探花直播 of Cambridge - Princeton 探花直播

New datasets will train AI models to think like scientists

sc604 — Mon, 02 Dec 2024 15:59:08 +0000

探花直播initiative, called Polymathic AI, uses technology like that powering large language models such as OpenAI鈥檚 ChatGPT or Google鈥檚 Gemini. But instead of ingesting text, the project鈥檚 models learn using scientific datasets from across astrophysics, biology, acoustics, chemistry, fluid dynamics and more, essentially giving the models cross-disciplinary scientific knowledge.

鈥淭hese datasets are by far the most diverse large-scale collections of high-quality data for machine learning training ever assembled for these fields,鈥� said team member Michael McCabe from the Flatiron Institute in New York City. 鈥淐urating these datasets is a critical step in creating multidisciplinary AI models that will enable new discoveries about our universe.鈥�

On 2 December, the Polymathic AI team released two of its open-source training dataset collections to the public 鈥� a colossal 115 terabytes, from dozens of sources 鈥� for the scientific community to use to train AI models and enable new scientific discoveries. For comparison, GPT-3 used 45 terabytes of uncompressed, unformatted text for training, which ended up being around 0.5 terabytes after filtering.

探花直播full datasets are available to download for free on HuggingFace, a platform hosting AI models and datasets. 探花直播Polymathic AI team provides further information about the datasets in two papers accepted for presentation at the NeurIPS machine learning conference, to be held later this month in Vancouver, Canada.

鈥淛ust as LLMs such as ChatGPT learn to use common grammatical structure across languages, these new scientific foundation models might reveal deep connections across disciplines that we鈥檝e never noticed before,鈥� said Cambridge team lead聽Dr Miles Cranmer from Cambridge鈥檚 Institute of Astronomy. 鈥淲e might uncover patterns that no human can see, simply because no one has ever had both this breadth of scientific knowledge and the ability to compress it into a single framework.鈥�

AI tools such as machine learning are increasingly common in scientific research, and were recognised in two of this year鈥檚 Nobel Prizes. Still, such tools are typically purpose-built for a specific application and trained using data from that field. 探花直播Polymathic AI project instead aims to develop models that are truly polymathic, like people whose expert knowledge spans multiple areas. 探花直播project鈥檚 team reflects intellectual diversity, with physicists, astrophysicists, mathematicians, computer scientists and neuroscientists.

探花直播first of the two new training dataset collections focuses on astrophysics. Dubbed the Multimodal Universe, the dataset contains hundreds of millions of astronomical observations and measurements, such as portraits of galaxies taken by NASA鈥檚 James Webb Space Telescope and measurements of our galaxy鈥檚 stars made by the European Space Agency鈥檚 Gaia spacecraft.

探花直播other collection 鈥� called the Well 鈥� comprises over 15 terabytes of data from 16 diverse datasets. These datasets contain numerical simulations of biological systems, fluid dynamics, acoustic scattering, supernova explosions and other complicated processes.聽Cambridge researchers played a major role in developing both dataset collections, working alongside PolymathicAI and other international collaborators.

While these diverse datasets may seem disconnected at first, they all require the modelling of mathematical equations called partial differential equations. Such equations pop up in problems related to everything from quantum mechanics to embryo development and can be incredibly difficult to solve, even for supercomputers. One of the goals of the Well is to enable AI models to churn out approximate solutions to these equations quickly and accurately.

鈥淏y uniting these rich datasets, we can drive advancements in artificial intelligence not only for scientific discovery, but also for addressing similar problems in everyday life,鈥� said Ben Boyd, PhD student in the Institute of Astronomy.

Gathering the data for those datasets posed a challenge, said team member Ruben Ohana from the Flatiron Institute. 探花直播team collaborated with scientists to gather and create data for the project. 鈥� 探花直播creators of numerical simulations are sometimes sceptical of machine learning because of all the hype, but they鈥檙e curious about it and how it can benefit their research and accelerate scientific discovery,鈥� he said.

探花直播Polymathic AI team is now using the datasets to train AI models. In the coming months, they will deploy these models on various tasks to see how successful these well-rounded, well-trained AIs are at tackling complex scientific problems.

鈥淚t will be exciting to see if the complexity of these datasets can push AI models to go beyond merely recognising patterns, encouraging them to reason and generalise across scientific domains,鈥� said Dr Payel Mukhopadhyay from the Institute of Astronomy. 鈥淪uch generalisation is essential if we ever want to build AI models that can truly assist in conducting meaningful science.鈥�

鈥淯ntil now, haven鈥檛 had a curated scientific-quality dataset cover such a wide variety of fields,鈥� said Cranmer, who is also a member of Cambridge鈥檚 Department of Applied Mathematics and Theoretical Physics. 鈥淭hese datasets are opening the door to true generalist scientific foundation models for the first time. What new scientific principles might we discover? We're about to find out, and that's incredibly exciting.鈥�

探花直播Polymathic AI project is run by researchers from the Simons Foundation and its Flatiron Institute, New York 探花直播, the 探花直播 of Cambridge, Princeton 探花直播, the French Centre National de la Recherche Scientifique and the Lawrence Berkeley National Laboratory.

