Aurelia Butler

An entirely new predictive typing model can simulate different kinds of users, helping figure out ways to optimize how we use our phones. Developed by researchers at Aalto University, the new model captures the difference between typing with one or two hands or between younger and older users.
‘Typing on a phone requires manual dexterity and visual perception: we press buttons, proofread text, and correct mistakes. We also use our working memory. Automatic text correction functions can help some people, while for others they can make typing harder,’ says Professor Antti Oulasvirta of Aalto University.
The researchers created a machine-learning model that uses its virtual ‘eyes and fingers’ and working memory to type out a sentence, just like humans do. That means it also makes similar mistakes and has to correct them.
‘We created a simulated user with a human-like visual and motor system. Then we trained it millions of times in a keyboard simulator. Eventually, it learned typing skills that can also be used to type in various situations outside the simulator,’ explains Oulasvirta.
The predictive typing model was developed in collaboration with Google. New designs for phone keyboards are normally tested with real users, which is costly and time-consuming. The project’s goal is to complement those tests so keyboards can be evaluated and optimized more quickly and easily.
For Oulasvirta, this is part of a larger effort to improve user interfaces overall and understand how humans behave in task-oriented situations. He leads a research group at Aalto that uses computational models of human behaviour to probe these questions.
‘We can train computer models so that we don’t need observation of lots of people to make predictions. User interfaces are everywhere today — fundamentally, this work aims to create a more functional society and smoother everyday life,’ he says.
The researchers will present their findings at the CHI Conference in May. More

Contrary to previous assumptions, nerve cells in the human neocortex are wired differently than in mice. Those are the findings of a new study conducted by Charité — Universitätsmedizin Berlin and published in the journal Science.* The study found that human neurons communicate in one direction, while in mice, signals tend to flow in loops. This increases the efficiency and capacity of the human brain to process information. These discoveries could further the development of artificial neural networks.
The neocortex, a critical structure for human intelligence, is less than five millimeters thick. There, in the outermost layer of the brain, 20 billion neurons process countless sensory perceptions, plan actions, and form the basis of our consciousness. How do these neurons process all this complex information? That largely depends on how they are “wired” to each other.
More complex neocortex — different information processing
“Our previous understanding of neural architecture in the neocortex is based primarily on findings from animal models such as mice,” explains Prof. Jörg Geiger, Director of the Institute for Neurophysiology at Charité. In those models, the neighboring neurons frequently communicate with each other as if they are in dialogue. One neuron signals another, and then that one sends a signal back. That means the information often flows in recurrent loops.”
The human neocortex is much thicker and more complex than that of a mouse. Nonetheless, researchers had previously assumed — in part due to lack of data — that it follows the same basic principles of connectivity. A team of Charité researchers led by Geiger has now used exceptionally rare tissue samples and state-of-the-art technology to demonstrate that this is not the case.
A clever method of listening in on neuronal communication
For the study, the researchers examined brain tissue from 23 people who had undergone neurosurgery at Charité to treat drug-resistant epilepsy. During surgery, it was medically necessary to remove brain tissue in order to gain access to the diseased structures beneath it. The patients had consented to the use of this access tissue for research purposes.

To be able to observe the flows of signals between neighboring neurons in the outermost layer of the human neocortex, the team developed an improved version of what is known as the “multipatch” technique. This allowed the researchers to listen in on the communications taking place between as many as ten neurons at once (for details, see “About the method”). As a result, they were able to take the necessary number of measurements to map the network in the short time before the cells ceased their activity outside the body. In all, they analyzed the communication channels among nearly 1,170 neurons with about 7,200 possible connections.
Feed-forward instead of in cycles
They found that only a small fraction of the neurons engaged in reciprocal dialogue with each other. “In humans, the information tends to flow in one direction instead. It seldom returns to the starting point either directly or via cycles,” explains Dr. Yangfan Peng, first author of the publication. He worked on the study at the Institute for Neurophysiology and is now based at the Department of Neurology and the Neuroscience Research Center at Charité. The team used a computer simulation that they devised according to the same principles underlying the human network architecture to demonstrate that this forward-directed signal flow has benefits in terms of processing data.
The researchers gave the artificial neural network a typical machine learning task: recognizing the correct numbers from audio recordings of spoken digits. The network model that mimicked the human structures achieved more correct responses to this speech recognition task than the one modeled on mice. It was also more efficient, with the same performance requiring the equivalent of 380 neurons in the mouse model, but only 150 in the human one.
An economic role model for AI?
“The directed network architecture we see in humans is more powerful and conserves resources because more independent neurons can handle different tasks simultaneously,” Peng explains. “This means that the local network can store more information. It isn’t clear yet whether our findings within the outermost layer of the temporal cortex extend to other cortical regions, or how well they might explain the unique cognitive abilities of humans.”
In the past, AI developers have looked to biological models for inspiration in designing artificial neural networks, but have also optimized their algorithms independently of the biological models. “Many artificial neural networks already use some form of this forward-directed connectivity because it delivers better results for some tasks,” Geiger says. “It’s fascinating to see that the human brain also shows similar network principles. These insights into cost-efficient information processing in the human neocortex could provide further inspiration for refining AI networks.” More

