Amazing Science

A series of major floods in Australia has made global headlines in recent years. People around the world were shocked to see Sydney, the city known for the 2000 Olympics, the Harbour Bridge, the Opera House, sunshine and Bondi beach culture inundated with flash floods this year. But were these floods a freak occurrence or a sign of things to come?

Recent research has found an alarming increase of at least 40% in the rate at which rain falls in the most intense rapid rain bursts in Sydney over the past two decades. This rapid increase in peak rainfall intensity has never been reported elsewhere, but may be happening in other parts of the world.

The findings of a recent study, published in Science, have major implications for the city’s preparedness for flash flooding. More intense downpours are likely to overwhelm stormwater systems that were designed for past conditions.

The study demonstrated the increases in the rate of rainfall of rapid rain bursts for each of three weather radars (at Newcastle, Terrey Hills and Wollongong). All radars showed a rate of change of at least 20% per decade. To increase the climate scientists' confidence, they also calculated the change for rapid rain bursts observed by two radars (Wollongong and Terrey Hills) at the same time. The rates of change in these storms are much higher (80-90% per decade) than for storms detected only by a single radar station.

One possible explanation could be these are more extreme storms that are well developed and can be seen by two radars simultaneously. So, it is possible that the change in rainfall rates for well-developed rapid rain bursts is even greater. A dangerous situation.

What if in our attempt to build artificial intelligence we don’t simulate neurons in code and mimic neural networks in Python, but instead build actual physical neurons connected by physical synapses in ways very similar to our own biological brains? And in so doing create neural networks that are 1000X more energy efficient than existing AI frameworks? That’s precisely what Rain Neuromorphics is trying to do: build a non-biological yet very human-style artificial brain.

Which at one and the same time uses much less energy and is much faster at learning than existing AI projects. And that learns, in short, kind of like we meatspace humans do. Plus, that is built with analog chips, not digital. “We have kind of two missions that are very complimentary: One of them is to build a brain and the other one is to actually understand it,” Gordon Wilson, the soft-spoken but deep-thinking CEO of Rain Neuromorphics told me in a recent TechFirst podcast. “Ultimately, we see these as kind of like Lego pieces that due to their low-power footprint, we’ll be able to concatenate together using things like chiplet integration, advanced packaging, and ultimately scale out these systems to be brain scale — 86 billion neurons, 500 trillion synapses — and low-power enough that they can exist in autonomous devices.”

Wilson seems to be in the habit of very quietly and unassumingly saying things that are essentially completely mind-blowing and world-altering. So quietly you almost miss the gargantuan scale of the scheme. Which, in this case, is nothing less than the Frankenstein project.

Wilson and co-founders Jack Kendall and Juan Nino started four years ago with a small seed round. Late last year the team taped together a demonstration chip that proves out at least some of their theories about building brain-analog hardware for artificial intelligence workloads via a completely analog chip. And just a month ago the team was rewarded with a $25 million funding round to finish that design, engineer it to be manufacturable, and bring it to market. Who is one of the investors? Artificial intelligence heavyweight and Open AI CEO Sam Altman.

Key to the project is the fact that Rain Neuromorphics is building an analog chip. This is very different than 99.9% of the computer chips on the market that reduce reality as they see it to binary: on or off, zeroes or ones. Those chips have to model the facts and relationships and verbs of computer programs with very precise digital math.

Analog chips, on the other hand, represent reality in a very natural way. “Digital chips are ... built on the very bottom on zeros and ones, on this Boolean logic of on or off, and all of the other logic is then constructed on top of that,” explains Wilson. “When you zoom down to the bottom of an analog chip, you don’t have zeros or ones, you have gradients of information. You have voltages and currents and resistances. You have physical quantities you are measuring, that represent the mathematical operations you’re performing, and you’re exploiting the relationship between those physical quantities to then perform these very complex neural operations.”

How does that work?vBy making physics do the work of computation for us, rather than brute-forcing it through a reality-screen of ones and zeroes.vSo when you’re building out a neural network and modeling it on how the human brain so incredibly efficiently learns, stores data, and executes decisions, you are more measuring conclusions than arriving at them, using the artificial neurons and synapses that you’ve built.

