Chess is a canonical example of a task that requires rigorous reasoning and long-term planning. Modern decision Transformers - trained similarly to LLMs - are able to learn competent gameplay, but it is unclear to what extent they truly capture the rules of chess. To investigate this, we train a 270M parameter chess Transformer and test it on out-of-distribution scenarios, designed to reveal failures of systematic generalization. Our analysis shows that Transformers exhibit compositional generalization, as evidenced by strong rule extrapolation: they adhere to fundamental syntactic rules of the game by consistently choosing valid moves even in situations very different from the training data. Moreover, they also generate high-quality moves for OOD puzzles. In a more challenging test, we evaluate the models on variants including Chess960 (Fischer Random Chess) - a variant of chess where starting positions of pieces are randomized. We found that while the model exhibits basic strategy adaptation, they are inferior to symbolic AI algorithms that perform explicit search, but gap is smaller when playing against users on Lichess. Moreover, the training dynamics revealed that the model initially learns to move only its own pieces, suggesting an emergent compositional understanding of the game.

There have been key advancements to building universal approximators for multi-goal collections of reinforcement learning value functions -- key elements in estimating long-term returns of states in a parameterized manner. We extend this to hierarchical reinforcement learning, using the options framework, by introducing hierarchical universal value function approximators (H-UVFAs). This allows us to leverage the added benefits of scaling, planning, and generalization expected in temporal abstraction settings. We develop supervised and reinforcement learning methods for learning embeddings of the states, goals, options, and actions in the two hierarchical value functions: $Q(s, g, o; \theta)$ and $Q(s, g, o, a; \theta)$. Finally we demonstrate generalization of the HUVFAs and show they outperform corresponding UVFAs.

Does your PC really need to search for aliens? How about pitching in your resources to help make AI art, instead? A new community effort, Stable Horde, allows you to donate your PC's extra GPU cycles to create AI art and use your donated time to create AI art in just a fraction of the time instead. Stable Horde is a grass-roots effort where you can donate your PC's idle time to help others create fabulous AI art -- or you can use the "horde" of PCs to create your own AI art, too. Stable Horde is similar to both SETI@Home (which went into "hibernation" in 2020) or Folding@Home.

I needed some explosive chemicals from the old sawmill on the edge of town, but hundreds of freakers were gathered there, feasting on a mass grave. So I set explosive traps around the building's edges, planned a course through its twists and turns, then tossed a napalm-filled molotov cocktail into the building. They were on me in an instant -- hundreds of hungry monsters ready to rip me limb from limb. I ran for it, hoping not to blow myself up with my own bombs or become the freakers' next meal. After an hour of fighting, dying, and trying again, I was finally victorious. I was out of ammo, explosives and medical supplies, but the horde was dead.

Another month has come and gone and as we bid July adieu it's important to remember that a new month means new video games. For Xbox Games Pass subscribers ($10 a month here, with a 14-day free trial option) this means seven new games that you can download and play with your monthly subscription. I hesitate to say "free" since you're paying for a sub, but it's still a great deal. Of these, I've only played a couple. I loved Hitman Season 1.

A fresh instalment of the bombastic shooter franchise was teased in 2014 but never came to fruition. Instead, developer Croteam released The Talos Principle, a critically acclaimed puzzler about androids and AI. Now, though, we have a teaser trailer for Serious Sam 4: Planet Badass, which shows the titular hero driving through the (French? Of course, he's soon attacked by a "headless kamikaze," which Sam dispatches with a casual shotgun blast. The camera then pans back to reveal a horde of gruesome enemies and the message: See you at E3 2018.

Abstract: Unsupervised learning can be a challenging task due to the absence of an outcome variable. When on top it concerns outlier detection, the scarcity of this type of observations makes the problem even more challenging. This talk will be about applying the isolation forest algorithm in a financial context in order to detect unusual customer behavior.

As we start looking towards more comprehensive exploration of the Moon and of Mars, the assumption is that we're working on sending humans to the surface of those worlds. It's going to be exponentially more difficult and dangerous than sending robots, but that's what exploration is all about, right? The idea is using robotic telepresence for planetary exploration. From orbit, the authors argue, a small team of humans would remote operate rovers and other robotic systems and as a result they could do more exploration while keeping the overall mission safer and cheaper. We already use telerobotics for planetary exploration--we've got robots all over the solar system sending us data and then patiently doing what we tell them to do.

Last month in Montreal, researchers huddled around a monitor at Maluuba, an artificial intelligence startup Microsoft acquired in January, to learn the answer to a minor mystery of computer science: What happens when you score a million points at classic Atari game Ms. Pac-Man? Such a question might seem to lack a certain urgency, considering the game and its original arcade version were released in 1982. But they would soon get an answer: An inhuman, machine-learning powered player they had built was chomping towards a seven-digit score. The moment proved somewhat anticlimactic. "It just reset to zero, it was kind of disappointing," says Rahul Mehrotra, a program manager at Maluuba, who was part of the small crowd.

According to a trailer released for French-speaking audiences, Overwatch's next big event is headed to consoles and PC tomorrow, April 11th. "Insurrection" sends you and five teammates into the past against hordes of robotic Omnics on the King's Row map. Set as a "declassified" archival mission detailing Tracer's first outing for Overwatch, the update will have more than 100 new character models, emote poses and graffiti tags waiting for you. Be sure and grab this quickly, though, as the event only lasts until May 1st. Character skins from the new "Insurrection" event also leaked onto Xbox Live a few days ago, too, as fans search for more details to whet their appetite for the new update.