It’s possible that I shall make an ass of myself. But in that case one can always get out of it with a little dialectic. I have, of course, so worded my proposition as to be right either way (K.Marx, Letter to F.Engels on the Indian Mutiny)
Monday, November 25, 2024
First globally broadcast remote robot-assisted laparoscopic hysterectomy
KeAi Communications Co., Ltd.
image:
The surgical team and the Edge Multi-Port Endoscopic Surgical Robot MP1000 (Edge MP1000) surgical system
Telemedicine offers patients in isolated regions the convenience of remote consultation and treatment, alleviating issues related to the uneven distribution and development of medical resources. However, the implementation of remote surgery still faces technical and operational hurdles, primarily revolving around data transmission speed and surgical precision.
In a study published in the in the KeAi journal Intelligent Surgery, researchers from China completed a remote robot-assisted laparoscopic hysterectomy across 1,200 km by using the Edge Multi-Port Endoscopic Surgical Robot MP1000 (Edge MP1000) and 5G communication technology. The surgery, which was broadcast live across the network, demonstrates the feasibility of this technology, and marks a new stage in remote medical technology.
The surgeon was located at the Chinese People's Liberation Army General Hospital in Beijing. Operation instructions were entered using the surgical console, and were then transmitted via the dedicated network data line to the slave console in Wuhan and converted into actual operative actions. The three-dimensional images captured by the endoscope from the slave console served as visual signals, and were transmitted back to the surgeon console's screen in real time via a 5G network dedicated line, forming a closed-loop operation.
Notably, the remote communication host system of the surgical robot was capable of monitoring and recording the surgical process and network status, while providing intelligent assistance functions. There was sufficient interaction between the teams on both sides, allowing smooth communication and tacit cooperation.
“This achieved the specific requirements of low latency, high precision and high reliability for surgical operations,” says Yuanguang Meng, lead researcher of the case report. “Data showed that the bidirectional latency during the surgery was only 19 ms, with a maximum jitter of about 3 ms in rare moments, and a frame drop rate of approximately 0.2%.”
The patient recovered well postoperatively and was discharged on the fourth day with no postoperative complications.
###
Contact the author: Wei Zhang, zw6676@163.com
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 100 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
Feasibility and safety evaluation of ultra-remote robot-assisted laparoscopic hysterectomy
COI Statement
Professor Yuanguang Meng and Wei Zhang are the editorial board members for Intelligent Surgery and were not involved in the editorial review or decision to publish this article. All authors declare that there are no competing interests.
Common way to test for leaks in large language models may be flawed
UVA researchers collaborated to study the effectiveness of membership inference attacks
University of Virginia School of Engineering and Applied Science
image:
This slide shows how a membership inference attack might start. Assessing the product of an app asked to generate an image of a professor teaching students in “the style of” artist Monet could lead to inferences that one of Monet’s bridge paintings assisted the AI’s training.
Large language models are everywhere, including running in the background of the apps on the device you’re using to read this. The auto-complete suggestions in your texts and emails, the query responses composed by Gemni, Copilot and ChatGPT, and the images generated from DALL-E are all built using LLMs.
And they're all trained on real documents and images.
Computer security expert David Evans at the University of Virginia School of Engineering and Applied Science and his colleagues recently reported that a common method that artificial intelligence developers use to test if an LLM’s training data is at risk of exposure doesn’t work as well as once thought.
Presented at the Conference for Language Modeling last month, the paper states in its abstract, “We find that MIAs barely outperform random guessing for most settings across varying LLM sizes and domains.”
What’s an MIA? A Leak?
When creating large language models, developers essentially take a vacuum cleaner approach. They suck up as much text as they can, often from crawling sections of the internet, as well as more private sources, such as email or other data repositories, to train their artificial intelligence applications to understand properties of the world in which they work.
That’s important when it comes to the security of that training data, which could include writing or images millions of internet users posted.
The possibilities for vulnerability, either for content creators or for those who train LLMs, are expansive.
Membership inference attacks, or MIAs, are the primary tool that AI developers use to measure information exposure risks, known as leaks, explained Evans, a professor of computer science who runs the Security Research Group at UVA and a co-author of the research.
Evans and recently graduated Ph.D student Anshuman Suri, the second author on the paper, who is now a postdoctoral researcher at Northeastern University, collaborated with researchers at the University of Washington on the study.
