Transformers have recently gained prominence in long time series forecasting by elevating accuracies in a variety of use cases. Regrettably, in the race for better predictive performance the overhead of model architectures has grown onerous, leading to models with computational demand infeasible for most practical applications. To bridge the gap between high method complexity and realistic computational resources, we introduce the Residual Cyclic Transformer, ReCycle. ReCycle utilizes primary cycle compression to address the computational complexity of the attention mechanism in long time series. By learning residuals from refined smoothing average techniques, ReCycle surpasses state-of-the-art accuracy in a variety of application use cases. The reliable and explainable fallback behavior ensured by simple, yet robust, smoothing average techniques additionally lowers the barrier for user acceptance. At the same time, our approach reduces the run time and energy consumption by more than an order of magnitude, making both training and inference feasible on low-performance, low-power and edge computing devices. Code is available at https://github.com/Helmholtz-AI-Energy/ReCycle

The JET nuclear fusion reactor in Oxfordshire, UK, set many records over its 40 years of operation, pushing forward our understanding of how to spark and contain reactions of the type normally only found within the heart of a star. Although it has now reached the end of its life, it will break records once more as the team behind the experiment attempt something that has never been done before: recycling a fusion reactor. Nuclear fusion breakthrough: Is cheap, clean energy finally here?

Since their introduction, diffusion models have quickly become the prevailing approach to generative modeling in many domains. They can be interpreted as learning the gradients of a time-varying sequence of log-probability density functions. This interpretation has motivated classifier-based and classifier-free guidance as methods for post-hoc control of diffusion models. In this work, we build upon these ideas using the score-based interpretation of diffusion models, and explore alternative ways to condition, modify, and reuse diffusion models for tasks involving compositional generation and guidance. In particular, we investigate why certain types of composition fail using current techniques and present a number of solutions. We conclude that the sampler (not the model) is responsible for this failure and propose new samplers, inspired by MCMC, which enable successful compositional generation. Further, we propose an energy-based parameterization of diffusion models which enables the use of new compositional operators and more sophisticated, Metropolis-corrected samplers. Intriguingly we find these samplers lead to notable improvements in compositional generation across a wide set of problems such as classifier-guided ImageNet modeling and compositional text-to-image generation.

Methods for improving the efficiency of deep network training (i.e. the resources required to achieve a given level of model quality) are of immediate benefit to deep learning practitioners. Distillation is typically used to compress models or improve model quality, but it's unclear if distillation actually improves training efficiency. Can the quality improvements of distillation be converted into training speed-ups, or do they simply increase final model quality with no resource savings? We conducted a series of experiments to investigate whether and how distillation can be used to accelerate training using ResNet-50 trained on ImageNet and BERT trained on C4 with a masked language modeling objective and evaluated on GLUE, using common enterprise hardware (8x NVIDIA A100). We found that distillation can speed up training by up to 1.96x in ResNet-50 trained on ImageNet and up to 1.42x on BERT when evaluated on GLUE. Furthermore, distillation for BERT yields optimal results when it is only performed for the first 20-50% of training. We also observed that training with distillation is almost always more efficient than training without distillation, even when using the poorest-quality model as a teacher, in both ResNet-50 and BERT. Finally, we found that it's possible to gain the benefit of distilling from an ensemble of teacher models, which has O(n) runtime cost, by randomly sampling a single teacher from the pool of teacher models on each step, which only has a O(1) runtime cost. Taken together, these results show that distillation can substantially improve training efficiency in both image classification and language modeling, and that a few simple optimizations to distillation protocols can further enhance these efficiency improvements.

AMP Robotics has extended its partnership with Evergreen, a producer of food-grade recycled polyethylene terephthalate (rPET). Evergreen now has 15 of AMP's robotic sorting systems installed or planned across three facilities. In addition to six robots in Clyde, Evergreen has added six in Riverside, California, and will soon add three in Albany, New York. AMP's technology identifies and sorts green and clear PET from post-consumer bales of plastic soft drink bottles at speeds up to three times faster and at a higher accuracy than manual sorters can achieve. Evergreen then recycles the material into reusable flakes or pellets, which it sells to end markets as feedstock for new containers and packaging.

Don't tell Frank Herbert (or the people at Thinx), but he actually came up with a pretty genius pair of menstrual underwear. Only, well, his was outerwear--and it did a lot more than collect blood and endometrial lining. Herbert's invention is, of course, the stillsuit. One of the iconic pieces of tech in his novel Dune--and an iconic piece of sci-fi tech, period--it's an invention born of necessity. Arrakis, where most of the novel takes place, is a desert; to survive, the planet's native Fremen construct form-fitting suits that collect all of their moist excretions--sweat, urine, feces, droplets from exhaled breath--and recycle them into potable water.

It's no secret that kids – and cats – like to play with cardboard boxes, sometimes more than with what's inside. Now, Amazon is introducing new boxes with a built-in play factor. Starting this week, some Amazon orders will start to be delivered in "more environmentally-friendly" boxes that can be turned into a rocket, car or fort for your pet, a robot costume or a mini-golf windmill, the Seattle-based retail giant shared exclusively with USA TODAY. The new boxes are part of Amazon's ongoing "Less Packaging, More Smiles" program and include a call to action to recycle the boxes and a QR code that directs consumers to Amazon.com/ThisBox for how to make the cardboard creations. Save better, spend better: Money tips and advice delivered right to your inbox.

Earlier this week, Sonos notified customers it would end support for a number of its legacy products released in the first 10 years of its existence, including the popular first-generation PLAY:5 speakers. But some of these affected products were still on the market as late as 2015, such as the Connect and Connect:Amp, which would make them only five years old from the date of purchase. On Thursday, January 23, due to huge backlash from customers and the technology media, the CEO, Patrick Spence, authored a corporate blog post apologizing for this end of support and stated that the company would, in fact, support its older products with bug and security fixes beyond May of 2020 even if new feature updates were not possible. The reason why Sonos initially decided to end support is that these first-generation products lack sufficient processing power and storage to accommodate new features. End of service is relatively normal for consumer electronics products that are five years old or more, particularly personal computers and mobile devices.

Daniela Rus is a robot evangelist. She challenged a packed audience in the Interdisciplinary Science and Engineering Complex on Tuesday to imagine a world where robots free us to be more creative by taking care of all our physical tasks--from playing with our pets to performing surgery without an incision. As director of the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory, Rus delivered the inaugural lecture in Northeastern's Distinguished Speaker Series in Robots. "Imagine a world where you're being driven home by your autonomous car," said Rus. "Your car is connected to your refrigerator, which tells it what ingredients you need for dinner. The car is also connected to the grocery store, which is run by robots that fill your bags so they are ready when you drive up. Then you bring the food home to the robot cook and you happily let your children help in the kitchen even though they make a mess, because the mess will be taken care of by the cleaning robot."

We all know that it's important to recycle, but that's sometimes easier said than done. Luckily, Clarke the robot is here to help. Admit it -- you're not entirely sure how to recycle. With so many different materials in play, how are you supposed to know what needs to be thrown into a landfill and what can be reused? Humans might not be the best at the Three R's (reduce, reuse, and recycle, of course), but another "R" is here to save us -- a robot, affectionately named Clarke.