BOULDER, Colo., Oct. 27, 2022 (GLOBE NEWSWIRE) -- In a recent paper published earlier this month in Nature Scientific Reports, researchers at the National Institutes of Health (NIH) conducted what may be the largest technical assessment to date of the 7K SomaScan® Platform to measure its performance and to support the platform's growing popularity among users. The NIH researchers found the platform's extensive human proteome coverage, sensitivity and consistently low variability to be a particular strength and describe the key advantages over other proteomic approaches.

For the first time, we show that non-invasive brain recordings can be used to decode images with similar performance as invasive measures.

Decoding visual stimuli from brain recordings aims to deepen our understanding of the human visual system and build a solid foundation for bridging human vision and computer vision through the Brain-Computer Interface. However, due to the scarcity of data annotations and the complexity of underlying brain information, it is challenging to decode images with faithful details and meaningful semantics.

In this work, AI scientists present MinD-Vis: Sparse Masked Brain Modeling with Double-Conditioned Diffusion Model for Vision Decoding. Specifically, by boosting the information capacity of representations learned in a large-scale resting-state fMRI dataset, they were able to show that the MinD-Vis framework reconstructed highly plausible images with semantically matching details from brain recordings with very few training pairs. The benchmarked model and its correlated method outperformed state-of-the-arts in both semantic mapping (100-way semantic classification) and generation quality (FID) by 66% and 41%, respectively. Exhaustive ablation studies are conducted to analyze this framework.

A human visual decoding system that only reply on limited annotations. State-of-the-art 100-way top-1 classification accuracy on GOD dataset: 23.9%, outperforming the previous best by 66%. State-of-the-art generation quality (FID) on GOD dataset: 1.67, outperforming the previous best by 41%.

The technology could allow scientists to both capture CO2 and transform it into useful chemicals such as carbon monoxide and synthetic natural gas in one circular process. 

Dr Melis Duyar, Senior Lecturer of Chemical Engineering at the University of Surrey commented:  “Capturing CO2 from the surrounding air and directly converting it into useful products is exactly what we need to approach carbon neutrality in the chemicals sector. This could very well be a milestone in the steps needed for the UK to reach its 2050 net-zero goals. We need to get away from our current thinking on how we produce chemicals, as current practices rely on fossil fuels which are not sustainable. With this technology we can supply chemicals with a much lower carbon footprint and look at replacing fossil fuels with carbon dioxide and renewable hydrogen as the building blocks of other important chemicals.” 

The technology uses patent-pending switchable Dual Function Materials (DFMs), that capture carbon dioxide on their surface and catalyse the conversion of captured CO2 directly into chemicals. The “switchable” nature of the DFMs comes from their ability to produce multiple chemicals depending on the operating conditions or the composition of the added reactant. This makes the technology responsive to variations in demand for chemicals as well as availability of renewable hydrogen as a reactant.  

Dr Duyar continued: “These outcomes are a testament to the research excellence at Surrey, with continuously improving facilities, internal funding schemes and a collaborative culture.” 

Loukia-Pantzechroula Merkouri, Postgraduate student leading this research at the University of Surrey added:  “Not only does this research demonstrate a viable solution to the production of carbon neutral fuels and chemicals, but it also offers an innovative approach to combat the ever-increasing CO2 emissions contributing to global warming.”  

Leprosy is one of the world’s oldest and most persistent diseases but the bacteria that cause it may also have the surprising ability to grow and regenerate a vital organ.

Is there life in Martian caves?

Last year, MIT researchers announced that they had built “liquid” neural networks, inspired by the brains of small species: a class of flexible, robust machine learning models that learn on the job and can adapt to changing conditions, for real-world safety-critical tasks, like driving and flying. The flexibility of these “liquid” neural nets meant boosting the bloodline to our connected world, yielding better decision-making for many tasks involving time-series data, such as brain and heart monitoring, weather forecasting, and stock pricing.