Members of the Polymathic AI team from the 探花直播 of Cambridge include PhD students, postdoctoral researchers and faculty across four departments: the Department of Applied Mathematics and Theoretical Physics, the Department of Pure Mathematics and Mathematical Statistics, the Institute of Astronomy and the Kavli Institute for Cosmology.

What can exploding stars teach us about how blood flows through an artery? Or swimming bacteria about how the ocean鈥檚 layers mix? A collaboration of researchers, including from the 探花直播 of Cambridge, has reached a milestone toward training artificial intelligence models to find and use transferable knowledge between fields to drive scientific discovery.

Alex Meng, Aaron Watters and the Well Collaboration

A mosaic of simulations included in the Well collection of datasets

探花直播text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright 漏探花直播 of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways 鈥� 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

First map of every neuron in an adult fly brain complete

jg533 — Wed, 02 Oct 2024 15:04:57 +0000

This landmark achievement has been conducted by the FlyWire Consortium, a large international collaboration including researchers from the 探花直播 of Cambridge, the MRC Laboratory of Molecular Biology in Cambridge, Princeton 探花直播, and the 探花直播 of Vermont. It is published today in two papers in the journal Nature.

探花直播diagram of all 139,255 neurons in the adult fly brain is the first of an entire brain for an animal that can walk and see. Previous efforts have completed the whole brain diagrams for much smaller brains, for example a fruit fly larva which has 3,016 neurons, and a nematode worm which has 302 neurons.

探花直播researchers say the whole fly brain map is a key first step to completing larger brains. Since the fruit fly is a common tool in research, its brain map can be used to advance our understanding of how neural circuits work.

Dr Gregory Jefferis, from the 探花直播 of Cambridge and the MRC Laboratory of Molecular Biology, one of the co-leaders of the research, said: 鈥淚f we want to understand how the brain works, we need a mechanistic understanding of how all the neurons fit together and let you think. For most brains we have no idea how these networks function.聽

鈥淔lies can do all kinds of complicated things like walk, fly, navigate, and the males sing to the females. Brain wiring diagrams are a first step towards understanding everything we鈥檙e interested in 鈥� how we control our movement, answer the telephone, or recognise a friend.鈥�

Dr Mala Murthy from Princeton 探花直播, one of the co-leaders of the research, said: 鈥淲e have made the entire database open and freely available to all researchers. We hope this will be transformative for neuroscientists trying to better understand how a healthy brain works. In the future we hope that it will be possible to compare what happens when things go wrong in our brains, for example in mental health conditions.鈥澛�

Dr Marta Costa from the 探花直播 of Cambridge, who was also involved in the research, said 鈥淭his brain map, the biggest so far, has only been possible thanks to technical advances that didn鈥檛 seem possible ten years ago. It is a true testament to the way that innovation can drive research forward. 探花直播next steps will be to generate even bigger maps, such as a mouse brain, and ultimately, a human one.鈥�

探花直播scientists found that there were substantial similarities between the wiring in this map and previous smaller-scale efforts to map out parts of the fly brain. This led the researchers to conclude that there are many similarities in wiring between individual brains 鈥� that each brain isn鈥檛 a unique structure.

When comparing their brain diagram to previous diagrams of small areas of the brain, the researchers also found that about 0.5% of neurons have developmental variations that could cause connections between neurons to be mis-wired. 探花直播researchers say it will be important聽to understand, through future research, if these changes are linked to individuality or brain disorders.聽

Making the map

3D rendering of all 140,000 neurons in the fruit fly brain.

3D rendering of all ~140k neurons in the fruit fly brain. Credit: Data source FlyWire.ai; Rendering by Philipp Schlegel ( 探花直播 of Cambridge/MRC LMB).

A whole fly brain is less than one聽millimetre wide. 探花直播researchers started with one female brain cut into seven thousand slices, each only 40 nanometres thick, that were previously scanned using high resolution electron microscopy in the laboratory of project co-leader Davi Bock at聽Janelia Research Campus in the US.

Analysing over 100 terabytes of image data (equivalent to the storage in 100 typical laptops) to extract the shapes of about 140,000 neurons and 50 million connections between them is too big a challenge for humans to complete manually. 探花直播researchers built on AI developed at Princeton 探花直播 to identify and map neurons and their connections to each other.

