Goto

Collaborating Authors

 jumanji


Beyond the Boundaries of Proximal Policy Optimization

Tan, Charlie B., Toledo, Edan, Ellis, Benjamin, Foerster, Jakob N., Huszár, Ferenc

arXiv.org Artificial Intelligence

Proximal policy optimization (PPO) is a widely-used algorithm for on-policy reinforcement learning. This work offers an alternative perspective of PPO, in which it is decomposed into the inner-loop estimation of update vectors, and the outer-loop application of updates using gradient ascent with unity learning rate. Using this insight we propose outer proximal policy optimization (outer-PPO); a framework wherein these update vectors are applied using an arbitrary gradient-based optimizer. The decoupling of update estimation and update application enabled by outer-PPO highlights several implicit design choices in PPO that we challenge through empirical investigation. In particular we consider non-unity learning rates and momentum applied to the outer loop, and a momentum-bias applied to the inner estimation loop. Methods are evaluated against an aggressively tuned PPO baseline on Brax, Jumanji and MinAtar environments; non-unity learning rates and momentum both achieve statistically significant improvement on Brax and Jumanji, given the same hyperparameter tuning budget.


Mava: a research library for distributed multi-agent reinforcement learning in JAX

de Kock, Ruan, Mahjoub, Omayma, Abramowitz, Sasha, Khlifi, Wiem, Tilbury, Callum Rhys, Formanek, Claude, Smit, Andries, Pretorius, Arnu

arXiv.org Artificial Intelligence

Multi-agent reinforcement learning (MARL) research is inherently computationally expensive and it is often difficult to obtain a sufficient number of experiment samples to test hypotheses and make robust statistical claims. Furthermore, MARL algorithms are typically complex in their design and can be tricky to implement correctly. These aspects of MARL present a difficult challenge when it comes to creating useful software for advanced research. Our criteria for such software is that it should be simple enough to use to implement new ideas quickly, while at the same time be scalable and fast enough to test those ideas in a reasonable amount of time. In this preliminary technical report, we introduce Mava, a research library for MARL written purely in JAX, that aims to fulfill these criteria. We discuss the design and core features of Mava, and demonstrate its use and performance across a variety of environments. In particular, we show Mava's substantial speed advantage, with improvements of 10-100x compared to other popular MARL frameworks, while maintaining strong performance. This allows for researchers to test ideas in a few minutes instead of several hours. Finally, Mava forms part of an ecosystem of libraries that seamlessly integrate with each other to help facilitate advanced research in MARL. We hope Mava will benefit the community and help drive scientifically sound and statistically robust research in the field. The open-source repository for Mava is available at https://github.com/instadeepai/Mava.


Jumanji: a Diverse Suite of Scalable Reinforcement Learning Environments in JAX

Bonnet, Clément, Luo, Daniel, Byrne, Donal, Surana, Shikha, Coyette, Vincent, Duckworth, Paul, Midgley, Laurence I., Kalloniatis, Tristan, Abramowitz, Sasha, Waters, Cemlyn N., Smit, Andries P., Grinsztajn, Nathan, Sob, Ulrich A. Mbou, Mahjoub, Omayma, Tegegn, Elshadai, Mimouni, Mohamed A., Boige, Raphael, de Kock, Ruan, Furelos-Blanco, Daniel, Le, Victor, Pretorius, Arnu, Laterre, Alexandre

arXiv.org Artificial Intelligence

Open-source reinforcement learning (RL) environments have played a crucial role in driving progress in the development of AI algorithms. In modern RL research, there is a need for simulated environments that are performant, scalable, and modular to enable their utilization in a wider range of potential real-world applications. Therefore, we present Jumanji, a suite of diverse RL environments specifically designed to be fast, flexible, and scalable. Jumanji provides a suite of environments focusing on combinatorial problems frequently encountered in industry, as well as challenging general decision-making tasks. By leveraging the efficiency of JAX and hardware accelerators like GPUs and TPUs, Jumanji enables rapid iteration of research ideas and large-scale experimentation, ultimately empowering more capable agents. Unlike existing RL environment suites, Jumanji is highly customizable, allowing users to tailor the initial state distribution and problem complexity to their needs. Furthermore, we provide actor-critic baselines for each environment, accompanied by preliminary findings on scaling and generalization scenarios. Jumanji aims to set a new standard for speed, adaptability, and scalability of RL environments.


From Tron to Jumanji: the greatest ever movies about video games

The Guardian

When most people think of movies about games, they immediately recall the astonishing light-cycle chase in Steven Lisberger's visually daring film. With its underlying themes of dehumanisation and corporate greed in the digital era, Tron was more than an action romp with pretty effects and a cool arcade setting; a fact underlined by a committed lead performance from Jeff Bridges. In this comedy action movie, Guy (Ryan Reynolds) discovers that not only is he living in a video game – one that bears a not insignificant resemblance to the lawless cities of Grand Theft Auto Online – but he's not even a hero: he's a non-player character. This realisation sets off a chain of high-octane events inside and outside the game, as a couple of its developers try to rescue it from a money-grubbing, unethical studio head. This beloved Edgar Wright comic-book action comedy is saturated in video game sound, imagery and style, and structured rather like a Street Fighter campaign, as Scott (Michael Cera) faces off against the seven evil exes of his girlfriend Ramona Flowers (Mary Elizabeth Winstead).


Video lame: has Hollywood's warped relationship with gaming gone too far?