But these models become computationally expensive as their number of neurons and synapses increase and require clunky computer programs to solve their underlying, complicated math. And all of this math, similar to many physical phenomena, becomes harder to solve with size, meaning computing lots of small steps to arrive at a solution. 

Now, the same team of scientists has discovered a way to alleviate this bottleneck by solving the differential equation behind the interaction of two neurons through synapses to unlock a new type of fast and efficient artificial intelligence algorithms. These modes have the same characteristics of liquid neural nets — flexible, causal, robust, and explainable — but are orders of magnitude faster, and scalable. This type of neural net could therefore be used for any task that involves getting insight into data over time, as they’re compact and adaptable even after training — while many traditional models are fixed. 

The models, dubbed a “closed-form continuous-time” (CfC) neural network, outperformed state-of-the-art counterparts on a slew of tasks, with considerably higher speedups and performance in recognizing human activities from motion sensors, modeling physical dynamics of a simulated walker robot, and event-based sequential image processing. On a medical prediction task, for example, the new models were 220 times faster on a sampling of 8,000 patients. 

The new paper on the work is published in Nature Machine Intelligence.

Do the laws of physics place a hard limit on how far technology can develop, or could an advanced civilization re-write those laws?

How different does sunset appear from Mars than from Earth?

For comparison, two images of our common star were taken at sunset, one from Earth and one from Mars. These images were scaled to have same angular width and featured here side-by-side. A quick inspection will reveal that the Sun appears slightly smaller from Mars than from Earth. This makes sense since Mars is 50% further from the Sun than Earth. More striking, perhaps, is that the Martian sunset is noticeably bluer near the Sun than the typically orange colors near the setting Sun from Earth. The reason for the blue hues from Mars is not fully understood, but thought to be related to forward scattering properties of Martian dust.

The terrestrial sunset was taken in March 2012 from Marseille, France, while the Martian sunset was captured in 2015 by NASA's Curiosity rover from Gale crater on Mars. 

Accurately moving to a musical beat was thought to be a skill innately unique to humans. However, new research now shows that rats also have this ability. The optimal tempo for nodding along was found to depend on the time constant in the brain (the speed at which our brains can respond to something), which is similar across all species. This means that the ability of our auditory and motor systems to interact and move to music may be more widespread among species than previously thought. This new discovery offers not only further insight into the animal mind, but also into the origins of our own music and dance.

Can you move to the beat, or do you have two left feet? Apparently, how well we can time our movement to music depends somewhat on our innate genetic ability, and this skill was previously thought to be a uniquely human trait. While animals also react to hearing noise, or might make rhythmic sounds, or be trained to respond to music, this isn't the same as the complex neural and motor processes that work together to enable us to naturally recognize the beat in a song, respond to it or even predict it. This is referred to as beat synchronicity.

Only relatively recently, research studies (and home videos) have shown that some animals seem to share our urge to move to the groove. A new paper by a team at the University of Tokyo provides evidence that rats are one of them. "Rats displayed innate -- that is, without any training or prior exposure to music -- beat synchronization most distinctly within 120-140 bpm (beats per minute), to which humans also exhibit the clearest beat synchronization," explained Associate Professor Hirokazu Takahashi from the Graduate School of Information Science and Technology.

"The auditory cortex, the region of our brain that processes sound, was also tuned to 120-140 bpm, which we were able to explain using our mathematical model of brain adaptation." But why play music to rats in the first place? "Music exerts a strong appeal to the brain and has profound effects on emotion and cognition. To utilize music effectively, we need to reveal the neural mechanism underlying this empirical fact," said Takahashi. "I am also a specialist of electrophysiology, which is concerned with electrical activity in the brain, and have been studying the auditory cortex of rats for many years."

The team had two alternate hypotheses: The first was that the optimal music tempo for beat synchronicity would be determined by the time constant of the body. This is different between species and much faster for small animals compared to humans (think of how quickly a rat can scuttle). The second was that the optimal tempo would instead be determined by the time constant of the brain, which is surprisingly similar across species.