An international research team led by scientists from the CNRS1 has discovered that the magnetic nanobubbles2 known as skyrmions can be moved by electrical currents, attaining record speeds up to 900 m/s.
Anticipated as future bits in computer memory, these nanobubbles offer enhanced avenues for information processing in electronic devices. Their tiny size3 provides great computing and information storage capacity, as well as low energy consumption.
Until now, these nanobubbles moved no faster than 100 m/s, which is too slow for computing applications. However, thanks to the use of an antiferromagnetic material4 as medium, the scientists successfully had the skyrmions move 10 times faster than previously observed.
These results, which were published in Science on 19 March, offer new prospects for developing higher-performance and less energy-intensive computing devices.
This study is part of the SPIN national research programme5 launched on 29 January, which supports innovative research in spintronics, with a view to helping develop a more agile and enduring digital world.
notes :
1 — The French laboratories involved are SPINTEC (CEA/CNRS/Université Grenoble Alpes), the Institut Néel (CNRS), and the Charles Coulomb Laboratory (CNRS/Université de Montpellier).

2 — A skyrmion consists of elementary nanomagnets (“spins”) that wind to form a highly stable spiral structure, like a tight knot.
3 — The size of a skyrmion can reach a few nanometres, which is to say approximately a dozen atoms.
4 — Antiferromagnetic stacks consist of two nano-sized ferromagnetic layers (such as cobalt) separated by a think non-magnetic layer, with opposite magnetisation.
5 — The SPIN priority research programme and equipment (PEPR) is an exploratory programme in connection with the France 2030 investment plan. More

With more than 200 types of cancer and every cancer individually unique, ongoing efforts to develop precision oncology treatments remain daunting. Most of the focus has been on developing genetic sequencing assays or analyses to identify mutations in cancer driver genes, then trying to match treatments that may work against those mutations.
But many, if not most, cancer patients do not benefit from these early targeted therapies. In a new study published on April 18, 2024, in the journal Nature Cancer, first author Sanju Sinha, Ph.D., assistant professor in the Cancer Molecular Therapeutics Program at Sanford Burnham Prebys, with senior authors Eytan Ruppin, M.D., Ph.D., and Alejandro Schaffer, Ph.D., at the National Cancer Institute, part of the National Institutes of Health (NIH) — and colleagues — describe a first-of-its-kind computational pipeline to systematically predict patient response to cancer drugs at single-cell resolution.
Dubbed PERsonalized Single-Cell Expression-Based Planning for Treatments in Oncology, or PERCEPTION, the new artificial intelligence-based approach dives deeper into the utility of transcriptomics — the study of transcription factors, the messenger RNA molecules expressed by genes that carry and convert DNA information into action.
“A tumor is a complex and evolving beast. Using single-cell resolution can allow us to tackle both of these challenges,” says Sinha. “PERCEPTION allows for the use of rich information within single-cell omics to understand the clonal architecture of the tumor and monitor the emergence of resistance.” (In biology, omics refers to the sum of constituents within a cell.)
Sinha says, “The ability to monitor the emergence of resistance is the most exciting part for me. It has the potential to allow us to adapt to the evolution of cancer cells and even modify our treatment strategy.”
Sinha and colleagues used transfer learning — a branch of AI — to build PERCEPTION.
“Limited single-cell data from clinics was our biggest challenge. An AI model needs large amounts of data to understand a disease, not unlike how ChatGPT needs huge amounts of text data scraped from the internet.”
PERCEPTION uses published bulk-gene expression from tumors to pre-train its models. Then, single-cell data from cell lines and patients, even though limited, was used to tune the models.

