̽��ֱ�� of Cambridge - Cornell ̽��ֱ��

Machine learning models can produce reliable results even with limited training data

sc604 — Tue, 19 Sep 2023 10:03:16 +0000

̽��ֱ��researchers, from the ̽��ֱ�� of Cambridge and Cornell ̽��ֱ��, found that for partial differential equations – a class of physics equations that describe how things in the natural world evolve in space and time – machine learning models can produce reliable results even when they are provided with limited data.

Their results, reported in the Proceedings of the National Academy of Sciences, could be useful for constructing more time- and cost-efficient machine learning models for applications such as engineering and climate modelling.

Most machine learning models require large amounts of training data before they can begin returning accurate results. Traditionally, a human will annotate a large volume of data – such as a set of images, for example – to train the model.

“Using humans to train machine learning models is effective, but it’s also time-consuming and expensive,” said first author Dr Nicolas Boullé, from the Isaac Newton Institute for Mathematical Sciences. “We’re interested to know exactly how little data we actually need to train these models and still get reliable results.”

Other researchers have been able to train machine learning models with a small amount of data and get excellent results, but how this was achieved has not been well-explained. For their study, Boullé and his co-authors, Diana Halikias and Alex Townsend from Cornell ̽��ֱ��, focused on partial differential equations (PDEs).

“PDEs are like the building blocks of physics: they can help explain the physical laws of nature, such as how the steady state is held in a melting block of ice,” said Boullé, who is an INI-Simons Foundation Postdoctoral Fellow. “Since they are relatively simple models, we might be able to use them to make some generalisations about why these AI techniques have been so successful in physics.”

̽��ֱ��researchers found that PDEs that model diffusion have a structure that is useful for designing AI models. “Using a simple model, you might be able to enforce some of the physics that you already know into the training data set to get better accuracy and performance,” said Boullé.

̽��ֱ��researchers constructed an efficient algorithm for predicting the solutions of PDEs under different conditions by exploiting the short and long-range interactions happening. This allowed them to build some mathematical guarantees into the model and determine exactly how much training data was required to end up with a robust model.

“It depends on the field, but for physics, we found that you can actually do a lot with a very limited amount of data,” said Boullé. “It’s surprising how little data you need to end up with a reliable model. Thanks to the mathematics of these equations, we can exploit their structure to make the models more efficient.”

̽��ֱ��researchers say that their techniques will allow data scientists to open the ‘black box’ of many machine learning models and design new ones that can be interpreted by humans, although future research is still needed.

“We need to make sure that models are learning the right things, but machine learning for physics is an exciting field – there are lots of interesting maths and physics questions that AI can help us answer,” said Boullé.

Reference

Nicolas Boullé, Diana Halikias, and Alex Townsend. ‘Elliptic PDE learning is provably data-efficient.’ PNAS (2023). DOI: 10.1073/pnas.2303904120

Researchers have determined how to build reliable machine learning models that can understand complex equations in real-world situations while using far less training data than is normally expected.

It’s surprising how little data you need to end up with a reliable model

Nicolas Boullé

Andriy Onufriyenko via Getty Images

Digital generated image of multi coloured glowing data over landscape.

̽��ֱ��text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright © ̽��ֱ�� of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways – as here, on our main website under its Terms and conditions, and on a range of channels including social media that permit your use and sharing of our content under their respective Terms.

Yes

Tree-dwelling mammals survived after asteroid strike destroyed forests

cmm201 — Thu, 14 Oct 2021 12:10:32 +0000

Overall, the study supports the hypothesis that the widespread destruction of forests following the asteroid’s impact favoured ground-dwelling mammals over their arboreal counterparts. However, it also provides strong evidence that some tree-dwelling animals also survived the cataclysm, possibly nesting in branches through the Cretaceous-Paleogene (K-Pg) extinction event.

“ ̽��ֱ��recovery of terrestrial vertebrate life following the end-Cretaceous asteroid impact was one of the most important events in the history of life on Earth,” said senior author Dr Daniel Field, from Cambridge’s Department of Earth Sciences. “In this study, we drew on our previous work at Cambridge to investigate whether there were similarities in the ecological attributes of avian and mammalian survivors of the end-Cretaceous mass extinction.”