Anshuman Suri, who shared first authorship on the paper, is now a postdoctoral researcher at Northeastern University. The UVA researchers collaborated with researchers at the University of Washington on the study. (Contributed photo)
The study was supported by the National Science Foundation’s Center for Trustworthy Machine Learning.
The main value of a membership inference test on an LLM is as a privacy audit, Evans explained. “It is a way to measure how much information the model is leaking about specific training data.
For example, using adversarial software to assess the product of an app asked to generate an image of a professor teaching students in “the style of” artist Monet could lead to inferences being generated that one of Monet’s bridge paintings assisted the AI’s training.
“An MIA is also used to test if — and if so, by how much — the model has memorized texts verbatim,” Suri added.
Given the potential for legal liability, developers would want to know how solid their foundational pipes are.
This slide shows how a membership inference attack might start. Assessing the product of an app asked to generate an image of a professor teaching students in “the style of” artist Monet could lead to inferences that one of Monet’s bridge paintings assisted the AI’s training. (Contributed)
How Private Is That LLM? How Effective is That MIA?
The researchers performed a large-scale evaluation of five commonly used MIAs. All of the adversarial tools were trained on the popular, open-source language modeling data set known as “the Pile.” A nonprofit research group called EleutherAI released the large language model collection publicly in December 2020.
Microsoft and Meta, along with major universities such as Stanford, have all trained the LLMs of selected applications on the data set.
What’s in the training data? Subsets of data collected from Wikipedia entries, PubMed abstracts, United States Patent and Trademark Office backgrounds, YouTube subtitles, Google DeepMind mathematics and more — representing 22 popular, information-rich web locations in total.
The problem is that language data is not like records for training a traditional model, so it is very difficult to define what a training member is.
The Pile was not filtered based on who gave consent, although researchers can use Eleuther’s tools to refine the model, based on the types of ethical concerns they might have.
“We found that the current methods for conducting membership inference attacks on LLMs are not actually measuring membership inference well, since they suffer from difficulty defining a good representative set of non-member candidates for the experiments,” Evans said.
One reason is that the fluidity of language, as opposed to other types of data, can lead to ambiguity as to what constitutes a member of a dataset.
“The problem is that language data is not like records for training a traditional model, so it is very difficult to define what a training member is,” he said, noting that sentences can have subtle similarities or dramatic differences in meaning based on small changes in word choices.
“It is also very difficult to find candidate non-members that are from the same distribution, and using training time cut-offs for this is error-prone since the actual distribution of language is always changing.”
That’s what has caused past published research showing MIAs as effective to in fact be demonstrating distribution inference instead, Evans and his colleagues assert.
The discrepancy “can be attributed to a distribution shift, e.g., members and non-members are seemingly drawn from identical domain but with different temporal ranges,” the paper states.
Their Python-based, open-source research is now available under an umbrella project called MIMIR, so that other researchers can conduct more revealing membership inference tests.
Worried? Relative Risk Still Low
Evidence so far is that inference risks for individual records in pre-training data is low, but there is no guarantee.
“We expect there is less inference risk for LLMs because of the huge size of the training corpus, and the way training is done, that individual text is often only seen a few times by the model in training,” Evans said.
At the same time, the interactive nature of these types of open source LLMs does open up more avenues that could be used in the future to have stronger attacks.
“We do know, however, that if an adversary uses existing LLMs to train on their own data, known as fine-tuning, their own data is way more susceptible to error than the data seen during the model’s original training phase,” Suri said.
The researchers’ bottom line is that measuring LLM privacy risks is challenging, and the AI community is just beginning to learn how to do it.
Publication Information
“Do Membership Inference Attacks Work on Large Language Models?” by Michael Duan, Anshuman Suri, Niloofar Mireshghallah, Sewon Min, Weijia Shi, Luke Zettlemoyer, Yulia Tsvetkov, Yejin Choi, David Evans and Hannaneh Hajishirzim, was published for peer review July 10, before being accepted for the Conference on Language Modeling, held Oct. 7-9 at the University of Pennsylvania.
Language comprehension impacts medical prescriptions for Ontario's long-term care Francophone, Allophone residents: uOttawa study
Findings in BMC Geriatrics highlight importance of a workforce delivering culturally and linguistically concordant care to avoid inappropriate prescribing of antipsychotics
University of Ottawa
Patients living in linguistically discordant long-term care homes in Ontario are at higher odds of being inappropriately prescribed psychosis medication, says a new University of Ottawa study highlighting the importance of delivering care in the patient’s preferred language.
