|September 8 · Issue #56 · View online |
Hey ho everyone,
we are back this week with an oven-fresh issue.
Happy reading and hacking,
As always if you like receiving this newsletter, you can help us by sharing it with your friends and colleagues.
| Meet Michelangelo: Uber's Machine Learning Platform |
Uber Engineering showcases some of the fruits of their self-driving car research; Michelangelo their machine learning-as-a-service system that enables teams to easily build, deploy, and operate ML solutions.
| NIPS 2017 Conference Registration still open |
Registration for NIPS is still open, but will soon sell out.
| Facebook and Microsoft introduce new open ecosystem for interchangeable AI frameworks |
Facebook and Microsoft introduced Open Neural Network Exchange (ONNX) format, a standard for representing deep learning models that enables models to be transferred between frameworks. A much needed common language in current AI platform and framework zoo. ONNX is the first step toward an open ecosystem where AI developers can easily move between state-of-the-art tools and choose the combination that is best for them.
| Computer Vision in Retail |
Technology from Startup Standard Cognition
is powering autonomous checkout at a prototype store in Santa Clara, the system correctly identifies an impressive array of products a prospective customer takes from the shelf 98% of them time. With Amazon Go testing a similar system autonomous checkout could be with us sooner rather than later.
| New AI Camera System Developed For Remote, Outdoors Applications |
Most AI camera systems require a dry, dust-free operating environment and a high speed local network. A new device called DNNCam™ is different. With on-board processing and data storage, DNNCam™ can function as a stand-alone AI system. With a dustproof and waterproof case, DNNCam™ operates in nearly any environment. Learn more here
| Compressing Deep Neural Nets |
Great post outlining in-depth how to reduce the size of an existing network by compressing it by removing connections between neurons that don’t really add much to the final result.
The example in this post makes MobileNet-224 25% smaller reducing its parameters from 4 million to 3 million.
| Understanding Attentive Recurrent Comparators – Sanyam Agarwal – Medium |
A very clear explanation of the recent paper Attentive Recurrent Comparator
, an important paper in the important effort to advance data-efficient machine learning.
| Deep Meaning Beyond Thought Vectors |
A great survey of recent work in natural language understanding ranging from attention over Graph-LSTMs to neural symbolic machines.
| Neural Language Modeling From Scratch (Part 1) |
This post lays out the foundations of natural language modeling building up from the fundamentals of word embeddings to using RNNs to improve performance and handling bias through drop out and weight tying. For a deeper dive every section includes references to the relevant papers.
| GitHub - Bidirectional LSTM-CRF for Sequence Labeling |
anago performs named-entity recognition (NER), part-of-speech tagging (POS tagging), semantic role labeling (SRL) for a range of languages.
| GitHub - onnx/onnx: Open Neural Network Exchange |
The code for Facebook’s and Microsoft’s ONNX initiative.
| Automated Crowdturfing Attacks and Defenses in Online Review Systems |
In this paper, the authors identify a new class of attacks that leverage deep learning language models (Recurrent Neural Networks or RNNs) to automate the generation of fake online reviews for products and services. Not only are these attacks cheap and therefore more scalable, but they can control rate of content output to eliminate the signature burstiness that makes crowdsourced campaigns easy to detect.
| Fast Image Processing with Fully-Convolutional Networks |
The authors present a fascinating new approach to accelerating a wide variety of image processing operators using a fully-convolutional network that is trained on input-output pairs that demonstrate the operator’s action.
| Developing Bug-Free Machine Learning Systems With Formal Mathematics |
Great paper in which the authors outline an approach to use an interactive proof assistant to arrive at a formal theory for proving the correctness of a machine learning system, thereby tackling the many factors which make ensuring correctness notoriously difficult such as noisy data, non-convex objectives, model misspecification, and numerical instability.