Connect with us

On a table in his lab at the University of Pennsylvania, physicist Sam Dillavou has connected an array of breadboards via a web of brightly colored wires. The setup looks like a DIY home electronics project—and not a particularly elegant one. But this unassuming assembly, which contains 32 variable resistors, can learn to sort data like a machine-learning model.

While its current capability is rudimentary, the hope is that the prototype will offer a low-power alternative to the energy-guzzling graphical processing unit (GPU) chips widely used in machine learning. 

“Each resistor is simple and kind of meaningless on its own,” says Dillavou. “But when you put them in a network, you can train them to do a variety of things.”

breadboards connected in a grid
Sam Dillavou’s laboratory at the University of Pennsylvania is using circuits composed of resistors to perform simple machine learning classification tasks. FELICE MACERA

A task the circuit has performed: classifying flowers by properties such as petal length and width. When given these flower measurements, the circuit could sort them into three species of iris. This kind of activity is known as a “linear” classification problem, because when the iris information is plotted on a graph, the data can be cleanly divided into the correct categories using straight lines. In practice, the researchers represented the flower measurements as voltages, which they fed as input into the circuit. The circuit then produced an output voltage, which corresponded to one of the three species.

This is a fundamentally different way of encoding data from the approach used in GPUs, which represent information as binary 1s and 0s. In this circuit, information can take on a maximum or minimum voltage or anything in between. The circuit classified 120 irises with 95% accuracy.

Now the team has managed to make the circuit perform a more complex problem. In a preprint currently under review, the researchers have shown that it can perform a logic operation known as XOR, in which the circuit takes in two binary numbers and determines whether the inputs are the same. This is a “nonlinear” classification task, says Dillavou, and “nonlinearities are the secret sauce behind all machine learning.”

Their demonstrations are a walk in the park for the devices you use every day. But that’s not the point: Dillavou and his colleagues built this circuit as an exploratory effort to find better computing designs. The computing industry faces an existential challenge as it strives to deliver ever more powerful machines. Between 2012 and 2018, the computing power required for cutting-edge AI models increased 300,000-fold. Now, training a large language model takes the same amount of energy as the annual consumption of more than a hundred US homes. Dillavou hopes that his design offers an alternative, more energy-efficient approach to building faster AI.

Training in pairs

To perform its various tasks correctly, the circuitry requires training, just like contemporary machine-learning models that run on conventional computing chips. ChatGPT, for example, learned to generate human-sounding text after being shown many instances of real human text; the circuit learned to predict which measurements corresponded to which type of iris after being shown flower measurements labeled with their species.

Training the device involves using a second, identical circuit to “instruct” the first device. Both circuits start with the same resistance values for each of their 32 variable resistors. Dillavou feeds both circuits the same inputs—a voltage corresponding to, say, petal width—and adjusts the output voltage of the second circuit to correspond to the correct species. The first circuit receives feedback from that second circuit, and both circuits adjust their resistances so they converge on the same values. The cycle starts again with a new input, until the circuits have settled on a set of resistance levels that produce the correct output for the training examples. In essence, the team trains the device via a method known as supervised learning, where an AI model learns from labeled data to predict the labels

Read More

————

By: Sophia Chen
Title: How a simple circuit could offer an alternative to energy-intensive GPUs
Sourced From: www.technologyreview.com/2024/06/05/1093250/how-a-simple-circuit-could-offer-an-alternative-to-energy-intensive-gpus/
Published Date: Wed, 05 Jun 2024 08:00:00 +0000

Continue Reading

Tech

Meet the architect creating wood structures that shape themselves

ICD ITKE HygroShell2022 P01a scaled

Humanity has long sought to tame wood into something more predictable. Sawmills manufacture lumber from trees selected for consistency. Wood is then sawed into standard sizes and dried in kilns to prevent twisting, cupping, or cracking. Generations of craftsmen have employed sophisticated techniques like dovetail joinery, breadboard ends, and pocket flooring to keep wood from distorting in their finished pieces.

