Retrieval-augmented generation (RAG) systems face a fundamental challenge in aligning independently developed retrievers and large language models (LLMs). Existing approaches typically involve modifying either component or introducing simple intermediate modules, resulting in practical limitations and sub-optimal performance. Inspired by human search behavior -- typically involving a back-and-forth process of proposing search queries and reviewing documents, we propose C-3PO, a proxy-centric framework that facilitates communication between retrievers and LLMs through a lightweight multi-agent system. Our framework implements three specialized agents that collaboratively optimize the entire RAG pipeline without altering the retriever and LLMs. These agents work together to assess the need for retrieval, generate effective queries, and select information suitable for the LLMs. To enable effective multi-agent coordination, we develop a tree-structured rollout approach for reward credit assignment in reinforcement learning. Extensive experiments in both in-domain and out-of-distribution scenarios demonstrate that C-3PO significantly enhances RAG performance while maintaining plug-and-play flexibility and superior generalization capabilities.

One of the most appealing aspects of games set in the Star Wars universe is that you get to concoct scenes and stories we would never see in the movies. Whether you're playing Knights of the Old Republic, Jedi: Fallen Order or the old Star Wars role-playing board game designed by Greg Costikyan in the 1990s, there will be individual moments unrepeatable on the big screen. I know this, because I just won a round of the new trading card game Star Wars Unlimited thanks to a heroic C-3PO wielding Luke Skywalker's lightsaber. On a basic level, Star Wars Unlimited works like most modern trading card games, such as Yu-Gi-Oh! You and an opponent each have a deck of cards, most of which feature a single character or vehicle, with a number for health and another number for power/damage.

We consider a certifiable object pose estimation problem, where -- given a partial point cloud of an object -- the goal is to not only estimate the object pose, but also to provide a certificate of correctness for the resulting estimate. Our first contribution is a general theory of certification for end-to-end perception models. In particular, we introduce the notion of $\zeta$-correctness, which bounds the distance between an estimate and the ground truth. We show that $\zeta$-correctness can be assessed by implementing two certificates: (i) a certificate of observable correctness, that asserts if the model output is consistent with the input data and prior information, (ii) a certificate of non-degeneracy, that asserts whether the input data is sufficient to compute a unique estimate. Our second contribution is to apply this theory and design a new learning-based certifiable pose estimator. We propose C-3PO, a semantic-keypoint-based pose estimation model, augmented with the two certificates, to solve the certifiable pose estimation problem. C-3PO also includes a keypoint corrector, implemented as a differentiable optimization layer, that can correct large detection errors (e.g. due to the sim-to-real gap). Our third contribution is a novel self-supervised training approach that uses our certificate of observable correctness to provide the supervisory signal to C-3PO during training. In it, the model trains only on the observably correct input-output pairs, in each training iteration. As training progresses, we see that the observably correct input-output pairs grow, eventually reaching near 100% in many cases. Our experiments show that (i) standard semantic-keypoint-based methods outperform more recent alternatives, (ii) C-3PO further improves performance and significantly outperforms all the baselines, and (iii) C-3PO's certificates are able to discern correct pose estimates.

Change detection (CD) aims to identify changes that occur in an image pair taken different times. Prior methods devise specific networks from scratch to predict change masks in pixel-level, and struggle with general segmentation problems. In this paper, we propose a new paradigm that reduces CD to semantic segmentation which means tailoring an existing and powerful semantic segmentation network to solve CD. This new paradigm conveniently enjoys the mainstream semantic segmentation techniques to deal with general segmentation problems in CD. Hence we can concentrate on studying how to detect changes. We propose a novel and importance insight that different change types exist in CD and they should be learned separately. Based on it, we devise a module named MTF to extract the change information and fuse temporal features. MTF enjoys high interpretability and reveals the essential characteristic of CD. And most segmentation networks can be adapted to solve the CD problems with our MTF module. Finally, we propose C-3PO, a network to detect changes at pixel-level. C-3PO achieves state-of-the-art performance without bells and whistles. It is simple but effective and can be considered as a new baseline in this field. Our code is at https://github.com/DoctorKey/C-3PO.

