|
|
|
An Epidemic of ML Misinformation
|
Covering the news and developments in AI is hard. People are quick to trumpet technological advances as the next great discovery, and slow to understand the limitations that may prevent that from being true. Check out our Spotlight where Gary Marcus discusses the epidemic.
Sobering up AI hype in the ML industry is sorely needed, and this article brings up very good points about the misaligned incentives behind suggesting that an ML algorithm's impact is more remarkable than reality. However, Marcus's unique brand of AI criticism deserves a healthy dose of skepticism in turn. From his "AI can't fix fake news" op-ed last year to his Rebooting AI book, Marcus leaves industry experts with the impression that unless a Machine Learning algorithm generates results that are bulletproof (e.g. detect all fake news with perfect F1, or leave no room for a Turning tester to doubt they are talking to a machine) as opposed to results that are merely statistically significant, then it's not a worthwhile result. The truth likely lies somewhere between the poles of AI hype and Marcus's hyper-cynicism.
|
|
|
|
|
AI Skills
|
|
Out of Fold Predictions
|
Jason Brownlee PhD has written an excellent primer on the steps for evaluating the effectiveness of your ML model. His tutorial walks you through the process and intuition of K-Fold cross-validation. Read more
|
Making Recommendations with Graphs
|
Uber AI applies the latest in graph based embedding technology to transform how Uber Eats matches users, restaurants and dishes. This wave of algorithms signals a new era in recommendation technology making traditional matrix factorization a thing of ML history.. Read more
|
|
|
Tools & Libraries
|
|
Ray for the Curious
|
Are you curious how you might be able to scale out your Python applications without having to refactor every line? Ray is a library designed take advantage of distributed compute, and in some cases only requires you to add decorators to your functions. Read more
|
Safety Gym
|
Reinforcement learning may require safety boundaries in specific applications like self-driving vehicles. OpenAI released a suite of tools to help. Read more
|
Private Federated Learning
|
Just because your data is private does not mean you can't use it to train models along with your competitors. Read up on the intuition behind using sensitive data to produce the most powerful models. Read more
|
|
|
Solutions & Products
|
|
IBM and a Department Store Build a Recommender
|
IBM partnered with a large department store to test out the validity of a new recommendation system. This blog shows the features considered and reasoning behind the ultimate recommender chosen. Read more
|
Capital One AI Chief sees path to Explainable AI
|
Capital One AI Chief Nitzan Mekel-Bobrov speaks on the opacity of deep learning, and how traditional statistics is not inherently more interpretable. Read more
|
|
|
Research
|
|
Deep Double Descent: A Game Changer in How We Understand How Deep Learning Works
|
Open AI covers a detailed investigation into ongoing research on the "Double Descent" phenomenon that Deep Learing experts have observed in the way deep neural networks learn. This paper offers new empirical insight into the way that DL training differs from traditional ML training and the trade off between data, capacity and time across a number of standard DL architectures. A must read for any modern ML professional. Read more
|
Using WaveNet Technology to Reunite Speech-Impaired Users with their Original Voices
|
Heres a demonstration of an early proof of concept of how text-to-speech technologies can synthesize a high-quality, natural sounding voice using minimal recorded speech data. Read more
|
Artificial Intelligence Index - 2019 Annual Report
|
Stanford published its comprehensive annual report of what happened in the world of AI in 2019. It covers everything from the trends in demographics of researchers to the surging interest in conferences focused on AI. Read more
|
|
|
|
See all upcoming community events here.
|
January 15, 9:30–10:30 am | Online meetup
|
Submissions due January 15
|
January 16, 6–9 pm | In-person | San Francisco, CA
|
February 12, 6–9:30 pm | In-person | San Francisco, CA
|
May 4 | In-person | San Francisco, CA
|