However, the AI still makes many errors in datasets of this size. 探花直播Princeton 探花直播 researchers established the FlyWire Consortium 鈥� made up of teams in more than 76 laboratories and 287 researchers around the world, as well as volunteers from the general public 鈥� which spent an estimated 33 person-years painstakingly proofreading all the data.

Dr Sebastian Seung, from Princeton 探花直播, who was one of the co-leaders of the research, said: 鈥淢apping the whole brain has been made possible by advances in AI computing - it would have not been possible to reconstruct the entire wiring diagram manually. This is a display of how AI can move neuroscience forward. 探花直播fly brain is a milestone on our way to reconstructing a wiring diagram of a whole mouse brain.鈥�

探花直播researchers also annotated many details on the wiring diagram, such as classifying more than 8,000 cell types across the brain. This allows researchers to select particular systems within the brain for further study, such as the neurons involved in sight or movement.聽

Dr Philipp Schlegel, the first author of one of the studies, from the MRC Laboratory of Molecular Biology, said: 鈥淭his dataset is a bit like Google Maps but for brains: the raw wiring diagram between neurons is like knowing which structures on satellite images of the Earth correspond to streets and buildings. Annotating neurons is like adding the names for streets and towns, business opening times, phone numbers and reviews聽to the map 鈥� you need both for it to be really useful.鈥�

Simulating brain function

This is also the first whole brain wiring map 鈥� often called a connectome 鈥� to predict the function of all the connections between neurons.聽

Neurons use electrical signals to send messages. Each neuron can have hundreds of branches that connect it to other neurons. 探花直播points where these branches meet and transmit signals between neurons are called synapses. There are two main ways that neurons communicate across synapses: excitatory (which promotes the continuation of the electrical signal in the receiving neuron), or inhibitory (which reduces the likelihood that the next neuron will transmit signals).

Researchers from the team used AI image scanning technology to predict whether each synapse was inhibitory or excitatory.

Dr Gregory Jefferis added: 鈥淭o begin to simulate the brain digitally, we need to know not only the structure of the brain, but also how the neurons function to turn each other on and off.鈥�

鈥淯sing our data, which has been shared online as we worked, other scientists have already started trying to simulate how the fly brain responds to the outside world. This is an important start, but we will need to collect many different kinds of data to produce reliable simulations of how a brain functions.鈥�

Associate Professor Davi Bock, one of the co-leaders of the research from the 探花直播 of Vermont, said: 鈥� 探花直播hyper-detail of electron microscopy data creates its own challenges, especially at scale. This team wrote sophisticated software algorithms to identify patterns of cell structure and connectivity within all that detail.聽

鈥淲e now can make precise synaptic level maps and use these to better understand cell types and circuit structure at whole-brain scale. This will inevitably lead to a deeper understanding of how nervous systems process, store and recall information. I think this approach points the way forward for the analysis of future whole-brain connectomes, in the fly as well as in other species."

This research was conducted using a female fly brain. Since there are differences in neuronal structure between male and female fly brains, the researchers also plan to聽characterise a male brain in the future.聽

探花直播principal funders were the National Institutes of Health BRAIN Initiative, Wellcome, Medical Research Council, Princeton 探花直播 and National Science Foundation.

References

Schlegel, P. et al: Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature, Oct 2024. DOI:聽10.1038/s41586-024-07686-5

Dorkenwald, S. et al: Neuronal wiring diagram of an adult brain. Nature, Oct 2024. DOI: 10.1038/s41586-024-07558-y

聽

探花直播first wiring diagram of every neuron in an adult brain and the 50 million connections between them has been produced for a fruit fly.

Brain wiring diagrams are a first step towards understanding everything we鈥檙e interested in 鈥� how we control our movement, answer the telephone, or recognise a friend.

Gregory Jefferis

探花直播first complete map of every neuron in an adult fly brain.

FlyWire.ai; Rendering by Philipp Schlegel ( 探花直播 of Cambridge/MRC LMB).

3D rendering of all 140,000 neurons in the adult fruit fly brain.

Yes

Licence type:

Attribution-Noncommerical

Pythagoras was wrong: there are no universal musical harmonies, study finds

ta385 — Tue, 27 Feb 2024 09:30:00 +0000

According to the Ancient Greek philosopher Pythagoras, 鈥榗onsonance鈥� 鈥� a pleasant-sounding combination of notes 鈥� is produced by special relationships between simple numbers such as 3 and 4. More recently, scholars have tried to find psychological explanations, but these 鈥榠nteger ratios鈥� are still credited with making a chord sound beautiful, and deviation from them is thought to make music 鈥榙issonant鈥�, unpleasant sounding.聽

But researchers from the 探花直播 of Cambridge, Princeton and the Max Planck Institute for Empirical Aesthetics, have now discovered two key ways in which Pythagoras was wrong.