But wood is inherently imprecise. Its grain reverses and swirls. Trauma and disease manifest in scars and knots.

Instead of viewing these natural tendencies as liabilities, Achim Menges, an architect and professor at the University of Stuttgart in Germany, sees them as wood’s greatest assets. Menges and his team at the Institute for Computational Design and Construction are uncovering new ways to build with the material by using computational design—which relies on algorithms and data to simulate and predict how wood will behave within a structure long before it is built. He hopes this work will enable architects to create more sustainable and affordable timber buildings by reducing the amount of wood required.

Menges’s recent work has focused on creating “self-shaping” timber structures like the HygroShell, which debuted at the Chicago Architecture Biennial in 2023. Constructed from prefabricated panels of a common building material known as cross-laminated timber, HygroShell morphed over a span of five days, unfurling into a series of interlaced sheets clad with wooden scale-like shingles that stretched to cover the structure as it expanded. Its final form, designed as a proof of concept, is a delicately arched canopy that rises to nearly 33 feet (10 meters) but is only an inch thick. In a time-lapse video, the evolving structure resembles a bird stretching its wings.

HygroShell takes its name from hygroscopicity, a property of wood that causes it to absorb or lose moisture with humidity changes. As the material dries, it contracts and tends to twist and curve. Traditionally, lumber manufacturers have sought to minimize these movements. But through computational design, Menges’s team can predict the changes and structure the material to guide it into the shape they want.

“From the start, I was motivated to understand computation not as something that divides the physical and the digital world but, instead, that deeply connects them.”

Achim Menges, architect and professor, University of Stuttgart in Germany

The result is a predictable and repeatable process that creates tighter curves with less material than what can be attained through traditional construction techniques. Existing curved structures made from cross-laminated timber (also known as mass timber) are limited to custom applications and carry premium prices, Menges says. Self-shaping, in contrast, could offer industrial-scale production of curved mass timber structures for far less cost.

To build HygroShell, the team created digital profiles of hundreds of freshly sawed boards using data about moisture content, grain orientation, and more. Those parameters were fed into modeling software that predicted how the boards were likely to distort as they dried and simulated how to arrange them to achieve the desired structure. Then the team used robotic milling machines to create the joints that held the panels together as the piece unfolded.

“What we’re trying to do is develop design methods that are so sophisticated they meet or match the sophistication of the material we deal with,” Menges says.

Menges views “self-shaping,” as he calls his technique, as a low-energy way of creating complex curved architectures that would otherwise be too difficult to build on most construction sites. Typically, making curves requires extensive machining and a lot more materials, at considerable cost. By letting the wood’s natural properties do the heavy lifting, and using robotic machinery to prefabricate the structures, Menges’s process allows for thin-walled timber construction that saves material and money.

Read More

————

By: John Wiegand
Title: Meet the architect creating wood structures that shape themselves
Sourced From: www.technologyreview.com/2024/06/24/1093513/achim-menges-architect-wood-buildings-sustainability/
Published Date: Mon, 24 Jun 2024 09:00:00 +0000

Continue Reading

Tech

The Download: hyperrealistic deepfakes, and using math to shape wood

This is today’s edition of The Download our weekday newsletter that provides a daily dose of what’s going on in the world of technology.

Synthesia’s hyperrealistic deepfakes will soon have full bodies

Startup Synthesia’s AI-generated avatars are getting an update to make them even more realistic: They will soon have bodies that can move, and hands that gesticulate.

The new full-body avatars will be able to do things like sing and brandish a microphone while dancing, or move from behind a desk and walk across a room. They will be able to express more complex emotions than previously possible, like excitement, fear, or nervousness.

These new capabilities, which are set to launch toward the end of the year, will add a lot to the illusion of realism. That’s a scary prospect at a time when deepfakes and online misinformation are proliferating. Read the full story and watch our reporter’s avatars meet each other.

