Variational Information Pursuit (V-IP) is a framework for making interpretable predictions by design by sequentially selecting a short chain of task-relevant, userdefined and interpretable queries about the data that are most informative for the task. While using queries related with semantic concepts allows for built-in interpretability in predictive models, applying V-IP to any task requires data samples with concept-labeling by domain experts, limiting the application of V-IP to smallscale tasks where manual data annotation is feasible. In this work, we extend the V-IP framework with Foundational Models (FMs) to address this limitation. More specifically, we use a two-step process, by first leveraging Large Language Models (LLMs) to generate a sufficiently large candidate set of task-relevant interpretable concepts, then using multimodal models to annotate each data sample by semantic similarity with each concept in the generated concept set. While other interpretableby-design frameworks such as Concept Bottleneck Models (CBMs) require an additional step of removing repetitive and non-discriminative concepts to have good interpretability and test performance, we mathematically and empirically justify that, with a sufficiently informative and task-relevant query (concept) set, the proposed FM+V-IP method does not require any type of concept filtering. In addition, we show that FM+V-IP with LLM generated concepts can achieve better test performance than V-IP with human annotated concepts, demonstrating the effectiveness of LLMs at generating efficient query sets. Finally, when compared to other interpretable-by-design frameworks such as CBMs, FM+V-IP can achieve competitive test performance using fewer number of concepts/queries in both cases with filtered or unfiltered concept sets.

Regularization techniques are crucial to improving the generalization performance and training efficiency of deep neural networks. Many deep learning algorithms rely on weight decay, dropout, batch/layer normalization to converge faster and generalize. Label Smoothing (LS) is another simple, versatile and efficient regularization which can be applied to various supervised classification tasks. Conventional LS, however, regardless of the training instance assumes that each non-target class is equally likely. In this work, we present a general framework for training with label regularization, which includes conventional LS but can also model instance-specific variants. Based on this formulation, we propose an efficient way of learning LAbel regularization by devising a Bi-level Optimization (LABO) problem. We derive a deterministic and interpretable solution of the inner loop as the optimal label smoothing without the need to store the parameters or the output of a trained model. Finally, we conduct extensive experiments and demonstrate our LABO consistently yields improvement over conventional label regularization on various fields, including seven machine translation and three image classification tasks across various

Concept Bottleneck Models (CBM) are inherently interpretable models that factor model decisions into human-readable concepts. They allow people to easily understand why a model is failing, a critical feature for high-stakes applications. CBMs require manually specified concepts and often under-perform their black box counterparts, preventing their broad adoption. We address these shortcomings and are first to show how to construct high-performance CBMs without manual specification of similar accuracy to black box models. Our approach, Language Guided Bottlenecks (LaBo), leverages a language model, GPT-3, to define a large space of possible bottlenecks. Given a problem domain, LaBo uses GPT-3 to produce factual sentences about categories to form candidate concepts. LaBo efficiently searches possible bottlenecks through a novel submodular utility that promotes the selection of discriminative and diverse information. Ultimately, GPT-3's sentential concepts can be aligned to images using CLIP, to form a bottleneck layer. Experiments demonstrate that LaBo is a highly effective prior for concepts important to visual recognition. In the evaluation with 11 diverse datasets, LaBo bottlenecks excel at few-shot classification: they are 11.7% more accurate than black box linear probes at 1 shot and comparable with more data. Overall, LaBo demonstrates that inherently interpretable models can be widely applied at similar, or better, performance than black box approaches.

Released in April, Nintendo Labo was one of the more unusual games of this year – or any year. The box contains cardboard sheets, rubber bands and string along with a game cartridge, inviting players to build ingenious little cardboard models that, when combined with the Nintendo Switch console and its controllers, become working interactive toys. It's rather like cardboard Lego, presented in a way that gently introduces the basics of engineering. Labo is not as playground-popular as Minecraft or Fortnite, but it's a rare video game that provides educational value as well as fun, and does so without forcing it down kids' throats. There are three Nintendo Labo sets available: the Variety Kit, the Robot Kit and the Vehicles Kit.

Nintendo Labo is a weird idea. Nintendo considers itself a toy company, not just a game company. That's one reason so many Nintendo products are more than just games, whether that's the Wii motion controllers or the amiibo line of toys or even, to some degree, the retro Nintendo consoles. Nintendo Labo takes the Nintendo Switch and uses it, along with cardboard cutouts, string and censors, to create a variety of interesting, interactive creations. These range from a robot pack you wear on your back to a piano that actually plays sounds.

In an industry now in its mid-to-late 30s, and still with a predominantly male workforce, the glut of recent blockbuster video games featuring father-child relationships surely reflects the preoccupations of the men who make them. God of War is the latest specimen: a game in which a monosyllabic muscleman is on a journey to scatter his late wife's ashes on the tallest mountain in Norse myth, while accompanied by his young son. Previously the God of War series, which debuted in 2005, had little time to explore the emotional landscape of its testosterone-pumped protagonist Kratos, whose only downtime from tearing the balls from mythological monsters was spent gruffly shagging mute slave girls. God of War was always something akin to Marvel does Greek mythology (which, to be fair, was pretty much how Homer did Greek mythology): all brutal set-pieces that, with their lingering camera angles and splattering money shots, treated violence as pornography. It was a peculiarly American vision for the mid-2000s video game action blockbuster, one that has aged quicker than its protagonist's tribal tattoos.

Over the last week, I've spent more than 20 hours folding and assembling cardboard, and I've learned a few things. One: You don't want to follow exactly in my footsteps. And two: Nintendo's Labo is ingenious. It's something few other companies could have produced and turns the Switch into so much more than a game console: With Labo, it's an engine powering a whole new world of DIY creations. Building Labo kits can be a pretty huge time sink. But for some, that might be a good thing.

Nintendo has a history of making people ask why. Why make a console that can come apart and plug into a TV; why did it soldier on for so long with cartridges for games; why is Mario a plumber and wear a corresponding outfit despite not apparently having done it for decades; why the Wii U? It has never stopped, all the way up to its latest release: the Switch, which came at a risky time for the company but helped them pull off exactly what it needed. The company's newest product, Labo, is marketed as an accessory for that console but is actually a huge box full of pieces of perforated cardboard that can be popped out, folded and assembled into a variety of accessories: everything from a small remote control car that drives around using vibration to an entire robot suit that can be strapped on to operate a virtual version of the same robot in a game. It is perhaps the company's most why-inducing release yet. But the answer has, for the most part, always been the same. And it is the same this time, too.

I built a car today. And it only took me 10 minutes. A remote-controlled car is the first project you make as part of the Nintendo Labo, a new type of build-it-yourself cardboard accessory that works with the Switch console to combine real-world and digital play. Then, using the console and its touch screen, I used the vibrations to drive it around. The car is just one project you can make using Labo, which hit store shelves Friday. It is a pretty out-there product from Nintendo.

I built a car today. And it only took me 10 minutes. A remote-controlled car is the first project you make as part of the Nintendo Labo, a new type of build-it-yourself cardboard accessory that works with the Switch console to combine real-world and digital play. Then, using the console and its touch screen, I used the vibrations to drive it around. The car is just one project you can make using Labo, which hit store shelves Friday. It is a pretty out-there product from Nintendo.