Their study, published in Nature Communications, shows that in normal listening contexts, we do not actually prefer chords to be perfectly in these mathematical ratios.

鈥淲e prefer slight amounts of deviation. We like a little imperfection because this gives life to the sounds, and that is attractive to us,鈥� said co-author, Dr Peter Harrison, from Cambridge鈥檚 Faculty of Music and Director of its Centre for Music and Science.

探花直播researchers also found that the role played by these mathematical relationships disappears when you consider certain musical instruments that are less familiar to Western musicians, audiences and scholars. These instruments tend to be bells, gongs, types of xylophones and other kinds of pitched percussion instruments. In particular, they studied the 鈥榖onang鈥�, an instrument from the Javanese gamelan built from a collection of small gongs.

鈥淲hen we use instruments like the bonang, Pythagoras's special numbers go out the window and we encounter entirely new patterns of consonance and dissonance,鈥� said Dr Harrison, a Fellow of Churchill College.

鈥� 探花直播shape of some percussion instruments means that when you hit them, and they resonate, their frequency components don鈥檛 respect those traditional mathematical relationships. That's when we find interesting things happening.鈥�

鈥淲estern research has focused so much on familiar orchestral instruments, but other musical cultures use instruments that, because of their shape and physics, are what we would call 鈥榠nharmonic鈥�.聽

探花直播researchers created an online laboratory in which over 4,000 people from the US and South Korea participated in 23 behavioural experiments. Participants were played chords and invited to give each a numeric pleasantness rating or to use a slider to adjust particular notes in a chord to make it sound more pleasant. 探花直播experiments produced over 235,000 human judgments.

探花直播experiments explored musical chords from different perspectives. Some zoomed in on particular musical intervals and asked participants to judge whether they preferred them perfectly tuned, slightly sharp or slightly flat. 探花直播researchers were surprised to find a significant preference for slight imperfection, or 鈥榠nharmonicity鈥�. Other experiments explored harmony perception with Western and non-Western musical instruments, including the bonang.

聽

Instinctive appreciation of new kinds of harmony

探花直播researchers found that the bonang鈥檚 consonances mapped neatly onto the particular musical scale used in the Indonesian culture from which it comes. These consonances cannot be replicated on a Western piano, for instance, because they would fall between the cracks of the scale traditionally used.聽

鈥淥ur findings challenge the traditional idea that harmony can only be one way, that chords have to reflect these mathematical relationships. We show that there are many more kinds of harmony out there, and that there are good reasons why other cultures developed them,鈥� Dr Harrison said.

Importantly, the study suggests that its participants 鈥� not trained musicians and unfamiliar with Javanese music 鈥� were able to appreciate the new consonances of the bonang鈥檚 tones instinctively.

鈥淢usic creation is all about exploring the creative possibilities of a given set of qualities, for example, finding out what kinds of melodies can you play on a flute, or what kinds of sounds can you make with your mouth,鈥� Harrison said.

鈥淥ur findings suggest that if you use different instruments, you can unlock a whole new harmonic language that people intuitively appreciate, they don鈥檛 need to study it to appreciate it. A lot of experimental music in the last 100 years of Western classical music has been quite hard for listeners because it involves highly abstract structures that are hard to enjoy. In contrast, psychological findings like ours can help stimulate new music that listeners intuitively enjoy.鈥�

Exciting opportunities for musicians and producers

Dr Harrison hopes that the research will encourage musicians to try out unfamiliar instruments and see if they offer new harmonies and open up new creative possibilities.聽

鈥淨uite a lot of pop music now tries to marry Western harmony with local melodies from the Middle East, India, and other parts of the world. That can be more or less successful, but one problem is that notes can sound dissonant if you play them with Western instruments.聽

鈥淢usicians and producers might be able to make that marriage work better if they took account of our findings and considered changing the 鈥榯imbre鈥�, the tone quality, by using specially chosen real or synthesised instruments. Then they really might get the best of both worlds: harmony and local scale systems.鈥�

Harrison and his collaborators are exploring different kinds of instruments and follow-up studies to test a broader range of cultures. In particular, they would like to gain insights from musicians who use 鈥榠nharmonic鈥� instruments to understand whether they have internalised different concepts of harmony to the Western participants in this study.