̽��ֱ��K-Pg mass extinction event occurred when a meteor slammed into Earth at the end of the Cretaceous period. ̽��ֱ��impact and its aftereffects killed roughly 75% of the animal and plant species on the planet, including whole groups like the non-avian dinosaurs.

For the study, the researchers analysed patterns of substrate preferences among all modern mammals and their ancestors, working backwards from the present day to before the K-Pg extinction event by tracing these traits along numerous phylogenetic trees — diagrams that illustrate the evolutionary relationships among species.

“Our study takes advantage of an ongoing revolution in our understanding of the tree of life, only made possible by researchers working in association with natural history collections,” said co-lead author Jacob Berv, from the ̽��ֱ�� of Michigan, USA. “By integrating data from such collections with modern statistical techniques, we can address new questions about major transitions in evolutionary history.”

̽��ֱ��researchers classified each mammalian species as arboreal, non-arboreal, or semi-arboreal. To be considered arboreal, the species had to nearly always nest in trees. Categorising some species could be tricky. For example, many bat species spend a lot of time among trees but nest in caves, so bats were mostly categorised as non-arboreal or semi-arboreal.

“We were able to see that leading up to the K-Pg event, there was a spike in transitions from arboreal and semi-arboreal to non-arboreal habitat use across our models,” said co-lead author Jonathan Hughes, from Cornell ̽��ֱ��, USA.

̽��ֱ��work builds on a previous study led by Field, which used the same analytical method — known as ancestral state reconstruction — to show that all modern birds evolved from ground-dwelling ancestors after the asteroid strike.

“ ̽��ֱ��fossil record of many vertebrate groups is sparse in the immediate aftermath of the extinction,” said Field, who is also curator of ornithology at the Cambridge Museum of Zoology. “Analytical approaches like ancestral state reconstruction allow us to establish hypotheses for how groups like birds and mammals made it through this cataclysm, which palaeontologists can then test when additional fossils are found.”

̽��ֱ��analysis helps illuminate ecological selectivity of mammals across the K-Pg boundary despite the relatively sparse fossil record of mammalian skeletal elements from the periods immediately preceding and following the asteroid’s impact.

How the tree-dwelling ancestors of primates survived the asteroid’s destruction is unclear. According to the authors, it’s possible that some forest fragments survived the calamity or that early primates and their relatives were ecologically flexible enough to modify their substrate preferences in a world mostly denuded of trees.

Reference:
Jonathan J Hughes et al. ‘Ecological selectivity and the evolution of mammalian substrate preference across the K–Pg boundary.’ Ecology and Evolution (2021). DOI: 10.1002/ece3.8114

Adapted from a Yale news release.

An asteroid strike 66 million years ago wiped out the non-avian dinosaurs and devastated the Earth’s forests, but tree-dwelling ancestors of primates may have survived it, according to a new study published in the journal Ecology and Evolution.

̽��ֱ��recovery of terrestrial vertebrate life following the end-Cretaceous asteroid impact was one of the most important events in the history of life on Earth

Daniel Field

Rod Waddington

Chimpanzee, Uganda

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. Images, including our videos, are Copyright © ̽��ֱ�� of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways – as here, on our main website under its Terms and conditions, and on a range of channels including social media that permit your use and sharing of our content under their respective Terms.

Yes

Licence type:

Attribution-ShareAlike

Cell ‘membrane on a chip’ could speed up screening of drug candidates for COVID-19

sc604 — Mon, 06 Jul 2020 07:57:17 +0000

̽��ֱ��researchers, from the ̽��ֱ�� of Cambridge, Cornell ̽��ֱ�� and Stanford ̽��ֱ��, say their device could mimic any cell type--bacterial, human or even the tough cells walls of plants. Their research recently pivoted to how COVID-19 attacks human cell membranes and, more importantly, how it can be blocked.

̽��ֱ��devices have been formed on chips while preserving the orientation and functionality of the cell membrane and have been successfully used to monitor the activity of ion channels, a class of protein in human cells which are the target of more than 60% of approved pharmaceuticals. ̽��ֱ��results are published in two recent papers in Langmuir and ACS Nano.