—Melissa Heikkilä

Meet the architect creating wood structures that shape themselves

Humanity has long sought to tame wood into something more predictable, but it is inherently imprecise. Its grain reverses and swirls. Trauma and disease manifest in scars and knots.

Instead of viewing these natural tendencies as liabilities, Achim Menges, an architect and professor at the University of Stuttgart in Germany, sees them as wood’s greatest assets.

Menges and his team at the Institute for Computational Design and Construction are uncovering new ways to build with wood by using algorithms and data to simulate and predict how wood will behave within a structure long before it is built. He hopes this will help create more sustainable and affordable timber buildings by reducing the amount of wood required. Read our story all about him and his work.

—John Wiegand

This story is from the forthcoming print issue of MIT Technology Review, which explores the theme of Play. It’s set to go live on Wednesday June 26, so if you don’t already, subscribe now to get a copy when it lands.

Live: How generative AI could transform games

Generative AI could soon revolutionize how we play video games, creating characters that can converse with you freely, and experiences that are infinitely detailed, twisting and changing every time you experience them.

Together, these could open the door to entirely new kinds of in-game interactions that are open-ended, creative, and unexpected. One day, the games we love playing may not have to end. Read our executive editor Niall Firth’s story all about what that future could look like. 

If you want to learn more, register now to join our next exclusive subscriber-only Roundtable discussion at 11.30ET today! Niall and our editorial director Allison Arieff will be talking about games without limits, the future of play, and much more.

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Big Tech firms are going all-in on experimental clean energy projects
Due to the fact AI is so horribly polluting. But the projects range from ‘long shot’ to ‘magical thinking’. (WP $)
Making the grid smarter, rather than bigger, could help. (Semafor)
How virtual power plants are shaping tomorrow’s energy system. (MIT Technology Review)

2 Google is about to be hit with a ton of AI-related lawsuits
Its AI Overviews keep libeling people—and they’re lawyering up. (The Atlantic $)
Why Google’s AI Overviews gets things wrong. (MIT Technology Review)
Another AI-powered search engine, Perplexity, is running into the exact same issues. (Wired $)
Worst of all? There’s currently no way to fix the underlying problem. (MIT Technology Review)

3 Apple is exploring a deal with Meta
To integrate Meta’s generative AI models into Apple Intelligence. (Wall Street Journal $)
Apple is delaying launching AI features in Europe due to regulatory concerns. (Quartz)

4 NASA is indefinitely delaying the return of Starliner
In order to give it more time to review data. (Ars Technica)

5 Chinese tech companies are pushing their staff beyond breaking point
As growth slows and competition rises, work-life balance is going out the window. (FT $)

6 Used electric vehicles are now less expensive than gas cars in the US
It’s a worrying statistic that reflects the cratering demand for EVs. (Insider $)
The problem with plug-in hybrids? Their drivers. (MIT Technology Review)

7 Check out these photos of San Francisco’s AI scene
The city is currently buzzing with people hoping to make their fortune off the back of the boom. (WP $)

8 The next wave of weight loss drugs is coming
The hope is that they might be cheaper, and come with fewer side effects. (NBC)

9 Elon Musk is obsessed with getting us to have more babies
He’s funding

Read More

————

By: Charlotte Jee
Title: The Download: hyperrealistic deepfakes, and using math to shape wood
Sourced From: www.technologyreview.com/2024/06/24/1094179/the-download-hyperrealistic-deepfakes-math-shape-wood/
Published Date: Mon, 24 Jun 2024 12:10:00 +0000

Continue Reading

Tech

Is this the end of animal testing?

In a clean room in his lab, Sean Moore peers through a microscope at a bit of intestine, its dark squiggles and rounded structures standing out against a light gray background. This sample is not part of an actual intestine; rather, it’s human intestinal cells on a tiny plastic rectangle, one of 24 so-called “organs on chips” his lab bought three years ago.