Reference

R Marjieh, P M C Harrison, H Lee, F Deligiannaki, and N Jacoby, 鈥�Timbral effects on consonance disentangle psychoacoustic mechanisms and suggest perceptual origins for musical scales鈥�, Nature Communications (2024). DOI: 10.1038/s41467-024-45812-z

探花直播tone and tuning of musical instruments has the power to manipulate our appreciation of harmony, new research shows. 探花直播findings challenge centuries of Western music theory and encourage greater experimentation with instruments from different cultures.

There are many more kinds of harmony out there

Peter Harrison

Pythagoras was wrong: there are no universal harmonies! | Cambridge research

Andrew Otto via Flikr under a CC license

A man playing a bonang

Yes

Licence type:

Attribution

Scientists begin building AI for scientific discovery using tech behind ChatGPT

sc604 — Fri, 13 Oct 2023 10:55:15 +0000

While ChatGPT deals in words and sentences, the team鈥檚 AI will learn from numerical data and physics simulations from across scientific fields to aid scientists in modelling everything from supergiant stars to the Earth鈥檚 climate.

探花直播team launched the initiative, called Polymathic AI earlier this week, alongside the publication of a series of related scientific papers on the arXiv.org open access repository.

鈥淭his will completely change how people use AI and machine learning in science,鈥� said Polymathic AI principal investigator Shirley Ho, a group leader at the Flatiron Institute鈥檚 Center for Computational Astrophysics in New York City.

探花直播idea behind Polymathic AI 鈥渋s similar to how it鈥檚 easier to learn a new language when you already know five languages,鈥� said Ho.

Starting with a large, pre-trained model, known as a foundation model, can be both faster and more accurate than building a scientific model from scratch. That can be true even if the training data isn鈥檛 obviously relevant to the problem at hand.

鈥淚t鈥檚 been difficult to carry out academic research on full-scale foundation models due to the scale of computing power required,鈥� said co-investigator Miles Cranmer, from Cambridge鈥檚 Department of Applied Mathematics and Theoretical Physics and Institute of Astronomy. 鈥淥ur collaboration with Simons Foundation has provided us with unique resources to start prototyping these models for use in basic science, which researchers around the world will be able to build from 鈥� it鈥檚 exciting.鈥�

鈥淧olymathic AI can show us commonalities and connections between different fields that might have been missed,鈥� said co-investigator Siavash Golkar, a guest researcher at the Flatiron Institute鈥檚 Center for Computational Astrophysics. 鈥淚n previous centuries, some of the most influential scientists were polymaths with a wide-ranging grasp of different fields. This allowed them to see connections that helped them get inspiration for their work. With each scientific domain becoming more and more specialised, it is increasingly challenging to stay at the forefront of multiple fields. I think this is a place where AI can help us by aggregating information from many disciplines.鈥�

探花直播Polymathic AI team includes researchers from the Simons Foundation and its Flatiron Institute, New York 探花直播, the 探花直播 of Cambridge, Princeton 探花直播 and the Lawrence Berkeley National Laboratory. 探花直播team includes experts in physics, astrophysics, mathematics, artificial intelligence and neuroscience.

Scientists have used AI tools before, but they鈥檝e primarily been purpose-built and trained using relevant data. 鈥淒espite rapid progress of machine learning in recent years in various scientific fields, in almost all cases, machine learning solutions are developed for specific use cases and trained on some very specific data,鈥� said co-investigator Francois Lanusse, a cosmologist at the Centre national de la recherche scientifique (CNRS) in France. 鈥淭his creates boundaries both within and between disciplines, meaning that scientists using AI for their research do not benefit from information that may exist, but in a different format, or in a different field entirely.鈥�

Polymathic AI鈥檚 project will learn using data from diverse sources across physics and astrophysics (and eventually fields such as chemistry and genomics, its creators say) and apply that multidisciplinary savvy to a wide range of scientific problems. 探花直播project will 鈥渃onnect many seemingly disparate subfields into something greater than the sum of their parts,鈥� said project member Mariel Pettee, a postdoctoral researcher at Lawrence Berkeley National Laboratory.

鈥淗ow far we can make these jumps between disciplines is unclear,鈥� said Ho. 鈥淭hat鈥檚 what we want to do 鈥� to try and make it happen.鈥�

ChatGPT has well-known limitations when it comes to accuracy (for instance, the chatbot says 2,023 times 1,234 is 2,497,582 rather than the correct answer of 2,496,382). Polymathic AI鈥檚 project will avoid many of those pitfalls, Ho said, by treating numbers as actual numbers, not just characters on the same level as letters and punctuation. 探花直播training data will also use real scientific datasets that capture the physics underlying the cosmos.

Transparency and openness are a big part of the project, Ho said. 鈥淲e want to make everything public. We want to democratise AI for science in such a way that, in a few years, we鈥檒l be able to serve a pre-trained model to the community that can help improve scientific analyses across a wide variety of problems and domains.鈥�

An international team of scientists, including from the 探花直播 of Cambridge, have launched a new research collaboration that will leverage the same technology behind ChatGPT to build an AI-powered tool for scientific discovery.

