or subscribe with
Join 3,800+ readers for one email each week.
Digests » 106
Customer Data Infrastructure, offering Segment API compatibility, multiple hosting options, fixed infrastructure based pricing & powerful real time transformations.
this week's favorite
People enjoy food photography because they appreciate food. Food culture has been spreading more than ever in the current digital era, with many people sharing pictures of food they are eating across social media. Instagram queries for #food leads to at least 300 M posts.
Learn a valuable representation of time for your Machine Learning Model.
As technology has advanced, the way we accomplish things in our lives has shifted. While new tech is influencing such basic human activities as communication, for instance, encouraging us to reconsider what it means to connect with one another, technological innovations are also changing approaches to building tech itself.
The policy gradient theorem is a foundational result in reinforcement learning. It turns the derivative of the total reward (hard!) into the derivative of the policy (easy!). Unfortunately, the usual proof uses more algebraic manipulation than you'd like in a foundational result. Instead, this proof starts with an unusual view of reinforcement learning problems, a view from which you only need a small amount of maths to get to the theorem.
We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via bipartite matching, and a transformer encoder-decoder architecture.