The Guardian

As recent efforts – Tomb Raider, Assassin's Creed, Warcraft – continue to show, video games rarely make great movies. Dwayne Johnson's new epic Rampage might change all this, just as giant, genetically modified wolves might fly, but the source material was hardly that compelling to start with, partly because it was already a mish-mash of movie tropes. In the original Rampage arcade game, you could be King Kong, Godzilla or a werewolf and you basically had to re-enact a city-trashing scene out of a monster movie. Now, see the movie of the game of the movie! To turn it around, however, games already have taken over the movies.


'Jumanji' holds off 'The Post' to stay on top as newcomers fail to ignite holiday weekend box office

Los Angeles Times

A comedic adventure flick about being teleported into a video game outplayed a political drama over the four-day MLK weekend as "Jumanji: Welcome to the Jungle" landed at the top of the domestic box office. The Sony movie outpaced "The Post," about the Washington Post's 1971 decision to publish the Pentagon Papers. In its fourth week of release, "Jumanji: Welcome to the Jungle" added an estimated $35.4 million, for a cumulative $291.6 million, according to numbers from measurement firm comScore. It's on track to become one of Sony's top five domestic releases of all time. "The film shows the strong drawing power of Dwayne Johnson and Kevin Hart in this reboot of the original'Jumanji' that opened way back in December of 1995," said comScore senior media analyst Paul Dergarabedian.


'Jumanji: Welcome To The Jungle' Review Roundup: Sequel Serves Up Mixed Opinions

International Business Times

"Jumanji: Welcome to the Jungle," the sequel to the 1995 film starring the late Robin Williams, is now out in theaters and critics have a lot to say about it. Dwayne "The Rock" Johnson, Kevin Hart, Jack Black and Karen Gillan teamed up for the new flick, which now features the Jumanji board game as a video game instead, but while some reviewers think the star power is just the icing on the cake, others don't even think the big names save the movie. Here's what critics are saying about "Jumanji: Welcome to the Jungle." "'Jumanji: Welcome to the Jungle,' a sequel no one asked for to the 1995 Robin Williams original as well as the animated TV series, has enough star power and comic zest to deliver a fun time at the movies. OK, just barely, but we'll take what we can get. "This updated'Jumanji' can't outrun the clichés on its tail, but its puppy-eagerness to please is hard to resist." "Although the action sequences are a bit rote and it always feels like the movie could go further with its satire of video games, the film succeeds thanks to its outstanding cast who have excellent comedic chemistry.


John Hughes meets Indiana Jones in reimagined 'Jumanji: Welcome to the Jungle'

Los Angeles Times

Don't consider "Jumanji: Welcome to the Jungle" a straight remake of the 1995 Robin Williams board game adventure film. It's more of a spiritual sequel or reimagining, loosely based on the same Chris Van Allsburg book. If it was a jungle in there for Williams, it's a jungle out there for Dwayne Johnson and pals in this video game-inspired romp directed by Jake Kasdan. No longer is Jumanji a simple board game where a roll of the dice can unleash a supernatural jungle explosion in the living room. This time around, "Jumanji" is an old video game console and cartridge dusted off by a mismatched group of high-schoolers stuck with detention one afternoon. The breakfast club fires it up, selecting their avatars: neurotic nerd Spencer (Alex Wolff) chooses Dr. Smolder Bravestone; hulking jock Fridge (Ser'Darius Blain) picks zoologist Moose Finbar; weirdo smartypants Martha (Morgan Turner) is Ruby Roundhouse; while the selfie-obsessed Bethany (Madison Iseman) goes for the "curvy cartographer" Professor Shelly Oberon.


Is 'Jumanji 2' A Video Game? Board Game Eliminated From Remake

International Business Times

"Jumanji: Welcome to the Jungle" trailer made its debut Thursday without the board game in sight. Instead, the makers of the film decided to give the original a facelift, swapping out the outdated board game concept for a digitized video game. As a result, fans of the 1995 film may not be satisfied with this modernized transformation. Eliminating the board game came naturally to the makers behind the reboot. During a CinemaCon interview in March, the film's Director Jake Kasdan -- "Bad Teacher" -- spoke with Forbes about what led to the use of video game tropes instead of the outdated board game concept.


'Jumanji' trailer turns Dwayne Johnson, Kevin Hart into video game avatars

Los Angeles Times

Today in Entertainment: President Trump mocks'Morning Joe' hosts; new'Jumanji' trailer released MSNBC calls out President Trump for his disparaging comments on TV hosts'Jumanji' trailer turns Dwayne Johnson, Kevin Hart into video game avatars Motion picture academy invites 774, the largest class ever Johnny Galecki's ranch home burns in San Luis Obispo's Hill fire Booted from KROQ, Rodney Bingenheimer will return to radio via Sirius XM'Jumanji' trailer turns Dwayne Johnson, Kevin Hart into video game avatars'Jumanji' trailer turns Dwayne Johnson, Kevin Hart into video game avatars Welcome to the jungle, kids. Four detention-bound teenagers are thrust into a mysterious game in the first "Jumanji: Welcome to the Jungle" trailer, literally forced into the bodies of their avatars -- played by Dwayne Johnson, Kevin Hart, Karen Gillan and Jack Black. "I think we got sucked into Jumanji and we've become the avatars that we chose," Johnson's geeky Spencer observes after they land in the jungle with new identities. Instead of a board game, however, the Sony-Columbia reboot features Jumanji as a retro video game console. Judging from the trailer, unlike its 1995 predecessor, the film appears to lean quite heavily on comedy in addition to action and adventure.