PERCEPTION was successfully validated by predicting the response to monotherapy and combination treatment in three independent, recently published clinical trials for multiple myeloma, breast and lung cancer.
In each case, PERCEPTION correctly stratified patients into responder and non-responder categories. In lung cancer, it even captured the development of drug resistance as the disease progressed, a notable discovery with great potential.
Sinha says that PERCEPTION is not ready for clinics, but the approach shows that single-cell information can be used to guide treatment. He hopes to encourage the adoption of this technology in clinics to generate more data, which can be used to further develop and refine the technology for clinical use.
“The quality of the prediction rises with the quality and quantity of the data serving as its foundation,” says Sinha. “Our goal is to create a clinical tool that can predict the treatment response of individual cancer patients in a systematic, data-driven manner. We hope these findings spur more data and more such studies, sooner rather than later.”
Additional authors on the study include Rahulsimham Vegesna, Sumit Mukherjee, Ashwin V. Kammula, Saugato Rahman Dhruba, Nishanth Ulhas Nair, Peng Jiang, Alejandro Schäffer, Kenneth D. Aldape and Eytan Ruppin, National Cancer Institute (NCI); Wei Wu, Lucas Kerr, Collin M. Blakely and Trever G. Biovona, University of California, San Francisco; Mathew G. Jones and Nir Yosef, University of California, Berkeley; Oleg Stroganov and Ivan Grishagin, Rancho BioSciences; Craig J. Thomas, National Institutes of Health; and Cyril H. Benes, Harvard University.
This research was supported in part by the Intramural Research Program of the NIH; NCI; and NIH grants R01CA231300, R01CA204302, R01CA211052, R01CA169338 and U54CA224081. More

Multiple sclerosis (MS) is an insidious disease. Patients suffer because their immune system is attacking their own nerve fibres, which inhibits the transmission of nerve signals. People with MS experience mild to severe impairment of their motor function and sensory perception in a variety of ways. These impairments disrupt their daily activities and reduce their overall quality of life. As individual as the symptoms and progression of the disease are, so too is the way it is managed. To monitor the disease progression and be able to recommend effective treatments, physicians ask their patients on a regular basis to describe their symptoms, such as fatigue.
Going off memory
Patients are thus faced with the tricky task of having to provide information about their state of health and what they have been capable of over the past few weeks and even months from memory. The data gathered in this way can be inaccurate and incomplete because patients might misremember details or tailor their responses to social expectations. And since these responses have a significant impact on how the progression of the disease is recorded, it could be mismanaged.
“Physicians would benefit from having access to reliable, frequent and long-term measurements of patients’ health parameters that give an accurate and comprehensive view of their state of health,” explains Shkurta Gashi. She is lead author of a new study and postdoc in the groups led by ETH Professors Christian Holz and Gunnar Rätsch at the Department of Computer Science as well as a fellow of the ETH AI Center.
Together with colleagues from ETH Zurich, the University Hospital Zurich, and the University of Zurich, Gashi has now shown that fitness trackers and smartphones can provide this kind of reliable long-term data with a high temporal resolution. Their study was published in the journal NPJ Digital Medicine.
Digital markers for MS
The researchers recruited a group of volunteers — 55 with MS and a further 24 serving as control subjects — and provided each person with a fitness tracking armband. Over the course of two weeks, the researchers collected data from these wearable devices as well as from participants’ smartphones. They then performed statistical tests and a machine learning analysis of this data to identify reliable and clinically useful information.

