f(AI)nder View
March 2019
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

Company Nodis - partner of the March issue
TruTint - Transparent, color, fast switching smart glass
Nodis' TruTint smart glass technology is transforming windows, giving people the ability to change the tint, color and temperature characteristics of windows instantly. TruTint is the only color smart glass solution.
Buildings consume 40% of a city's electricity and generate 45% of its CO2 emissions. TruTint smart windows can reduce electricity usage significantly (by up to 40%), enable carbon neutral buildings.
Nodis is headquartered in Singapore and funded by Singapore government (National Research Foundation).
Nodis is Winner of Shell IdeaRefinery, Grand Prize Winner of K-Startup Grand Challenge.
AI ecosystem
AI industry analytics, trends, foresights, insights, challenges, regulations, etc.
Latest innovation survey from BCG. Strong innovators are extending their edge over weaker rivals by embracing AI—in their products and services as well as in how they create them. Top innovators are also AI leaders (Google, Amazon, Apple, Microsoft, Netflix, and IBM, for openers) and that many others—are actively leveraging AI.
Read more
Association for Computing Machinery (ACM) named Yoshua Bengio, Geoffrey Hinton, and Yann LeCun recipients of the 2018 ACM A.M. Turing Award for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing. The ACM A.M. Turing Award often is referred to as the "Nobel Prize of Computing" .
Read more
The report from MMC Ventures is based on 400 discussions with ecosystem participants, and aims to go beyond the hype and explain the reality of AI today. Authors provide an accessible introduction to AI and its applications, describe the state of AI adoption, technology and talent in 2019, explore the dynamics of Europe's AI startups.
Read more
The experts are convinced that in time they can build a high-performance quantum computer. In the long term, such machines will very likely shape new computing and business paradigms by solving computational problems that are currently out of reach. Quantum computing has the potential to revolutionize information processing the way quantum science revolutionized physics a century ago.
Read more
The goal of this classification is to bring understanding to the complex market for AI vendor solutions. This is meant to help end users better understand how to evaluate, procure, and implement AI solutions and compare the different offerings at different layers in the stack. Authors looked closely at 3000+ companies that offer AI-specific or AI-enhanced products for sale in the market. The classification sorts the vendor landscape into a "stack" of four layers.
Read more
The study from Microsoft and IDC Asia/Pacific surveyed over 1,600 business leaders and over 1,580 workers across 15 markets in Asia Pacific. While 80% of business leaders polled agreed that AI is instrumental for their organization's competitiveness, only 41% of organizations in the region have embarked on their AI journeys. Those organizations that have adopted AI expect it to increase their competitiveness 100% by 2021.
Read more
Applications
Rather than generate audio sequentially, GANSynth generates an entire sequence in parallel, synthesizing audio significantly faster than real-time on a modern GPU and ~50,000 times faster than a standard WaveNet. GANSynth generates the entire audio clip from a single latent vector, allowing for easier disentanglement of global features such as pitch and timbre.
Read more
DeepMind and Google started applying ML algorithms to 700 megawatts of wind power capacity. Using a neural network trained on widely available weather forecasts and historical turbine data, the DeepMind system was configured to predict wind power output 36 hours ahead of actual generation. The model recommends how to make optimal hourly delivery commitments to the power grid a full day in advance.
Read more
Does A.I. beat doctors?
Brief review of the top A.I. algorithms that were recently applied in healthcare:
spotting DNA mutations in tumors, classifying heart images, diagnosing skin cancer, detecting breast cancer risk, predicting suicide risk, predicting death risk among inpatients, heart attack predicting algorithm and others.
Read more
Researchers are now applying deep generative modeling techniques to the generation and optimization of molecules. The review of the state of the art is based on 45 papers on the subject published in the past two years. The four classes of techniques are described: recursive neural networks, autoencoders, generative adversarial networks, and reinforcement learning.
Read more
A parametric t-SNE approach based on deep feed-forward neural networks was applied to the chemical space visualization problem.
It is able to retain more information than certain dimensionality reduction techniques used for this purpose (principal component analysis (PCA), multidimensional scaling (MDS)). The new method will help to create new chemical compounds and navigate in the space of the existing chemicals.