Yuichiro Chino via Getty Images

Network and data connection on a dark blue background.

探花直播text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright 漏探花直播 of Cambridge and licensors/contributors as identified.聽 All rights reserved. We make our image and video content available in a number of ways 鈥� 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

How sure is sure? Incorporating human error into machine learning

sc604 — Wed, 09 Aug 2023 23:40:58 +0000

Human error and uncertainty are concepts that many artificial intelligence systems fail to grasp, particularly in systems where a human provides feedback to a machine learning model. Many of these systems are programmed to assume that humans are always certain and correct, but real-world decision-making includes occasional mistakes and uncertainty.

Researchers from the 探花直播 of Cambridge, along with 探花直播Alan Turing Institute, Princeton, and Google DeepMind, have been attempting to bridge the gap between human behaviour and machine learning, so that uncertainty can be more fully accounted for in AI applications where humans and machines are working together. This could help reduce risk and improve trust and reliability of these applications, especially where safety is critical, such as medical diagnosis.

探花直播team adapted a well-known image classification dataset so that humans could provide feedback and indicate their level of uncertainty when labelling a particular image. 探花直播researchers found that training with uncertain labels can improve these systems鈥� performance in handling uncertain feedback, although humans also cause the overall performance of these hybrid systems to drop. Their results will be reported at the AAAI/ACM Conference on Artificial Intelligence, Ethics and Society (AIES 2023) in Montr茅al.

鈥楬uman-in-the-loop鈥� machine learning systems 鈥� a type of AI system that enables human feedback 鈥� are often framed as a promising way to reduce risks in settings where automated models cannot be relied upon to make decisions alone. But what if the humans are unsure?

鈥淯ncertainty is central in how humans reason about the world but many AI models fail to take this into account,鈥� said first author Katherine Collins from Cambridge鈥檚 Department of Engineering. 鈥淎 lot of developers are working to address model uncertainty, but less work has been done on addressing uncertainty from the person鈥檚 point of view.鈥�

We are constantly making decisions based on the balance of probabilities, often without really thinking about it. Most of the time 鈥� for example, if we wave at someone who looks just like a friend but turns out to be a total stranger 鈥� there鈥檚 no harm if we get things wrong. However, in certain applications, uncertainty comes with real safety risks.

鈥淢any human-AI systems assume that humans are always certain of their decisions, which isn鈥檛 how humans work 鈥� we all make mistakes,鈥� said Collins. 鈥淲e wanted to look at what happens when people express uncertainty, which is especially important in safety-critical settings, like a clinician working with a medical AI system.鈥�

鈥淲e need better tools to recalibrate these models, so that the people working with them are empowered to say when they鈥檙e uncertain,鈥� said co-author Matthew Barker, who recently completed his MEng degree at Gonville聽& Caius College, Cambridge. 鈥淎lthough machines can be trained with complete confidence, humans often can鈥檛 provide this, and machine learning models struggle with that uncertainty.鈥�

For their study, the researchers used some of the benchmark machine learning datasets: one was for digit classification, another for classifying chest X-rays, and one for classifying images of birds. For the first two datasets, the researchers simulated uncertainty, but for the bird dataset, they had human participants indicate how certain they were of the images they were looking at: whether a bird was red or orange, for example. These annotated 鈥榮oft labels鈥� provided by the human participants allowed the researchers to determine how the final output was changed. However, they found that performance degraded rapidly when machines were replaced with humans.

鈥淲e know from decades of behavioural research that humans are almost never 100% certain, but it鈥檚 a challenge to incorporate this into machine learning,鈥� said Barker. 鈥淲e鈥檙e trying to bridge the two fields so that machine learning can start to deal with human uncertainty where humans are part of the system.鈥�

探花直播researchers say their results have identified several open challenges when incorporating humans into machine learning models. They are releasing their datasets so that further research can be carried out and uncertainty might be built into machine learning systems. 聽

鈥淎s some of our colleagues so brilliantly put it, uncertainty is a form of transparency, and that鈥檚 hugely important,鈥� said Collins. 鈥淲e need to figure out when we can trust a model and when to trust a human and why. In certain applications, we鈥檙e looking at probability over possibilities. Especially with the rise of chatbots, for example, we need models that better incorporate the language of possibility, which may lead to a more natural, safe experience.鈥�

鈥淚n some ways, this work raised more questions than it answered,鈥� said Barker. 鈥淏ut even though humans may be miscalibrated in their uncertainty, we can improve the trustworthiness and reliability of these human-in-the-loop systems by accounting for human behaviour.鈥�

探花直播research was supported in part by the Cambridge Trust, the Marshall Commission, the Leverhulme Trust, the Gates Cambridge Trust and the Engineering and Physical Sciences Research Council (EPSRC), part of UK Research and Innovation (UKRI).