What proved particularly meaningful was the data on physical activity and heart rate, which was collected from participants’ wearable devices. The higher the participants’ disease severity and fatigue levels, the lower their physical activity and heart rate variability proved to be. Compared to the controls, MS patients took fewer steps per day, engaged in an overall lower level of physical activity and registered more consistent intervals between heartbeats.
How often people used their smartphone also delivered important information about their disease severity and fatigue levels: the less often a study participant used their phone, the greater their level of disability and the more severe their level of fatigue. The researchers gained insights into motor function through a game-like smartphone test. Developed at ETH a few years ago, this test requires the user to tap the screen as quickly as possible to make a virtual person move as fast as possible. Monitoring how fast a person taps and how their tapping frequency changes over time allows the researchers to draw conclusions about their motor skills and physical fatigue.
“Altogether, the combination of data from the fitness tracker and smartphone lets us distinguish between healthy participants and those with MS with a high degree of accuracy,” Gashi says. “Combining information related to several aspects of the disease, including physiological, behavioural, motor performance and sleep information, is crucial for more effective and accurate monitoring of the disease.”
Reliable approach
This new approach gives MS sufferers a straightforward way of collecting reliable and clinically useful long-term data as they go about their day-to-day lives. The researchers expect that this type of data can lead to better treatments and more effective disease management techniques: more comprehensive, precise and reliable data helps experts make better decisions and possibly even propose effective treatments sooner than before. What’s more, evaluating this patient data lets the experts verify the effectiveness of different treatments.
The researchers have now made their data set available to other scientists. They also point out the need for a larger study and more data to develop reliable and generalizable models for automatic evaluation. In the future, such models could enable MS patients to experience a significant improvement in their lives thanks to data from fitness trackers and smartphones. More

EPFL researchers are targeting the next generation of soft actuators and robots with an elastomer-based ink for 3D printing objects with locally changing mechanical properties, eliminating the need for cumbersome mechanical joints.
For engineers working on soft robotics or wearable devices, keeping things light is a constant challenge: heavier materials require more energy to move around, and — in the case of wearables or prostheses — cause discomfort. Elastomers are synthetic polymers that can be manufactured with a range of mechanical properties, from stiff to stretchy, making them a popular material for such applications. But manufacturing elastomers that can be shaped into complex 3D structures that go from rigid to rubbery has been unfeasible until now.
“Elastomers are usually cast so that their composition cannot be changed in all three dimensions over short length scales. To overcome this problem, we developed DNGEs: 3D-printable double network granular elastomers that can vary their mechanical properties to an unprecedented degree,” says Esther Amstad, head of the Soft Materials Laboratory in EPFL’s School of Engineering.
Eva Baur, a PhD student in Amstad’s lab, used DNGEs to print a prototype ‘finger’, complete with rigid ‘bones’ surrounded by flexible ‘flesh’. The finger was printed to deform in a pre-defined way, demonstrating the technology’s potential to manufacture devices that are sufficiently supple to bend and stretch, while remaining firm enough to manipulate objects.
With these advantages, the researchers believe that DNGEs could facilitate the design of soft actuators, sensors, and wearables free of heavy, bulky mechanical joints. The research has been published in the journal Advanced Materials.
Two elastomeric networks; twice as versatile
The key to the DNGEs’ versatility lies in engineering two elastomeric networks. First, elastomer microparticles are produced from oil-in-water emulsion drops. These microparticles are placed in a precursor solution, where they absorb elastomer compounds and swell up. The swollen microparticles are then used to make a 3D printable ink, which is loaded into a bioprinter to create a desired structure. The precursor is polymerized within the 3D-printed structure, creating a second elastomeric network that rigidifies the entire object.

While the composition of the first network determines the structure’s stiffness, the second determines its fracture toughness, meaning that the two networks can be fine-tuned independently to achieve a combination of stiffness, toughness, and fatigue resistance. The use of elastomers over hydrogels — the material used in state-of-the-art approaches — has the added advantage of creating structures that are water-free, making them more stable over time. To top it off, DNGEs can be printed using commercially available 3D printers.
“The beauty of our approach is that anyone with a standard bioprinter can use it,” Amstad emphasizes.
One exciting potential application of DNGEs is in devices for motion-guided rehabilitation, where the ability to support movement in one direction while restricting it in another could be highly useful. Further development of DNGE technology could result in prosthetics, or even motion guides to assist surgeons. Sensing remote movements, for example in robot-assisted crop harvesting or underwater exploration, is another area of application.
Amstad says that the Soft Materials Lab is already working on the next steps toward developing such applications by integrating active elements — such as responsive materials and electrical connections — into DNGE structures. More