Moore, a pediatric gastroenterologist at the University of Virginia School of Medicine, hopes the chips will offer answers to a particularly thorny research problem. He studies rotavirus, a common infection that causes severe diarrhea, vomiting, dehydration, and even death in young children. In the US and other rich nations, up to 98% of the children who are vaccinated against rotavirus develop lifelong immunity. But in low-income countries, only about a third of vaccinated children become immune. Moore wants to know why.

His lab uses mice for some protocols, but animal studies are notoriously bad at identifying human treatments. Around 95% of the drugs developed through animal research fail in people. Researchers have documented this translation gap since at least 1962. “All these pharmaceutical companies know the animal models stink,” says Don Ingber, founder of the Wyss Institute for Biologically Inspired Engineering at Harvard and a leading advocate for organs on chips. “The FDA knows they stink.”

But until recently there was no other option. Research questions like Moore’s can’t ethically or practically be addressed with a randomized, double-blinded study in humans. Now these organs on chips, also known as microphysiological systems, may offer a truly viable alternative. They look remarkably prosaic: flexible polymer rectangles about the size of a thumb drive. In reality they’re triumphs of bioengineering, intricate constructions furrowed with tiny channels that are lined with living human tissues. These tissues expand and contract with the flow of fluid and air, mimicking key organ functions like breathing, blood flow, and peristalsis, the muscular contractions of the digestive system.

More than 60 companies now produce organs on chips commercially, focusing on five major organs: liver, kidney, lung, intestines, and brain. They’re already being used to understand diseases, discover and test new drugs, and explore personalized approaches to treatment.

As they continue to be refined, they could solve one of the biggest problems in medicine today. “You need to do three things when you’re making a drug,” says Lorna Ewart, a pharmacologist and chief scientific officer of Emulate, a biotech company based in Boston. “You need to show it’s safe. You need to show it works. You need to be able to make it.”

All new compounds have to pass through a preclinical phase, where they’re tested for safety and effectiveness before moving to clinical trials in humans. Until recently, those tests had to run in at least two animal species—usually rats and dogs—before the drugs were tried on people.

But in December 2022, President Biden signed the FDA Modernization Act, which amended the original FDA Act of 1938. With a few small word changes, the act opened the door for non-animal-based testing in preclinical trials. Anything that makes it faster and easier for pharmaceutical companies to identify safe and effective drugs means better, potentially cheaper treatments for all of us.

Moore, for one, is banking on it, hoping the chips help him and his colleagues shed light on the rotavirus vaccine responses that confound them. “If you could figure out the answer,” he says, “you could save a lot of kids’ lives.”

While many teams have worked on organ chips over the last 30 years, the OG in the field is generally acknowledged to be Michael Shuler, a professor emeritus of chemical engineering at Cornell. In the 1980s, Shuler was a math and engineering guy who imagined an “animal on a chip,” a cell culture base seeded with a variety of human cells that could be used for testing drugs. He wanted to position a handful of different organ cells on the same chip, linked to one another, which could mimic the chemical communication between organs and the way drugs move through the body. “This was science fiction,” says Gordana Vunjak-Novakovic, a professor of biomedical engineering at Columbia University whose lab works with cardiac tissue on chips. “There was no body on a chip. There is still no body on a chip. God knows if there will ever be a body on a chip.”

Shuler had hoped to develop a computer model of a multi-organ system, but there were too many unknowns. The living cell culture system he dreamed up was his bid to fill in the blanks. For a while he played with the concept, but the materials simply weren’t good enough to build what he imagined.

“You can force mice to menstruate, but it’s not really menstruation. You need the human being.”

Linda Griffith, founding

Read More

————

By: Harriet Brown
Title: Is this the end of animal testing?
Sourced From: www.technologyreview.com/2024/06/21/1093419/animal-testing-organ-on-chip-research/
Published Date: Fri, 21 Jun 2024 09:00:00 +0000

Continue Reading

Trending