Cell membranes play a central role in biological signalling, controlling everything from pain relief to infection by a virus, acting as the gatekeeper between a cell and the outside world. ̽��ֱ��team set out to create a sensor that preserves all of the critical aspects of a cell membrane—structure, fluidity, and control over ion movement—without the time-consuming steps needed to keep a cell alive.

̽��ֱ��device uses an electronic chip to measure any changes in an overlying membrane extracted from a cell, enabling the scientists to safely and easily understand how the cell interacts with the outside world.

̽��ֱ��device integrates cell membranes with conducting polymer electrodes and transistors. To generate the on-chip membranes, the Cornell team first optimised a process to produce membranes from live cells and then, working with the Cambridge team, coaxed them onto polymeric electrodes in a way that preserved all of their functionality. ̽��ֱ��hydrated conducting polymers provide a more ‘natural’ environment for cell membranes and allows robust monitoring of membrane function.

̽��ֱ��Stanford team optimised the polymeric electrodes for monitoring changes in the membranes. ̽��ֱ��device no longer relies on live cells that are often technically challenging to keep alive and require significant attention, and measurements can last over an extended time period.

“Because the membranes are produced from human cells, it’s like having a biopsy of that cell’s surface - we have all the material that would be present including proteins and lipids, but none of the challenges of using live cells,” said Dr Susan Daniel, associate professor of chemical and biomolecular engineering at Cornell and senior author of the ACS Langmuir paper.

“This type of screening is typically done by the pharmaceutical industry with live cells, but our device provides an easier alternative,” said Dr Róisín Owens from Cambridge’s Department of Chemical Engineering and Biotechnology, and senior author of the ACS Nano paper. “This method is compatible with high-throughput screening and would reduce the number of false positives making it through into the R&D pipeline.”

“ ̽��ֱ��device can be as small as the size of a human cell and easily fabricated in arrays, which allows us to perform multiple measurements at the same time,” said Dr Anna-Maria Pappa, also from Cambridge and joint first author on both papers.

To date, the aim of the research, supported by funding from the United States Defense Research Projects Agency (DARPA), has been to demonstrate how viruses such as influenza interact with cells. Now, DARPA has provided additional funding to test the device’s effectiveness in screening for potential drug candidates for COVID-19 in a safe and effective way.

Given the significant risks involved to researchers working on SARS-CoV-2, the virus which causes COVID-19, scientists on the project will focus on making virus membranes and fusing those with the chips. ̽��ֱ��virus membranes are identical to the SARS-CoV-2 membrane but don’t contain the viral nucleic acid. This way new drugs or antibodies to neutralise the virus spikes that are used to gain entry into the host cell can be identified. This work is expected to get underway on 1 August.

“With this device, we are not exposed to risky working environments for combating SARS-CoV-2. ̽��ֱ��device will speed up the screening of drug candidates and provide answers to questions about how this virus works,” said Dr Han-Yuan Liu, Cornell researcher and joint first author on both papers.

Future work will focus on scaling up production of the devices at Stanford and automating the integration of the membranes with the chips, leveraging the fluidics expertise from Stanford PI Juan Santiago who will join the team in August.

“This project has merged ideas and concepts from laboratories in the UK, California and New York, and shown a device that works reproducibly in all three sites. It is a great example of the power of integrating biology and materials science in addressing global problems,” said Stanford lead PI Professor Alberto Salleo.

References:
H-Y Liu et al. “Self-assembly of mammalian cell membranes on bioelectronic devices with functional transmembrane proteins.” ACS Langmuir (2020). DOI: 10.1021/acs.langmuir.0c00804

A-M. Pappa et al. “Optical and Electronic Ion Channel Monitoring from Native Human Membranes.” ACS Nano (2020). DOI: 10.1021/acsnano.0c01330

Researchers have developed a human cell ‘membrane on a chip’ that allows continuous monitoring of how drugs and infectious agents interact with our cells, and may soon be used to test potential drug candidates for COVID-19.

