the Awards Ceremony at Jul 2019.
Idea Submission - November 30, 2018.
Read more
A global competition challenging teams to develop and demonstrate how humans can collaborate with powerful AI technologies to tackle the world's Grand Challenges.
The AI XPRIZE currently has 62 competitors from 15 countries, each harnessing AI to solve a grand challenge.
Registration to join the competition is currently open for the Second Wildcard Round until December 14th, 2018.
Read more
Events
9-11 November, 2018. San Jose, USA.
Topics: video understanding, NLP, robots, drones, deep learning breakthrough, edge computing, AI in healthcare, AI in finance,
AI in enterprises, games, IoT.

Read more
22 November, 2018. Moscow, Russia.
How will business benefit from AI ?
International exhibition-conference devoted to application of artificial intelligence in business.

Read more
5-7 December, 2018. Berlin, Germany.
Comprehensive insight into the principles of Machine Learning and introduction to the world of ML tools, programming languages and technologies.
Read more
Follow us
f(AI)nder View - a monthly digest in the field of
Artificial Intelligence, Machine Learning, Deep Learning
exploring the various levels and faces of AI:
from basic to sophisticated ones,
from algorithms and technologies to business applications

f(AI)nder View is prepared by f(AI)nder Lab
Made on
Tilda