Tero Keski-Valkama

The main problem with classic RL is that it tries to do credit assignment to actions externally, by returned rewards and associated heuristics.

The credit is way more complex than this, and this trivializes it beyond reasonable limits.

To improve this a bit you'll patch it up by adding a critic. This critic now has the task of trying to assign the credit to actions. It's a complex task, and it turns out it makes sense to share the backbones between the actor and the critic, so the critic can also learn from whatever the actor learns.

Still, the actor and the critic are almost blind; they only try to learn the reward dynamics and only the correlates of those rewards from the input images or other observations. It's not much different from not having observations at all, just learning action sequences blindly.

To improve world dynamics learning you add a yet another hack, World Models, to do autoregressive prediction of observations. Of course to the same shared backbone.

This is starting to look like a proper cognitive architecture! But that's just a mirage. All the models share the same backbone.

The complexity comes just from the misguided heuristic structure of returned rewards, and this comes from misguided economic theories.

Enter Decision Transformers!

Instead of Bellman equation, it just returns the whole outcome to the start of the sequence, to condition it.

The reward dynamics become completely data defined and we're back to a unified model!

#LLMs #AI #Transformers

Blender Dumbass ( J.Y.Amihud )

From: blenderdumbass . org

I decided to re-watch the first Transformers film today, but differently. Instead of indulging in it as mindless entertainment, I decided I want to actually view it seriously, as I would with somebody like Steven Spielberg, who, by the way, was an executive producer on the film.

Read: blenderdumbass.org/reviews/tra

#transformers #MichaelBay #film #review #movies #cinemastodon #ShiaLabeouf #MeganFox #VFX #CGI

Transformers is Good if You Take it Seriously

I decided to re-watch the first Transformers film…

blenderdumbass . org
Blender Dumbass ( J.Y.Amihud )

From: blenderdumbass . org

So here is a movie from Paramount Pictures that was executive produced by Steven Spielberg with Stanley Tucci playing a villain and Mark Wahlberg playing a dad of a teenage girl. The movie touches on very hard emotional topics of sexual abuse and age discrimination. And Peter Jackson does a very good job... Oh... wait... it's not The ...

Read: blenderdumbass.org/reviews/tra

#transformers #MichaelBay #film #review #movies #cinemastodon #MarkWahlberg #StanleyTucci #VFX #CGI

Transformers 4 is a ballsy brave film

So here is a movie from Paramount Pictures that was…

blenderdumbass . org
Blender Dumbass ( J.Y.Amihud )

From: blenderdumbass . org

So here is a movie from Paramount Pictures that was executive produced by Steven Spielberg with Stanley Tucci playing a villain and Mark Wahlberg playing a dad of a teenage girl. The movie touches on very hard emotional topics of sexual abuse and age discrimination. And Peter Jackson does a very good job... Oh... wait... it's not The ...

Read: blenderdumbass.org/reviews/tra

#transformers #MichaelBay #film #review #movies #cinemastodon #MarkWahlberg #StanleyTucci #VFX #CGI

Transformers 4 is a ballsy brave film

So here is a movie from Paramount Pictures that was…

blenderdumbass . org
Blender Dumbass ( J.Y.Amihud )

#MichaelBay s #Transformers 4 had #MarkWalhberg playing a dad of teenage girl, while #StanleyTucci is also in the film. Hmm....

The Romeo and Juliet stuff... They simply had to...

JdeBP

@cstross

It is a Decepticon, that misunderstood the explanation of the Golden Arches.

#FastFood #Transformers

Habr

[Перевод] Соревнование VN1: чему я научился у прогнозистов

С сентября по октябрь 2024 года мне выпала честь организовать VN1 Forecasting Competition — соревнование по прогнозированию, в котором приняли участие около 250 команд и отдельных участников. Целью конкурса было предсказать объём продаж различных онлайн‑продавцов на 13 недель вперёд. Первый был разогревочным: участники могли протестировать разные модели, увидеть свои результаты и сравнить их с успехами соперников. Второй этап был решающим — участникам разрешалось подать только один окончательный прогноз, а его точность оставалась неизвестной до самого финала. После соревнования я собрал мнения и выводы 20 сильнейших участников: мы обсудили их подходы, трудности, с которыми они столкнулись, и найденные решения. В этой статье я поделюсь основными уроками, которые вынес из этого соревнования.

habr.com/ru/companies/bothub/a

#data_science #vn1 #прогнозирование #lgbm #transformers #theta

Соревнование VN1: чему я научился у прогнозистов

С сентября по октябрь 2024 года мне выпала честь организовать…

Хабр
strangefour

The Rescue Bots Academy kids want shiny new gadgets like the big bots have, so they get Hoist to invent some — but his prototypes need a lot of work! Then, the mayor has decided to carve the Rescue Bots’ faces into a mountain as a tourist attraction, and Hot Shot thinks he should be up there too! Can the team save Griffin Rock from his hubris? Join us this week for the Rescue Bots Academy episodes “If At First…” and “Mount Botmore”!
iaconunderground.net/podcast/s
#podcast #Transformers #RescueBots

Stasis Pod Episode #41: If At First…/Mount Botmore

The Academy kids want shiny new gadgets like the big…

Iacon Underground Radio Network
Blender Dumbass ( J.Y.Amihud )

From: blenderdumbass . org

There is this satisfaction when it comes to animation, that I know from personal experience where the longer you work on a shot, or a sequence, the more details there is in the animation, the more satisfying it is to later just watch this insane level of effort simply wash over you one frame after another.

Read: blenderdumbass.org/reviews/tra

#transformers #MichaelBay #film #review #movies #cinemastodon #kingArthur

Transformers 5 is Too Good to be Good

There is this satisfaction when it comes to animation,…

blenderdumbass . org