Researchers from the University of Ottawa’s Department of Family Medicine and the Institut de Savoir Montfort concluded Francophone and allophone residents were more likely to experience inappropriate prescription of antipsychotics compared to English-language patients in a similar setting. Linguistic discordance occurs when care cannot be delivered in a patient’s preferred language.
“The findings of this studynorth_eastexternal link add to the growing body of evidence supporting the notion that language discordance is a social determinant of health contributing to adverse events and poor patient outcomes,” says lead author Dr. Lise Bjerre, who alongside Dr. Peter Tanuseputro, conducted the population-based study of nearly 200,000 long-term care residents in Ontario over the span of nearly a decade.
“The findings highlight the importance of having a diverse workforce capable of delivering culturally and linguistically concordant care.”
According to Dr. Bjerre, the effects of language barriers could be mitigated by:
Asking individuals to specify their preferred language, which could be achieved by having this information on the patient’s health card.
Modifying processes to favour matching of patients to facilities and providers who can provide care in residents’ preferred language.
Ensuring multilingual staff and/or trained interpreters are readily available.
Training staff in French to better serve official language minority communities across the country (such as the program exists at uOttawa’s School of Pharmacy).
“Furthermore, collecting language data at the population level for both patients and health care providers – for example, by including the patient’s preferred language on the health care card, which is done in some provinces (Prince Edward Island) would facilitate evaluating the provision of language-concordant health care and how it relates to patient outcomes in different settings,” adds Dr. Bjerre, the University of Ottawa and Institut du Savoir Montfort Chair in Family.
The impact of patient-facility language discordance on potentially inappropriate prescribing of antipsychotics in long-term care home in Ontario, Canada: a retrospective population health cohort study
Swanson School of Engineering selected to receive $3.3 million to develop new electricity transmission technology
Funding is part of $11 million in funding from Department of Energy for High Voltage Direct Current Transmission Projects
University of Pittsburgh
image:
Part of the Swanson School's one-of-a-kind Electric Power Technologies Lab at the Energy Innovation Center.
The University of Pittsburgh is among four groundbreaking high-voltage direct current (HVDC) transmission research and development projects that are selected to receive a total of $11 million from the U.S. Department of Energy’s (DOE) Office of Electricity (OE) and Office of Renewable Energy and Energy Efficiency (EERE). The awards are part of the Innovative DEsigns for high-performAnce Low-cost HVDC Converters (IDEAL HVDC) funding opportunity.
Pitt’s Swanson School of Engineering will lead a $3.3 million university/industry partnership using artificial intelligence to optimize an HVDC converter design for increased power density and decreased cost.
According to DOE, these projects will help to affordably integrate more renewable energy generation on land or far from shore (e.g., offshore wind) onto the grid via HVDC lines, reduce transmission system costs by 35 percent by 2035, and promote widespread technology adoption. OE is providing $8.1 million in funding and $3 million is coming from EERE.
“This grant presents a great opportunity for us to explore and apply the modern HVDC R&D approach, with artificial intelligence-assisted design, to achieve the most demanding performance metrics while reducing costs,” explained YuAnn Li, assistant professor of electrical and computer engineering at Pitt and a Pitt Co-PI. “AI provides excellent computing capability to flash forward on innovative power converter topologies and control, that previously would not be able to be achieved.”
The IDEAL projects are primed to help reinvent the power grid, which serves as an interstate highway for high-voltage electricity. HVDC transmission systems are more efficient than traditional alternating current (AC) transmission systems to deliver electricity over long distances at a lower cost while minimizing power losses.
“This was a highly competitive program, and our region should be proud to have received this significant support from DOE,” noted Fang Z. Peng, R.K. Mellon professor of electrical and computer engineering, director of Pitt’s Energy GRID Institute and a Pitt Co-PI. “Thanks to the investments in our one-of-a-kind facilities at the EIC, Pitt has become a national leader in HVDC research and development as well as high voltage power electronic systems.”
DOE further explained that many renewable resources are in remote locations on land or planned far from shore (e.g. offshore wind), and HVDC transmission provides a cost-effective solution for renewable integration onto the grid. And high-voltage transmission can more capably transfer power between different regions of the country without disrupting the frequency of either system, also helping to reduce delivery costs.
