Artificiality Bites đź’Š Issue #15
Hello Human! This is a new issue from my weekly newsletter, holding a tiny compilation made of interesting articles from last week, projects, tutorials and tools; all related to Data, Artificial Intelligence and adjacent topics. Bon appetit!
đź“ť Interesting articles this week
- The State of AI Ethics October 2020
158p
Montreal AI Ethics Institute released its second State of AI Ethics report, covering the most important developments this past quarter, in order to help researchers and practitioners considering the societal impacts of AI-enabled solutions. - Rediscovering Semi-Supervised Learning
9'
This blog post tries to demonstrate how semi-supervised learning improves model performance, especially when using small samples of labeled data. - Estimating the Impact of Training Data with Reinforcement Learning
9'
In this article Google AI address the challenge of quantifying the value of training data in order to dismiss low quality or incorrectly labeled data, by using a novel approach based on meta-learning. - DeepMind Introduces Algorithms for Causal Reasoning in Probability Trees
3'
Deepmind presented new algorithms for causal reasoning in discrete probability trees, covering the entire causal hierarchy (association, intervention, and counterfactuals), and operating on arbitrary propositional and causal events.
đź’ˇ Projects / Apps / Papers
- Deepnote
The Jupyter-compatible platform with real time collaboration in the cloud is now open to public. - Deepform
DeepForm aims to extract information from political TV ad receipts by using Machine Learning. The project consists of a baseline ML model, training data set, and public benchmark where anyone can submit their solution for extracting data from forms. Read more here. - Lobe
Microsoft has released a desktop application called Lobe, which automatically trains a custom machine learning model from some given samples and can be shipped in your own app. - Wordtune
An AI writing companion that promises to help you say what you mean through clear, compelling, and engaging writing. - O’Reilly Answers
O’Reilly released a new feature on its learning platform that uses an advanced NLP engine which instantly scans thousands of O’Reilly publications, in order to find the snippet that would answer our question. Read more details at O’Reilly. - Customizing Triggers with Concealed Data Poisoning
Berkeley researchers expose a vulnerability in NLP models by which an adversary can insert concealed poisoned examples causing targeted errors for inputs containing a trigger phrase. - Creativity x Machine Learning
A vast collection of Machine Learning experiments curated by Emil Wallner.
🔧 Tutorials
- How to Build a World-Class ML Model for Melanoma Detection đź“ą
55'
First episode of the new Grandmaster Series by Nvidia. In this video you'll learn how the Kaggle Grandmasters of NVIDIA (KGMON) team built the winning ML model for the SIIM-ISIC Melanoma Classification Kaggle competition. - An introduction to transfer learning in NLP and HuggingFace đź“ą
68'
Machine Learning Tokyo invited Thomas Wolf, Co-founder at HuggingFace, for a talk on the recent breakthroughs in NLP and the correspondent HuggingFace libraries and models. - Building a Graph Visualization Tool
7'
How to build a graph visualization tool and some of the challenges that come with it. - Building an AI 8-Ball with RoBERTa
8'
Can an Artificial Neural Network answer yes/no questions?
📦 Repositories
- lucidrains/lambda-networks
Python implementation of LambdaNetworks, a new approach to image recognition that reaches SOTA with less compute. - dataframehq/whale
Whale is a tool which enables automatic data discovery for data warehouses. - alibaba/EasyTransfer
EasyTransfer is designed to make the development of transfer learning in NLP applications easier. Used in several Alibaba projects since 2017 and recently open-sourced. - pytorch/pytorch
PyTorch 1.7 was released, with CUDA 11 support, new APIs for FFTs, Windows support for distributed training and more.
🎓 Courses / Presentations
- MIT 6.S897 Machine Learning for Healthcare (MIT)
Massachusetts Institute of Technology has made available for everyone its introductory course to Machine Learning in healthcare, taught by David Sontag and Peter Szolovits in Spring 2019. - I Like Notebooks đź“ą
52'
Jeremy Howard did a presentation about how to use Jupyter Notebooks as a literate and exploratory programming environment, using nbdev.
🚀 Extra bits
👉 Newsletter en Español
đź‘‹ See you next week!