聽

Reference:
Katherine M Collins et al. 鈥楬uman Uncertainty in Concept-Based AI Systems.鈥� Paper presented at the Sixth AAAI/ACM Conference on Artificial Intelligence, Ethics and Society (AIES 2023), August 8-10, 2023. Montr茅al, QC, Canada.

Researchers are developing a way to incorporate one of the most human of characteristics 鈥� uncertainty 鈥� into machine learning systems.

Uncertainty is central in how humans reason about the world but many AI models fail to take this into account

Katherine Collins

PeopleImages via Getty Images

Futuristic image of a doctor looking at brain scans

探花直播text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright 漏探花直播 of Cambridge and licensors/contributors as identified.聽 All rights reserved. We make our image and video content available in a number of ways 鈥� 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

New findings that map the universe鈥檚 cosmic growth support Einstein鈥檚 theory of gravity

sc604 — Tue, 11 Apr 2023 14:00:00 +0000

探花直播findings, from the Atacama Cosmology Telescope collaboration involving researchers from the 探花直播 of Cambridge, provide further support to Einstein鈥檚 theory of general relativity, which has been the foundation of the standard model of cosmology for more than a century. 探花直播results offer new methods to demystify dark matter, the unseen mass thought to account for 85% of the matter in the universe.

For millennia, humans have been fascinated by the mysteries of the cosmos. From ancient civilisations such as the Babylonians, Greeks, and Egyptians to modern-day astronomers, the allure of the starry sky has inspired countless quests to unravel the secrets of the universe.

And although models that explain the cosmos have existed for centuries, the field of cosmology, where scientists use quantitative methods to understand the evolution and structure of the universe, is relatively new鈥攈aving only formed in the early 20th century with the development of Albert Einstein鈥檚 theory of general relativity.聽

Now, a set of papers submitted to 探花直播Astrophysical Journal聽by researchers from the Atacama Cosmology Telescope (ACT) collaboration has produced a new image that reveals the most detailed map of matter distributed across a quarter of the entire sky, reaching deep into the cosmos. It confirms Einstein鈥檚 theory about how massive structures grow and bend light, with a test that spans the entire age of the universe.

鈥淲e have mapped the invisible dark matter across the sky to the largest distances, and clearly see features of this invisible world that are hundreds of millions of light-years across,鈥� said co-author Professor Blake Sherwin from Cambridge鈥檚 Department of Applied Mathematics and Theoretical Physics, where he leads a group of ACT researchers. 鈥淚t looks just as our theories predict.鈥�

Although dark matter makes up a large chunk of the universe and shaped its evolution, it has remained hard to detect because it doesn鈥檛 interact with light or other forms of electromagnetic radiation. As far as we know, dark matter only interacts with gravity.聽

To track it down, the more than 160 collaborators who have built and gathered data from the National Science Foundation鈥檚 Atacama Cosmology Telescope in the high Chilean Andes observe light emanating following the dawn of the universe鈥檚 formation, the Big Bang鈥攚hen the universe was only 380,000 years old. Cosmologists often refer to this diffuse light that fills our entire universe as the 鈥渂aby picture of the universe,鈥� but formally, it is known as the cosmic microwave background radiation (CMB).

探花直播team tracks how the gravitational pull of large, heavy structures including dark matter warps the CMB on its 14-billion year journey to us, like how a magnifying glass bends light as it passes through its lens.

鈥淲e鈥檝e made a new mass map using distortions of light left over from the Big Bang,鈥� said Mathew Madhavacheril from the 探花直播 of Pennsylvania, lead author of one of the papers. 鈥淩emarkably, it provides measurements that show that both the 鈥榣umpiness鈥� of the universe, and the rate at which it is growing after 14 billion years of evolution, are just what you鈥檇 expect from our standard model of cosmology based on Einstein's theory of gravity.鈥澛�

鈥淥ur results also provide new insights into an ongoing debate some have called 鈥� 探花直播Crisis in Cosmology鈥�,鈥� said Sherwin. This crisis stems from recent measurements that use a different background light, one emitted from stars in galaxies rather than the CMB. These have produced results that suggest the dark matter was not lumpy enough under the standard model of cosmology and led to concerns that the model may be broken. However, the team鈥檚 latest results from ACT were able to precisely assess that the vast lumps seen in this image are the exact right size.聽