Read more
The review describes applications of ML methods in particle physics and cosmology, quantum many body physics, quantum computing, and chemical and material physics. Authors highlight research and development into novel computing architectures aimed at accelerating ML.
In each of the sections are described recent successes as well as domain-specific methodology and challenges.
Read more
Introductions
Modern neural networks are often criticized as being a "black box." Understanding what's going on inside neural nets isn't solely a question of scientific curiosity — our lack of understanding handicaps our ability to audit neural networks and, in high stakes contexts, ensure they are safe. Activation atlases are a new way to see some of what goes on inside that box. Activation atlases build on feature visualization, a technique for studying what the hidden layers of neural networks can represent.
Read more
Reinforcement learning problem suffers from serious scaling issues. Hierarchical reinforcement learning (HRL) is a computational approach intended to address these issues by learning to operate on different levels of temporal abstraction. The key innovation of the HRL is to extend the set of available actions so that the agent can now choose to perform not only elementary actions, but also macro-actions, i.e. sequences of lower-level actions.
Read more
3 billion smartphones and 7 billion connected devices are constantly generating new data. Centralized approach of data analysis can be problematic if the data is sensitive or expensive to centralize. TensorFlow Federated (TFF) is an open source framework for experimenting with ML and other computations on decentralized data. It implements an approach called Federated Learning (FL), which enables many participating clients to train shared ML models, while keeping their data locally.
Read more
Snorkel MeTaL is used to construct a simple model (pretrained BERT + linear task heads) and incorporate a variety of supervision signals (traditional supervision, transfer learning, multi-task learning, weak supervision, and ensembling) in a Massive Multi-Task Learning (MMTL) setting, achieving a new state-of-the-art score on the GLUE Benchmark and four of its nine component tasks.
Read more
It has long been speculated that the backpropagation-of-error algorithm (backprop) may be a model of how the brain learns. Backpropagation-through-time (BPTT) is the canonical temporal-analogue to backprop used to assign credit in recurrent neural networks in machine learning, but there's even less conviction about whether BPTT has anything to do with the brain.
Read more
AI systems are vulnerable to adversarial attacks, which limit the applications of AI technologies in key security fields. Improving the robustness of AI systems against adversarial attacks has played an increasingly important role in the further development of AI. The paper comprehensively summarizes the latest research progress on adversarial attack and defense technologies in deep learning.
Read more
Toolbox
In recent years, multiagent settings have become an effective platform for deep reinforcement learning research. There are still some challenges for multiagent reinforcement learning. Current environments are either complex but too narrow or open-ended but too simple.
The game genre of Massively Multiplayer Online Games simulates a large ecosystem of a variable number of players competing in persistent and extensive environments.
Read more
AI Habitat enables training of embodied AI agents (virtual robots) in a highly photorealistic & efficient 3D simulator, before transferring the learned skills to reality. Habitat is a platform for embodied AI research that consists of Habitat-Sim, Habitat-API, and Habitat Challenge.
Habitat-Sim is a flexible, high-performance 3D simulator with configurable agents, multiple sensors, and generic 3D dataset handling.
Read more
Increasing amounts of computational capacity allow researchers to train ever-larger neural networks with new capabilities. TF-Replicator is a software library that helps researchers deploy their TensorFlow models on GPUs and Cloud TPUs with minimal effort and no previous experience with distributed systems. TF-Replicator's programming model has now been open sourced as part of TensorFlow's tf.distribute.Strategy.
Read more
Spatially-adaptive normalization is a simple but effective layer for synthesizing photorealistic images given an input semantic layout. Experiments on several challenging datasets demonstrate the advantage of the proposed method over existing approaches, regarding both visual fidelity and alignment with input layouts. Finally, the model allows user control over both semantic and style as synthesizing images.
Read more
Deep reinforcement learning algorithms require large amounts of experience to learn an individual task. Current methods rely heavily on on-policy experience, limiting their sample efficiency. Authors address these challenges by developing an off-policy meta-RL algorithm that disentangles task inference and control. Proposed method outperforms prior algorithms in sample efficiency by 20-100X as well as in asymptotic performance on several meta-RL benchmarks.
