4681
Gradient Descent Models Are Kernel Machines
This paper shows that models which result from gradient descent training (e.g., deep neural nets) can be expressed as a weighted sum of simi...
New
The latest blog posts to take off on Hacker News. 6,335 posts.
4682
Using computer vision to help win $1M in Mountain Dew's Super Bowl contest
SUMMARY Mountain Dew's Super Bowl LV ad challenged viewers to count the exact number of bottles in the commercial for a $1 million prize, and this post builds an object detection model to assist with that count. The 30-second video was sampled at three frames per second in Roboflow to produce 92 annotated training…
4683
4684
Science fiction hasn’t prepared us to imagine machine learning
It resembles the Library of Babel more than HAL.
4685
A better way to build ML: Why you should be using Active Learning
Data labeling is often the biggest bottleneck in machine learning. Active learning lets you train machine learning models with much less labeled data. The best AI-driven companies, like Tesla, use active learning.
4686
Reinforcement Learning at Facebook
If you ever wanted to learn about machine learning you could do worse than have Jason Gauci teach you. Jason has worked on YouTube recommendations. He was an early contributor to TensorFlow the open-source machine learning platform. His thesis work was cited by DeepMind. But what I find so fascinating with Jason is he…
4687
Baking with machine learning (2020)
Like many people, I’ve been entertaining myself at home by baking a ton. I’ve gotten pretty good at following recipes, but I decided I wanted to take things one step further and understand the science behind what differentiates a cake from a bread or a cookie. I also like machine learning so I thought: what if I could…
4688
Machine learning saves us $1.7M a year on document previews
Recently, we translated the predictive power of machine learning (ML) into $1.7 million a year in infrastructure cost savings by optimizing how Dropbox generates and caches document previews. Machine learning at Dropbox already powers commonly-used features such as search , file and folder suggestions , and OCR in…
4689
4690
Replicating GPT-2 at Home
How I (almost) replicated OpenAI’s GPT-2 (124M version)
4691
4692
Decentralised content moderation
With Donald Trump suspended from Twitter and Facebook, and Parler kicked off AWS, there is renewed discussion about what sort of speech is acceptable online, and how it should be enforced. Let me say up front that I believe that these bans were justified. However, they do raise questions that need to be discussed…
4693
Machine Learning: The Great Stagnation
The bureaucrats are running the asylum
4694
4695
The neural network of the Stockfish chess engine
Last time, we briefly mentioned the high-level differences between Stockfish and Leela Chess. To recap, Stockfish evaluates about 100 million positions per second using rudimentary heuristics, wher…
4696
Using GPT-3 for plain language incident root cause from logs
Skylar Advisor combines real-time observability data, rich operational context, and customer-owned knowledge to deliver explainable AI guidance for IT operations.
4697
Research team demonstrates world’s fastest optical neuromorphic processor
An international team of researchers has demonstrated the world’s fastest and most powerful optical neuromorphic processor for artificial intelligence, which is capable of processing ultra-large scale data.
4698
Decoding the Peloton
From Voltages to "Cadence, Output, and Resistance" TL;DR - I’ve decoded (most of) the protocol that the Peloton bike uses to communicate with its head unit tablet and built a device,...
4699
Clip: Connecting Text and Images
We’re introducing a neural network called CLIP which efficiently learns visual concepts from natural language supervision. CLIP can be applied to any visual classification benchmark by simply providing the names of the visual categories to be recognized, similar to the “zero-shot” capabilities of GPT-2 and GPT-3.
4700
DALL·E: Creating Images from Text
We’ve trained a neural network called DALL·E that creates images from text captions for a wide range of concepts expressible in natural language.
4701
Why I'm Lukewarm on Graph Neural Networks
{{ content|truncate(160) }}
4702
How Ray Kurzweil’s 1998 predictions about 2019 are faring
This is the fourth…and LAST…entry in my series of blog posts analyzing the accuracy of Ray Kurzweil’s predictions about what things would be like in 2019. These predictions come from hi…
4703
Introduction to Reinforcement Learning (2015)
Build AI responsibly to benefit humanity
4704
Does GPT-2 Know Your Phone Number?
The BAIR Blog
4705
4706
MuZero: Mastering Go, chess, shogi and Atari without rules
In 2016, we introduced AlphaGo, the first artificial intelligence (AI) program to defeat humans at the ancient game of Go. Two years later, its successor - AlphaZero - learned from scratch to master Go, chess and shogi. Now, in a paper in the journal Nature, we describe MuZero, a significant step forward in the…
4707
Interactive C++ for Data Science
Interactive C++ for Data Science In our previous blog post “Interactive C++ with Cling” we mentioned that exploratory programming is an effective way to reduce the complexity of the problem. This post will discuss some applications of Cling developed to support data science researchers.
4708
Building My Own Chess Engine
Exploring the computational complexity of chess. Code snippets in Python so you can do the same.
4709
Netflix's Metaflow: Reproducible machine learning pipelines
A complete guide to building end-to-end machine learning pipelines, from training to deployment, with Metaflow and Cortex
4710
Building a simple neural net in Java
Building a simple neural net in Java.