"After conducting our research with 20 human participants and 10 rats, our results suggest that the optimal tempo for beat synchronization depends on the time constant in the brain," said Takahashi. "This demonstrates that the animal brain can be useful in elucidating the perceptual mechanisms of music." The rats were fitted with wireless, miniature accelerometers, which could measure the slightest head movements. Human participants also wore accelerometers on headphones. They were then played one-minute excerpts from Mozart's Sonata for Two Pianos in D Major, K. 448, at four different tempos: Seventy-five percent, 100%, 200% and 400% of the original speed. The original tempo is 132 bpm and results showed that the rats' beat synchronicity was clearest within the 120-140 bpm range. The team also found that both rats and humans jerked their heads to the beat in a similar rhythm, and that the level of head jerking decreased the more that the music was sped up.

"To the best of our knowledge, this is the first report on innate beat synchronization in animals that was not achieved through training or musical exposure," said Takahashi. "We also hypothesized that short-term adaptation in the brain was involved in beat tuning in the auditory cortex. We were able to explain this by fitting our neural activity data to a mathematical model of the adaptation. Furthermore, our adaptation model showed that in response to random click sequences, the highest beat prediction performance occurred when the mean interstimulus interval (the time between the end of one stimulus and the start of another) was around 200 milliseconds. This matched the statistics of internote intervals in classical music, suggesting that the adaptation property in the brain underlies the perception and creation of music."

Drivers can struggle to see when driving at night in a rainstorm or snowstorm, but a smart headlight system invented by researchers at Carnegie Mellon University's Robotics Institute can improve visibility by constantly redirecting light to shine between particles of precipitation. The system, demonstrated in laboratory tests, prevents the distracting and sometimes dangerous glare that occurs when headlight beams are reflected by precipitation back toward the driver.

The first headlights to adorn automobiles weren’t all that much better than squinting real hard and hoping any cows in the road had the good sense to move out of your way. The dim light cast by early kerosene oil and acetylene gas lamps made most travel after dark a fool’s errand. 

Today, of course, the latest generation of headlights work much like modern televisions with tightly packed arrays of pixelated lights blinking at up to 5,000 times a second, allowing drivers to essentially use high and low beams at the same time. Until very recently, however, cutting-edge features like that weren’t allowed on vehicles sold in the US due to an NHTSA regulation set in the 1960s. But thanks to a multi-year lobbying effort on the part of Toyota, those regulations changed this last February — now America’s roadways are about to become a bit brighter and a whole lot safer.

Following the short-lived idea of using open flames to light the way, the first electric headlights appeared on the 1912 Cadillac Model 30 and, by the next decade, were quickly becoming mandatory equipment across the nation. The first split-intensity headlights offering separate low and high beams were produced in 1915 but wouldn’t be included in a vehicle’s OEM design until in 1924 and the floor-mounted switch that controlled them wouldn’t be invented until three years after that — a full decade of having to get out of the car just to turn your lights on and blink between brightnesses!

The advent of sealed beam headlights with filaments for both low and high beams in 1954, and its widespread adoption by 1957, proved a massive technological leap. With low beams for dusk and evening driving, and high beams for late night travel on otherwise unlit roads, these new headlights would drastically extend the hours of day a car could safely be on the road.

The first halogen light, which would itself quickly become a global standard, debuted in 1962. But halogens at that time were about as popular in the US as the metric system — we still preferred tungsten incandescents. That changed with the passage of the Motor Vehicle Safety Act of 1966 and the formation of the National Highway Transportation Authority (the NHTSA) in 1968, which took the existing hodge-podge of state-level vehicular regulations and federalized them, as well as the formal adoption that year of Federal Motor Vehicle Safety Standard (FMVSS) 108, which dictated that all headlights be constructed of sealed beams.

Credit: Srinivasa Narasimhan/Carnegie Mellon University

Read more at Futurity: http://www.futurity.org

Text-to-Image models have introduced a remarkable leap in the evolution of machine learning, demonstrating high-quality synthesis of images from a given text-prompt. However, these powerful pretrained models still lack control handles that can guide spatial properties of the synthesized images.

