A primary challenge for visual-based Reinforcement Learning (RL) is to generalize effectively across unseen environments. Although previous studies have explored different auxiliary tasks to enhance generalization, few adopt image reconstruction due to concerns about exacerbating overfitting to task-irrelevant features during training. Perceiving the pre-eminence of image reconstruction in representation learning, we propose SMG (\blue{S}eparated \blue{M}odels for \blue{G}eneralization), a novel approach that exploits image reconstruction for generalization. SMG introduces two model branches to extract task-relevant and task-irrelevant representations separately from visual observations via cooperatively reconstruction. Built upon this architecture, we further emphasize the importance of task-relevant features for generalization.

Graph class incremental learning (GCIL) requires the model to classify emerging nodes of new classes while remembering old classes. Existing methods are designed to preserve effective information of old models or graph data to alleviate forgetting, but there is no clear theoretical understanding of what matters in information preservation. In this paper, we consider that present practice suffers from high semantic and structural shifts assessed by two devised shift metrics. We provide insights into information preservation in GCIL and find that maintaining graph information can preserve information of old models in theory to calibrate node semantic and graph structure shifts. We correspond graph information into low-frequency local-global information and high-frequency information in spatial domain.

We consider the problem of reconfiguring a two-dimensional connected grid arrangement of passive building blocks from a start configuration to a goal configuration, using a single active robot that can move on the tiles, remove individual tiles from a given location and physically move them to a new position by walking on the remaining configuration. The objective is to determine a reconfiguration schedule that minimizes the overall makespan, while ensuring that the tile configuration remains connected. We provide both negative and positive results. (1) We present a generalized version of the problem, parameterized by weighted costs for moving with or without tiles, and show that this is NP-complete. (2) We give a polynomial-time constant-factor approximation algorithm for the case of disjoint start and target bounding boxes. In addition, our approach yields optimal carry distance for 2-scaled instances.

Matter is finally taking off. The new smart standard has long promised to make it easier for smart home manufacturers to build devices that work across smart home ecosystems, while simultaneously making it easier for customers (like you) to find devices that work with their other devices, and are easy to set up. With Matter, compatible smart home devices will work with Google Assistant, Apple HomeKit, and yes, Alexa. Amazon recently switched on Matter support for millions of Echo devices, as well as the ability for many of those Echo devices to work as Thread border routers--basically meaning that they'll integrate with a modern smart home's mesh network. With all this commotion around Matter and Alexa, you might be wondering how to connect Matter devices to your own Alexa-based smart home.

Break out the champagne and roll out the red carpets, CES is back! After two rough, COVID-addled years that saw the world's greatest tech show reduced to a shell of its former self, the show is primed to spring back to its former glory for 2023. And our team of writers and editors will be on the ground in Las Vegas, bringing it all to you. But much has changed since the last "normal" CES of 2020. The economy has boomed and busted, supply chains have knotted, and attitudes over excess have shifted as climate change looms larger and larger in our global conversation.

Consumers often don't understand AI's power and impact. "AI is just everywhere in our lives today and the average consumer has no clue how it works or what undermines the technology," Roetzer told me. We need help understanding what responsible AI looks like. Tech companies don't have all the answers. The burden of building and using AI responsibly falls on technology companies, which don't always have incentives to build systems that prioritize people over profit.

"There is no shortage of commentary on what artificial intelligence will do to human jobs. It's easy to find a multiplicity of predictions, prescriptions, or denunciations," says Thomas H. Davenport, one of the co-authors of the book. "It is not so easy, however, to find descriptions of how people work day-to-day with smart machines." Davenport joined a Conversation That Matters about our emerging and ever-expanding relationship with a technology that scares a wide range of people including, Elon Musk and Bill Gates.

Machine learning (ML) is essential to video game development. Predicting specific in-game actions and identifying and reaching your most valuable players helps to drive better outcomes. To this end, we need to keep track of all experiments that are happening behind the scenes. Google Cloud recently announced the general availability of a new feature called Vertex AI Experiments that can now help gaming companies do just that – keep track of their ML experiments to uncover insights and best practices across their ML Engineering and Data Science teams. In this blog post, we will focus on a popular video game dataset coming from the EA Sports' FIFA video game series.

Imagine this scenario in the not-too-distant future. You're awakened at 6:11 a.m. by the gentle sounds of tinkling bells and birdsong, even though you live in a 12th-floor apartment. Your alarm clock uses radar to track your breathing, and wakes you gently, with sound and light, when it detects you're in a lighter phase of sleep. Your transition to wakefulness triggers a cascade of changes in your apartment. In the kitchen, coffee starts brewing.

Building 3D perception systems for autonomous vehicles that do not rely on high-density LiDAR is a critical research problem because of the expense of LiDAR systems compared to cameras and other sensors. Recent research has developed a variety of camera-only methods, where features are differentiably "lifted" from the multi-camera images onto the 2D ground plane, yielding a "bird's eye view" (BEV) feature representation of the 3D space around the vehicle. This line of work has produced a variety of novel "lifting" methods, but we observe that other details in the training setups have shifted at the same time, making it unclear what really matters in top-performing methods. We also observe that using cameras alone is not a real-world constraint, considering that additional sensors like radar have been integrated into real vehicles for years already. We find that batch size and input resolution greatly affect performance, while lifting strategies have a more modest effect--even a simple parameter-free lifter works well.