This type of screening is typically done by the pharmaceutical industry with live cells, but our device provides an easier alternative

Róisín Owens

Susan Daniel/Cornell ̽��ֱ��

Schematic of membrane on a chip device

Yes

Science Live

dwc34 — Mon, 22 Dec 2014 14:52:07 +0000

From intimate science cafes to massive science festivals, the public science events sector encompasses an enormous diversity of activity involving a huge range of practitioners and target audiences. As unique as each event is, public science events are all live, in-person programs, designed to engage the public with science in a social context that is at least as meaningful as the content and messages delivered. This activity is already taking place on a grand scale in both the US and UK, and having a direct impact in a variety of ways. ̽��ֱ��Science Live project is a first significant step forward in a long-term effort to widen access to the beneficial impacts of public science events, better understand these impacts and how they are produced, and uncover new opportunities for engaging society with science.

Researchers at Cornell ̽��ֱ�� and the ̽��ֱ�� of the West of England (UWE), Bristol will work with science festival managers at MIT and the ̽��ֱ�� of Cambridge to tackle several issues across the science events sectors in the UK and US. Ben Wiehe from the Science Festival Alliance office at the MIT Museum said:

“We know that millions of people are participating in live public science events every year in both the US and UK, but the contributions of these events are only just coming to be acknowledged as distinct, and there is little overall tracking of or advocacy for such activity. ̽��ֱ��organizers of these events are often not aware of each other’s efforts. We have a lot to learn from each other, but right now there is no straightforward way to share findings with the full range of event practitioners.”

̽��ֱ��Science Live pilot study is funded by the Science Learning+ initiative of the Wellcome Trust, the US-based National Science Foundation, and the UK-based Economic and Social Research Council. It aims to lead to a longer term effort to establish a set of key facts related to live science activities, and use those facts to build a coherent narrative explaining the role that live events play in the larger science learning ecosystem.

Laura Fogg Rogers, Science Communication Research Fellow at UWE said:

“This is an exciting time to drill down and find out what works best in communicating research to benefit society. This consortium has access to some of the best networks of public science events, including science festival networks in the UK and US, grass-roots activities, and government level initiatives. ̽��ֱ��potential to make an impact on the researcher and practitioner landscape is massive.”

NOTES FOR EDITORS

Science Live is a collaborative project of the MIT Museum and the ̽��ֱ�� of Cambridge, administered by the Science Festival Alliance office at the MIT Museum, and is led by Dr John Durant from Massachusetts Institute of Technology. Research partners include

Professor Bruce Lewenstein from Cornell ̽��ֱ��, Laura Fogg Rogers from ̽��ֱ�� of the West of England, and Dr Wai Yi Feng from the ̽��ֱ�� of Cambridge.

Science Live is funded through the Science Learning+ scheme, an international initiative established by the Wellcome Trust, the US National Science Foundation (NSF), and the UK Economic and Social Research Council (ESRC) and in collaboration with the MacArthur Foundation, the Gordon and Betty Moore Foundation, and the Noyce Foundation. ̽��ֱ��scheme was launched in April 2014 after a review of informal learning commissioned by the Wellcome Trust indicated a need for further research to be undertaken in this area.

̽��ֱ��UK contact is Dane Comerford, from the ̽��ֱ�� of Cambridge Public Engagement Team and the US contact is Ben Wiehe from the Science Festival Alliance office at the MIT Museum.

Dane Comerford

Public Engagement Manager, ̽��ֱ�� of Cambridge

Phone: +44 (0)1223 764069

E-mail: dane.comerford@admin.cam.ac.uk

Ben Wiehe

Manager, Science Festival Alliance

Phone: +1 617.806.6369

E-mail: wiehe@mit.edu

A new transatlantic pilot study aims to better understand what makes science events tick.

These activities range from festivals, science busking and trips to nature reserves, and involve about half of the UK population. ̽��ֱ��project, called Science Live, will explore the differences between the huge varieties of live science events and will develop research questions about how they affect their audiences.

This consortium has access to some of the best networks of public science events, including science festival networks in the UK and US, grass-roots activities, and government level initiatives. ̽��ֱ��potential to make an impact on the researcher and practitioner landscape is massive.

Laura Fogg Rogers, Science Communication Research Fellow at UWE, Bristol

Around half the UK population access science activities every year.

̽��ֱ��text in this work is licensed under a Creative Commons Licence. If you use this content on your site please link back to this page. For image rights, please see the credits associated with each individual image.

Yes