In this work, the authors introduce a universal approach to guide a pretrained text-to-image diffusion model, with a spatial map from another domain (e.g., sketch) during inference time. Unlike previous works, this new method does not require to train a dedicated model or a specialized encoder for the task. The key idea here is to train a Latent Guidance Predictor (LGP) - a small, per-pixel, Multi-Layer Perceptron (MLP) that maps latent features of noisy images to spatial maps, where the deep features are extracted from the core Denoising Diffusion Probabilistic Model (DDPM) network. The LGP is trained only on a few thousand of images and constitutes a differential guiding map predictor, over which the loss is computed and propagated back to push the intermediate images to agree with the spatial map. The per-pixel training offers flexibility and locality which allows the technique to perform well on out-of-domain sketches, including free-hand style drawings. A particular focus is taken on the sketch-to-image translation task, revealing a robust and expressive way to generate images that follow the guidance of a sketch of arbitrary style or domain.

A new type of material can learn and improve its ability to deal with unexpected forces thanks to a unique lattice structure with connections of variable stiffness, as described in a new paper. The new material is a type of architected material, which gets its properties mainly from the geometry and specific traits of its design rather than what it is made out of. Take hook-and-loop fabric closures like Velcro, for example. It doesn’t matter whether it is made from cotton, plastic or any other substance. As long as one side is a fabric with stiff hooks and the other side has fluffy loops, the material will have the sticky properties of Velcro.

The new material’s architecture is based on that of an artificial neural network—layers of interconnected nodes that can learn to do tasks by changing how much importance, or weight, they place on each connection. Theoretically, such a mechanical lattice with physical nodes could be trained to take on certain mechanical properties by adjusting each connection’s rigidity.

To find out if a mechanical lattice indeed would be able to adopt and maintain new properties—like taking on a new shape or changing directional strength—scientists started off by building a computer model. They then selected a desired shape for the material as well as input forces and had a computer algorithm tune the tensions of the connections so that the input forces would produce the desired shape. They did this training on 200 different lattice structures and found that a triangular lattice was best at achieving all of the shapes tested.

Once the many connections are tuned to achieve a set of tasks, the material will continue to react in the desired way. The training is—in a sense—remembered in the structure of the material itself.

The researchers then built a physical prototype lattice with adjustable electromechanical springs arranged in a triangular lattice. The prototype is made of 6-inch connections and is about 2 feet long by 1½ feet wide. And it worked beautifully. When the lattice and algorithm worked together, the material was able to learn and change shape in particular ways when subjected to different forces. The scientists call this new material a mechanical neural network.

Artificial intelligence has revealed that prehistoric footprints thought to be made by a vicious dinosaur predator were in fact from a timid herbivore. In an international collaboration, University of Queensland palaeontologist Dr Anthony Romilio used AI pattern recognition to re-analyse footprints from the Dinosaur Stampede National Monument, south-west of Winton in Central Queensland.

In San Francisco, California, the self-driving tech company – owned by Google parent Alphabet – moved a step closer to launching a fully autonomous commercialized ride-hailing service, as is currently being operated by its chief rival, General Motors-owned Cruise.

And in Phoenix, Arizona, Waymo’s driverless service has been made available to members of the general public in the central Downtown area. The breakthrough in San Francisco has come via the approval by the California Department of Motor Vehicles to an amendment of the company’s current permit to operate. Now Waymo will be able to charge fees for driverless services in its autonomous vehicles (AVs), such as deliveries.

Once it has operated a driverless service on public roads in the city for a total of 30 days, it will then be eligible to submit an application to the California Public Utilities Commission (CPUC) for a permit that would enable it to charge fares for passenger-only autonomous rides in its vehicles.

This is the same permit that provided the greenlight for Cruise’s commercial driverless ride-hail service at the start of June. The CPUC awarded a drivered deployment permit to Waymo in February which allowed the company to charge its ‘trusted testers’ for autonomous rides with a safety operator on board. The trusted tester program comprises vetted members of the public who have applied to use the service and have signed an NDA which means they will not talk about their experiences publicly. In downtown Phoenix, the extension of the driverless ride-hail service is the latest evidence of the incremental progress Waymo has made in the city.

