Sometimes I learn a piece of new information that makes me say, "That's enough. I am done learning for today. I don't want to read or think anymore."

Oklahoma governor Frank Keating's brother Mark wrote a novel about a man named Tom McVey blowing up a building in Oklahoma City, who was later apprehended because he committed a minor traffic violation... years before that very scenario played out with Tim McVeigh
One of the first victims of the anthrax attacks in 2001 was tabloid photographer Robert Stevens. One of his editors was Michael Irish. Irish's wife, Gloria, rented apartments over the summer of 2001 to suspected hijackers Marwan Al-Shehhi and Hamza Alghamdi.
The photographer who shot the iconic/haunting "Falling Man" photograph on 9/11 (supposedly with a digital camera that couldn't have possibly done the job) was also present when Bobby Kennedy was assassinated in 1968.

https://t.co/1s8F0yKBK9
Anyway these were three recent ones that just sorta made me wish I never learned how to read
I'll just keep going.

In 1997, FBI special agent Dwayne Fusilier's son helped make a Columbine school video that eerily mimicked the shooting that was to take place two years later. Despite this, Agent Fusilier did not recuse himself from investigating the case.

More from All

How can we use language supervision to learn better visual representations for robotics?

Introducing Voltron: Language-Driven Representation Learning for Robotics!

Paper: https://t.co/gIsRPtSjKz
Models: https://t.co/NOB3cpATYG
Evaluation: https://t.co/aOzQu95J8z

🧵👇(1 / 12)


Videos of humans performing everyday tasks (Something-Something-v2, Ego4D) offer a rich and diverse resource for learning representations for robotic manipulation.

Yet, an underused part of these datasets are the rich, natural language annotations accompanying each video. (2/12)

The Voltron framework offers a simple way to use language supervision to shape representation learning, building off of prior work in representations for robotics like MVP (
https://t.co/Pb0mk9hb4i) and R3M (https://t.co/o2Fkc3fP0e).

The secret is *balance* (3/12)

Starting with a masked autoencoder over frames from these video clips, make a choice:

1) Condition on language and improve our ability to reconstruct the scene.

2) Generate language given the visual representation and improve our ability to describe what's happening. (4/12)

By trading off *conditioning* and *generation* we show that we can learn 1) better representations than prior methods, and 2) explicitly shape the balance of low and high-level features captured.

Why is the ability to shape this balance important? (5/12)

You May Also Like

Still wondering about this 🤔


save as q