It's Monday morning of some week in 2050 and you're shuffling into your kitchen, drawn by the smell of fresh coffee C-3PO has brewed while he unloaded the dishwasher. "Here you go, Han Solo, I used the new flavor you bought yesterday," C-3PO tells you as he hands you the cup. C-3PO arrived barely a month ago and already has developed a wonderful sense of humor and even some snark. He isn't the real C-3PO, of course--you just named him that because you are a vintage movie buff--but he's the latest NeuroCyber model that comes closest to how people think, talk, and acquire knowledge. He's no match to the original C-3PO's fluency in 6 million forms of communication, but he's got full linguistic mastery and can learn from humans like humans do--from observation and imitation, whether it's using sarcasm or sticking dishes into slots. Unlike the early models of such assistants like Siri or Alexa who could recognize commands and act upon them, NeuroCybers can evolve into intuitive assistants and companions.

Anthony Daniels didn't want to meet a relatively unknown American movie director looking for someone to play a robot in a "low-budget, science fiction film." He just wasn't a fan of the genre, but his agent persisted, telling the aspiring actor "you never know what it could lead to." It's a funny anecdote when you consider that the director was George Lucas, the sci-fi flick was Star Wars: A New Hope and the part Daniels was auditioning for was a "nervous, persnickety and uptight" human-cyborg relations protocol droid named C-3PO. More than 40 years later, Daniels is the only actor to have appeared in all nine Star Wars movies -- from 1977's A New Hope to last year's The Rise of Skywalker, released on DVD last month. Now 74, he chronicles his journey, from classically trained actor and mime in London to one of the most beloved characters in the history of filmmaking (alongside his wing man, R2-D2) in a new memoir, I Am C-3PO: The Inside Story. The story about not wanting to audition is only one of the surprises that Daniels shares. Lucas actually tested 30 other actors to give voice to C-3PO after filming was complete, including actor Richard Dreyfuss, before being convinced by a voiceover pro that Daniel's take of the droid worked best. And he re-creates (in our video interview) some of his favorite lines, calling out the scene in The Rise of Skywalker when he's about to get his memory wiped. "I also felt that this was the last movie and I was saying goodbye and taking one last look at the fans around the world, the people who have been part of the whole thing," he says.

Roundup Let's get cracking with some machine-learning news. Starksy Robotics is no more: Self-driving truck startup Starsky Robotics has shut down after running out of money and failing to raise more funds. CEO Stefan Seltz-Axmacher bid a touching farewell to his upstart, founded in 2016, in a Medium post this month. He was upfront and honest about why Starsky failed: "Supervised machine learning doesn't live up to the hype," he declared. Neural networks only learn to pick up on certain patterns after they are faced with millions of training examples.

A common refrain in the media is that people don't like their boss and people are scared of robots. So I wondered about the truth and nuance to these emotions: how many people would prefer a robot to their boss? The old saying goes, "People join a company, but they leave a bad boss." As Gallup research demonstrates, 70% of how we feel about work--our emotional commitment--is driven by who our manager is. The ongoing employee engagement crisis is largely about managers who know how to manage tasks, but don't know how to lead people. And there is also growing press coverage about automation and the emergence of robots in the workplace.

We may not be close to getting artificial intelligence like that of C-3PO. But AI is setting the scene for a revolution in software development. For decades, software development has been done manually. From punching cards in FORTRAN to writing distributed systems in Go, the discipline has remained fundamentally the same: think deeply about a problem, come up with a clever approach (i.e., algorithm) and give the machine a set of instructions to execute. This method, which could be called "explicit programming," has been integral to everything from the mainframe to the smartphone, from the internet boom to the mobile revolution.

C-3PO and R2-D2 are an odd couple in the Star Wars universe. C-3PO is a cowardly droid who obeys pre-defined protocols and routine tasks, while R2-D2 is a curious and adventurous robot who learns from previous problems, uses logical thinking and larger concepts to solve new problems. But together they do things they could not do alone. Similarly, RPA (Robotic Process Automation) and Advanced Analytics are an odd but very complementary combination of new business technologies. Like the diligent but unimaginative C-3PO, RPA follows precise rules to execute repetitive business processes; and like the curious and adaptable R2-D2, Advanced Analytics learns to make complex judgments when faced with new situations.