Read more
Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms makes it difficult to search the architectures on large-scale tasks. ProxylessNAS can directly learn the architectures for large-scale target tasks and target hardware platforms and thus can address the high memory consumption issue of differentiable NAS and reduce the computational cost.
Read more
Datasets
GQA: a strong and robust question engine that leverages scene graph structures to create 22M diverse reasoning questions. Accompanying the dataset is a suite of new metrics that evaluate essential qualities such as consistency, grounding and plausibility.
Read more
A large-scale human-annotated collection of one million short videos corresponding to dynamic events unfolding within three seconds, involving people, animals, objects or natural phenomena, that capture the gist of a dynamic scene. Each video is tagged with one action or activity label among 339 different classes.
Read more
The first of its kind available to the global research community. 1-million images of human faces from the publicly availableYFCC-100M Creative Commons dataset. The faces annotated using 10 well-established and independent coding schemes from the scientific literature.
Read more
AI hardware
Industry is producing new types of accelerators for ML, DL, and high-performance computing for various types of hardware - GPUs, FPGAs, ASICs, IPU, NPUs, xPUs, etc. These are generally add-in cards or accelerator modules installed on an existing system. The add-ins come in different shapes, with a range of thermal characteristics, a variety of board wiring schemes, and, in some cases, unique sockets. What's needed is a common form factor so they can be used in the same system to leverage resources.
Read more
An announcement of releasing the first, experimental support for embedded platforms in TensorFlow Lite. This is a prototype of a development board built by SparkFun, and it has a Cortex M4 processor with 384KB of RAM and 1MB of Flash storage. This is running entirely locally on the embedded chip, with no need to have any internet connectivity. The model itself takes up less than 20KB of Flash storage space, the footprint of the TensorFlow Lite code is only another 25KB of Flash, and it only needs 30KB of RAM to operate.
Read more
Graph processing has become an important part of various areas, such as machine learning, computational sciences, medical applications, social network analysis, and many others. Field Programmable Gate Arrays (FPGAs) can be an energy-efficient solution to deliver specialized hardware for graph processing. The first survey and taxonomy on graph computations on FPGAs aims to facilitate understanding of this emerging domain. The survey describes and categorizes existing schemes and explains key ideas.
Read more
Single-board computers (SBCs) are revolutionary devices. The most famous SBC is undoubtedly the Raspberry Pi. The Coral Dev Board is the new kid on the block. It is an SBC with Google's custom Mendel operating system, designed for use with the TensorFlow Lite neural network. Google designed the Coral Dev Board for fast prototyping of machine learning hardware. The first thing which makes it unique is the Edge TPU Module.
Read more
Jetson Nano supports high-resolution sensors, can process many sensors in parallel and can run multiple modern neural networks on each sensor stream. It also supports many popular AI frameworks, making it easy for developers to integrate their preferred models and frameworks into the product. Jetson Nano joins the Jetson™ family lineup, which also includes the powerful Jetson AGX Xavier™ for fully autonomous machines and Jetson TX2 for AI at the edge.
Read more
Quantum annealing is a computing paradigm that has the ambitious goal of efficiently solving large-scale combinatorial optimization problems of practical importance. The article gives a brief introduction to the concept of quantum annealing and also discusses the state of the art and future prospective of superconducting qubits, the paradigm platform for quantum annealing, as well as ideas for alternative platforms based on Rydberg atoms and trapped ions.
Read more
Events
11-12 April, 2019. San Jose, USA.
Explore HPC architectures, brain-inspired computing and quantum approaches to AI processing, alongside industry trends, investment and M&A.
Read more
30 April -1 May, 2019. Dubai, UAE.
80+ participating countries, 100+ enterprises & startups, 39 solution sectors, 9 industry verticals, 130+ leaders on stage, 100+ hours of power talks.
Read more
Deep Learning World
6-7 May, 2019. Munich, Germany.
Premier conference covering the commercial deployment of deep learning. DLW runs parallel to the Predictive Analytics World for Industry 4.0 at the same venue.
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
Twitter
Made on
Tilda