Understanding how proteins interact with each other is crucial for developing new treatments and understanding diseases. Thanks to computational advances, a team of researchers led by Assistant Professor of Chemistry Alberto Perez has developed a groundbreaking algorithm to identify these molecular interactions.
Perez’s research team included two graduate students from UF, Arup Mondal and Bhumika Singh, and a handful of researchers from Rutgers University and Rensselaer Polytechnic Institute. The team published their findings in Angewandte Chemie, a leading chemistry journal based in Germany.
Named the AF-CBA Pipeline, this innovative tool offers unparalleled accuracy and speed in pinpointing the strongest peptide binders to a specific protein. It does this by using AI to simulate molecular interactions, sorting through thousands of candidate molecules to identify the molecule that interacts best with the protein of interest.
The AI-driven approach allows the pipeline to perform these actions in a fraction of the time it would take humans or traditional physics based-approaches to accomplish the same task.
“Think of it like a grocery store,” Perez explained. “When you want to buy the best possible fruit, you have to compare sizes and aspects. There are too many fruits to try them all of course, so you compare a few before making a selection. This AI method, however, can not only try them all, but can also reliably pick out the best one.”
Typically, the proteins of interest are the ones that cause the most damage to our bodies when they misbehave. By finding what molecules interact with these problematic proteins, the pipeline opens avenues for targeted therapies to combat ailments such as inflammation, immune dysregulation, and cancer.
“Knowing the structure of the strongest peptide binder in turn helps us in the rational designing of new drug therapeutics,” Perez said.
The groundbreaking nature of the pipeline is enhanced by its foundation on pre-existing technology: a program called AlphaFold. Developed by Google Deepmind, AlphaFold uses deep learning to predict protein structures. This reliance on familiar technology will be a boon for the pipeline’s accessibility to researchers and will help ensure its future adoption.
Moving forward, Perez and his team aim to expand their pipeline to gain further biological insights and inhibit disease agents. They have two viruses in their sights: murine leukemia virus and Kaposi’s sarcoma virus. Both viruses can cause serious health issues, especially tumors, and interact with as-of-now unknown proteins.
“We want to design novel libraries of peptides,” Perez said. “AF-CBA will allow us to identify those designed peptides that bind stronger than the viral peptides.” More

Soft skin coverings and touch sensors have emerged as a promising feature for robots that are both safer and more intuitive for human interaction, but they are expensive and difficult to make. A recent study demonstrates that soft skin pads doubling as sensors made from thermoplastic urethane can be efficiently manufactured using 3D printers.
“Robotic hardware can involve large forces and torques, so it needs to be made quite safe if it’s going to either directly interact with humans or be used in human environments,” said project lead Joohyung Kim, a professor of electrical & computer engineering at the University of Illinois Urbana-Champaign. “It’s expected that soft skin will play an important role in this regard since it can be used for both mechanical safety compliance and tactile sensing.
As reported in the journal IEEE Transactions on Robotics, the 3D-printed pads function as both soft skin for a robotic arm and pressure-based mechanical sensors. The pads have airtight seals and connect to pressure sensors. Like a squeezed balloon, the pad deforms when it touches something, and the displaced air activates the pressure sensor.
Kim explained, “Tactile robotic sensors usually contain very complicated arrays of electronics and are quite expensive, but we have shown that functional, durable alternatives can be made very cheaply. Moreover, since it’s just a question of reprogramming a 3D printer, the same technique can be easily customized to different robotic systems.”
The researchers demonstrated that this functionality can be naturally used for safety: if the pads detect anything near a dangerous area such as a joint, the arm automatically stops. They can also be used for operational functionality with the robot interpreting touches and taps as instructions.
Since 3D-printed parts are comparatively simple and inexpensive to manufacture, they can be easily adapted to new robotic systems and replaced. Kim noted that this feature is desirable in applications where cleaning and maintaining parts is expensive or infeasible.
“Imagine you want to use soft-skinned robots to assist in a hospital setting,” he said. “They would need to be regularly sanitized, or the skin would need to be regularly replaced. Either way,there’s a huge cost. However, 3D printing is a very scalable process, so interchangeable parts can be inexpensively made and easily snapped on and off the robot body.”
Tactile inputs like the kind provided by the new pads are a relatively unexplored facet of robotic sensing and control. Kim hopes that the ease of this new manufacturing technique will inspire more interest.
“Right now, computer vision and language models are the two major ways that humans can interact with robotic systems, but there is a need for more data on physical interactions, or ‘force-level’ data,” he said. “From the robot’s point of view, this information is the most direct interaction with its environment, but there are very few users — mostly researchers — who think about this. Collecting this force-level data is a target task for me and my group. More