“Pitt has been a leader in transformative electric power engineering research for more than a century, and technologies like HVDC will take the U.S. and the world in a new direction for safe, efficient, and secure electric power transmission and distribution,” said David Vorp, Senior Associate Dean for Research & Facilities at the Swanson School of Engineering and John A. Swanson Professor of Bioengineering. “The Pitt laboratories at the EIC have evolved into a remarkable site where we can partner with industry, utilities, and academia to develop game-changing power products.”
Other IDEAL HVDC Projects include:
GE Vernova Advanced Research: $3.3 million to develop a low-cost HVDC transmission access point substation to reduce HVDC life cycle costs by >30%. TAPS aims to provide access to affordable renewable energy to underserved and underrepresented communities.
Sandia National Laboratories: $1.8 million to increase the power density and reduce cost of HVDC converter stations by 10% by developing a technology of smaller 1.7 kilovolt (kV) switches that can be operated as a single 10 kV switch in a converter.
Virginia Polytechnic Institute and State University: $3 million to investigate promising circuit technologies to upgrade the existing HVDC converter design. This approach aims to reduce direct material technology costs by 15-20%.
“This represents another step forward in our mission to modernize the nation’s electric grid," said Gene Rodrigues, Assistant Secretary for Electricity. "By investing $11 million in innovative HVDC transmission projects, we're accelerating adoption of an innovative technology that can create pathways to integrate more low- cost renewable energy onto the power grid. This ensures that reliable, resilient, secure and affordable clean energy is available and accessible to all Americans.”
Jeff Marootian, Principal Deputy Assistant Secretary for the Office of Energy Efficiency and Renewable Energy, agreed. He said, “A modern grid requires a transmission network that can offer access to a diverse range of clean energy resources across geographic regions. These investments will help our efforts to improve energy reliability for consumers by better integrating both land and offshore power sources like wind onto the grid.”
###
uOttawa physicists make laser cast a shadow
In a recent study, researchers from the University of Ottawa have demonstrated a remarkable new phenomenon: a laser beam casting a visible shadow
University of Ottawa
image:
“We've shown that under certain conditions, light can actually block other light, creating a shadow”
Jeff Lundeen
— Associate Professor in the Department of Physics at uOttawa
In a recent study, researchers from the University of Ottawa have demonstrated a remarkable new phenomenon: a laser beam casting a visible shadow.
Led by Jeff Lundeen, Associate Professor in the Department of Physics at uOttawa, in collaboration with the Boyd Research Group, this discovery marks the first time such an effect has been observed and challenges our understanding of how light interacts with itself.
Typically, photons - light particles - pass through each other without interacting. However, this experiment revealed an effect in which the shadow of a laser beam shows the same behavior of shadows cast by solid objects.
"We've shown that under certain conditions, light can actually block other light, creating a shadow," explains Professor Lundeen. "This opens up exciting new possibilities for controlling and manipulating light in ways we never thought possible before."
The team's experimental setup involved shining a green laser beam through a ruby crystal while illuminating it from the side with blue light. This arrangement created a shadow on a surface, visible to the naked eye. The effect occurs due to a phenomenon called reverse saturation of absorption in the ruby crystal, which allows the green laser to block the passage of blue light, resulting in a dark region that follows the contours of the laser beam.
"What's particularly fascinating is how closely this laser shadow behaves like a traditional shadow," says Prof. Lundeen. "It follows the shape of the 'object' - in this case, our laser beam - and even conforms to the contours of surfaces it falls on, just like the shadow of a tree branch would."
The researchers developed a theoretical model to predict the shadow's contrast, which closely matched their experimental data. They found that the shadow's darkness increased proportionally with the power of the green laser beam, reaching a maximum contrast of 22% - comparable to a typical shadow on a sunny day.
This discovery expands our understanding of light-matter interactions and holds potential for practical applications. "We're excited about the possibilities this opens up in fields like optical switching, fabrication, and imaging technologies," Prof. Lundeen adds.
The study underscores the importance of fundamental research in reshaping our understanding of the physical world. As scientists continue exploring this phenomenon's implications, it may lead to new advancements in photonics, non-linear optics, and other light-based technologies.
