|August 16 · Issue #92 · View online |
Hey and welcome to another week in deep learning!
As always, we hope you’ll enjoy reading as much as we did and would appreciate you sharing this newsletter with friends and colleagues.
See you next week!
| Machine Learning Can Identify the Authors of Anonymous Code |
While getting identified by your voice, handwriting or fingerprint seems logical, being identified by your coding style is new and pretty impressive.
| Facebook's Yann LeCun explains why it lets researchers split their time with academia |
Here, Yann LeCun, head of AI at Facebook, explains the reasoning behind their unusual recruitment model, how it’s inspired by law and medicine and how good it actually works.
| AutoKeras: The Killer of Google’s AutoML |
Comparing the recently released ‘AutoKeras’ package to the commercial AutoML solution currently being offered by Google, George Seif
explains his reasoning on why he thinks AutoKeras is the superior solution for automated network architecture search.
| Academic Torrents |
Although not new, we recently stumbled across this scalable, secure, and fault-tolerant repository for data, which makes accessing large datasets like ImageNet or the Functional Map of the World much easier.
| Now anyone can train Imagenet in 18 minutes |
A team of researchers has managed to build a system, that’s capable of training an image classifier to 93% accuracy using ‘only’ 16 public AWS
cloud instances, each with 8 NVIDIA V100
GPUs which translates to costs of roughly 40$.
| How GOAT Taught a Machine to Love Sneakers |
An interesting approach, using a variational auto encoder to explore the latent parameter space of sneakers. Some interesting models in there!
| Making Neural Nets Work With Low Precision |
This great introduction to low precision or quantized inference and how it’s implemented in TensorFlow Lite. If you need to squeeze down your models, this is worth a look.
| Troubleshooting Convolutional Neural Nets |
An extensive collection of tips to help you, when you’re stuck during building or training your neural network.
| Transfer Learning in NLP |
This article gives an introduction to transfer learning in NLP and tries to use it for Amazon review sentiment analysis. Well written and comes with a notebook containing code for all the experiments.
| Interactive Machine Learning List |
As we love interactive visualizations and machine learning, this list was a great thing to find. You should definitely check it out and add anything you’re missing!
| GraphPipe - Dead Simple ML Model Serving via a Standard Protocol |
Made by Oracle, GraphPipe is a protocol and collection of software designed to simplify machine learning model deployment and decouple it from framework-specific model implementations.
| Getting Alexa to Respond to Sign Language Using Your Webcam and TensorFlow.js |
With accessibility in mind, Abhishek Singh
decided to make Alexa usable via sign language as well. He describes how he tackled the problem, which technologies he used and what results his work yielded.
| Exploring the Limits of Weakly Supervised Pretraining |
In this paper, the authors present a unique study of transfer learning with large convolutional networks trained to predict hashtags on billions of social media images.
| Helix: Accelerating Human-in-the-loop Machine Learning |
This paper proposes Helix, a declarative machine learning system that accelerates iterative development by optimizing workflow execution end-to-end and across iterations.