Over the past couple of years, the company has operated a paid rider-only service in some of Phoenix’s eastern suburbs, such as Gilbert, Mesa, Chandler and Tempe. Earlier this year it moved into the busier, more central downtown area, where driverless rides were made available for trusted testers. It also trialled an autonomous service for employees at Phoenix Sky Harbor International Airport, albeit with a safety operator on board. In early November, it was confirmed that airport rides would be offered to trusted testers, although again Waymo made clear that there would be a specialist in the driver’s seat, initially at least.

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations (Google, OpenAI, Microsoft, etc.) and are frequently kept private and away from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176 Billion-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). The developers find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, the BLOOM team will publicly release all models and code under the Responsible AI License.

How do you integrate the advantages of a benchtop laser that fills a room onto a semiconductor chip the size of a fingernail? A research team co-led by Qiang Lin, a professor of electrical and computer engineering at the University of Rochester, has set new milestones in addressing this challenge, with the first multi-color integrated Pockels laser that:

• Emits high-coherence light at telecommunication wavelengths
• Allows laser-frequency tuning at record speeds
• Is the first narrow linewidth laser with fast configurability at the visible band

The project, described in Nature Communications, was co-led by John Bowers, distinguished professor at University of California/Santa Barbara, and Kerry Vahala, professor at the California Institute of Technology. Lin Zhu, professor at Clemson University, also collaborated on the project. The technology “has the potential to reshape the landscape of integrated photonics,” write co-lead authors Mingxiao Li, a former PhD student in Lin’s Laboratory for Nanophotonics at Rochester’s Hajim School of Engineering & Applied Sciences, and Lin Chang, a former postdoctoral student at University of California/Santa Barbara. It will pave the way for new applications of integrated semiconductor lasers in LiDAR (Light Detection and Ranging) remote sensing that is used, for example, in self-driving cars. The technology could also lead to advances in microwave photonics, atomic physics, and AR/VR.

A ‘fully on-chip laser solution’

Integrated semiconductor lasers have been at the core of integrated photonics, enabling many advances over the last few decades in information technologies and basic science. “However, despite these impressive achievements, key functions are missing in current integrated lasers,” Li says. “Two major challenges, the lack of fast reconfigurability and the narrow spectral window, have become major bottlenecks that stall the progression of many evolving applications,” Chang adds.

The researchers say they’ve overcome these challenges by creating a new type of integrated semiconductor laser, based on the Pockels effect. The laser is integrated with a lithium-niobate- on-insulator platform.

The new technology includes these beneficial features:

• Fast frequency chirping, which will be invaluable in LiDAR sensor systems, which measure distance by recording the time between emission of a short pulse and reception of reflected light.
• Frequency conversion capabilities that overcome spectral bandwidth limitations of traditional integrated semiconductor lasers. This will “significantly relieve” the difficulties in developing new wavelength lasers.
• Narrow wavelength and fast reconfigurability, providing a “fully on-chip laser solution” to probe and manipulate atoms and ions in atomic physics, and benefit AR/VR and other applications at short wavelengths.

The researchers see applications for a new type of integrated semiconductor laser in LiDAR, atomic physics, AR/VR.

Moore’s Law predicts that the number of transistors per device will double every two years. Moore’s Law is and always has been driven by innovation. The above figure illustrates the number of transistors per device as we look to the past, the present and the future. For the first 40 years, the gains came primarily from innovations in our process. Going forward, gains will come from innovations in both process and packaging. INTEL's processes will continue to deliver historic density improvements, while its 2D and 3D stacking technologies give architects and designers more tools to increase the number of transistors per device. As the designers look forward to innovative technologies such as High NA, RibbonFET, PowerVia, Foveros Omni and Direct, and others, INTEL sees no end to innovation and therefore currently no end to Moore’s Law.

In summary, when we consider all the various process and advanced packaging innovations, there are numerous options available to continue to double the number of transistors per device at the cadence demanded by our customers. Moore’s Law only stops when innovation stops, and innovation continues unabated at Intel in process, packaging and architecture. We remain undeterred in our aspiration to deliver approximately 1 trillion transistors in a single device by 2030.