Surprising finding could lead to new ways of controlling light
Optica
image:
For the experiment, a high-power green laser was directed through ruby cube and illuminated with a blue laser from the side. The green laser increases the optical absorption of the blue illuminating laser beam, creating a matching region in the illuminating light and creating a darker area that appears as a shadow of the green laser beam.
WASHINGTON — Can light itself cast a shadow? It may sound like a philosophical riddle, but researchers have found that under certain conditions, a laser beam can act like an opaque object and cast a shadow. The discovery challenges the traditional understanding of shadows and opens new possibilities for technologies that could use a laser beam to control another laser beam.
“Laser light casting a shadow was previously thought impossible since light usually passes through other light without interacting,” said research team leader Raphael A. Abrahao from Brookhaven National Laboratory, previously at the University of Ottawa. “Our demonstration of a very counter-intuitive optical effect invites us to reconsider our notion of shadow.”
In Optica, Optica Publishing Group’s journal for high-impact research, the researchers describe how they used a ruby crystal and specific laser wavelengths to show that a laser beam could block light and create a visible shadow due to a nonlinear optical process. This effect occurs when light interacts with a material in an intensity-dependent way and can influence another optical field.
“Our understanding of shadows has developed hand-in-hand with our understanding of light and optics,” said Abrahao. “This new finding could prove useful in various applications such as optical switching, devices in which light controls the presence of another light, or technologies that require precise control of light transmission, like high-power lasers."
Lunch talk sparks idea
The new research is part of a larger exploration into how a light beam interacts with another light beam under special conditions and nonlinear optical processes. The idea started over a lunch conversation when it was pointed out that some experimental schematics made with 3D visualization software depict the shadow of a laser beam because they treat it as a cylinder without accounting for the physics of a laser beam. Some of the scientists wondered: Could this be done in a lab?
“What started as a funny discussion over lunch led to a conversation on the physics of lasers and the nonlinear optical response of materials,” said Abrahao. “From there, we decided to conduct an experiment to demonstrate the shadow of a laser beam.”
To do this, the researchers directed a high-power green laser through a cube made of standard ruby crystal and illuminated it with a blue laser from the side. When the green laser enters the ruby, it locally changes the material response to the blue wavelength. The green laser acts like an ordinary object while the blue laser acts like illumination.
The interaction between the two light sources created a shadow on a screen that was visible as a dark area where the green laser blocked the blue light. It met all the criteria for a shadow because it was visible to the naked eye, followed the contours of the surface it fell on and followed the position and shape of the laser beam, which acted as an object.
The laser shadow effect is a consequence of optical nonlinear absorption in the ruby. The effect occurs because the green laser increases the optical absorption of the blue illuminating laser beam, creating a matching region in the illuminating light with lower optical intensity. The result is a darker area that appears as a shadow of the green laser beam.
Shadow measurements
“This discovery expands our understanding of light-matter interactions and opens up new possibilities for utilizing light in ways we hadn’t considered before,” said Abrahao.
The researchers experimentally measured the dependence of the shadow's contrast on the laser beam's power, finding a maximum contrast of approximately 22%, similar to the contrast of a tree's shadow on a sunny day. They also developed a theoretical model and showed that it could accurately predict the shadow contrast.
The researchers say that from a technological perspective, the effect they demonstrated shows that the intensity of a transmitted laser beam can be controlled by applying another laser. Next, they plan to investigate other materials and other laser wavelengths that can produce similar effects.
Optica is an open-access journal dedicated to the rapid dissemination of high-impact peer-reviewed research across the entire spectrum of optics and photonics. Published monthly by Optica Publishing Group, the Journal provides a forum for pioneering research to be swiftly accessed by the international community, whether that research is theoretical or experimental, fundamental or applied. Optica maintains a distinguished editorial board of more than 60 associate editors from around the world and is overseen by Editor-in-Chief Prem Kumar, Northwestern University, USA. For more information, visit Optica.
Researchers showed that a laser beam can sometimes act like a solid object and cast a shadow that is visible to the naked eye. In the picture, the shadow appears as the horizontal line traversing the blue background.
Optica Publishing Group is a division of Optica, the society advancing optics and photonics worldwide. It publishes the largest collection of peer-reviewed content in optics and photonics, including 18 prestigious journals, the society’s flagship member magazine, and papers from more than 835 conferences, including 6,500+ associated videos. With over 400,000 journal articles, conference papers and videos to search, discover and access, Optica Publishing Group represents the full range of research in the field from around the globe.