鈥淲hen I first saw them, our measurements were in such good agreement with the underlying theory that it took me a moment to process the results,鈥� said Cambridge PhD candidate Frank Qu, lead author of one of the new papers. 鈥淏ut we still don鈥檛 know what the dark matter is, so it will be interesting to see how this possible discrepancy between different measurements will be resolved.鈥�

鈥� 探花直播CMB lensing data rivals more conventional surveys of the visible light from galaxies in their ability to trace the sum of what is out there,鈥� said Suzanne Staggs from Princeton 探花直播, Director of ACT. 鈥淭ogether, the CMB lensing and the best optical surveys are clarifying the evolution of all the mass in the universe.鈥澛�

鈥淲hen we proposed this experiment in 2003, this measurement wasn鈥檛 even on our agenda; we had no idea the full extent of information that could be extracted from our telescope,鈥� said Mark Devlin, from the 探花直播 of Pennsylvania, Deputy Director of ACT. 鈥淲e owe this to the cleverness of the theorists, the many people who built new instruments to make our telescope more sensitive, and the new analysis techniques our team came up with.鈥�

With ACT having been decommissioned in late 2022, further papers highlighting some of the other final results are slated for submission in the coming year. Observations will continue at the site with the Simons Observatory, including a new telescope due to begin in 2024 that can map the sky almost ten times faster.

探花直播pre-print articles highlighted in this release are available on act.princeton.edu and will appear on the open-access arXiv.org. They have been submitted to 探花直播Astrophysical Journal.聽

This work was supported by the U.S. National Science Foundation, Princeton 探花直播, the 探花直播 of Pennsylvania, and a Canada Foundation for Innovation award. Team members at the 探花直播 of Cambridge were supported by the European Research Council.

A new image reveals the most detailed map of dark matter distributed across a quarter of the entire sky, reaching deep into the cosmos.

We have mapped the invisible dark matter across the sky to the largest distances, and clearly see features of this invisible world that are hundreds of millions of light-years across

Blake Sherwin

ACT Collaboration

A new map of the dark matter made by the Atacama Cosmology Telescope. 探花直播orange regions show where there is more mass; purple where there is less.

探花直播text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright 漏探花直播 of Cambridge and licensors/contributors as identified.聽 All rights reserved. We make our image and video content available in a number of ways 鈥� 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

What did Megalodon eat? Anything it wanted 鈥� including other predators

sc604 — Thu, 23 Jun 2022 16:32:36 +0000

New research involving the 探花直播 of Cambridge shows that prehistoric megatooth sharks 鈥� the biggest sharks that ever lived 鈥� were the ultimate top predators, operating higher up the food chain than any other marine predators through history.

Learn from the pandemic to prevent environmental catastrophe, scientists argue

fpjl2 — Thu, 02 Jul 2020 10:24:20 +0000

探花直播dynamics of the SARS-CoV-2 pandemic share 'striking similarities'聽with the twin environmental crises of global heating and species extinction, argue a team of scientists and policy experts from the UK and US.

They say that lessons learned the hard way in containing COVID-19 鈥� the need for early intervention to reduce death and economic damage; the curbing of some aspects of people鈥檚 lifestyles for the good of all of us 鈥� should also be at the heart of averting environmental catastrophe.

鈥淲e鈥檝e seen the consequences of delayed action in the fight against COVID-19. 探花直播consequences of continued inaction in the face of catastrophic climate change and mass extinction are too grave to contemplate,鈥� said Prof Andrew Balmford, from the 探花直播 of Cambridge鈥檚 Department of Zoology.

Writing in the journal Current Biology, Balmford and colleagues argue that the spread of coronavirus shares common characteristics with both global heating and the impending 'sixth mass extinction'.

For example, each new COVID-19 case can spawn others and so lead to escalating infection rates, just as hotter climates alter ecosystems, increasing emissions of the greenhouse gases that cause warming. 鈥淏oth are dangerous feedback loops,鈥� argue the scientists.

探花直播team also draw comparisons of what they term 'lagged impacts'. For coronavirus, the delay 鈥� or lag 鈥� before symptoms materialise means infected people spread the disease before they feel effects and change behaviour.

探花直播researchers equate this with the lag between our destruction of habitats and the eventual extinction of species, as well as lags between the emissions we pump out and the full effects of global heating, such as sea-level rise. As with viral infection, behaviour change may come too late.

鈥淟ike the twin crises of extinction and climate, the SARS-CoV-2 pandemic might have seemed like a distant problem at first, one far removed from most people鈥檚 everyday lives,鈥� said coauthor Ben Balmford from the 探花直播 of Exeter.

鈥淏ut left unchecked for too long, the disease has forced major changes to the way we live. 探花直播same will be true of the environmental devastation we are causing, except the consequences could be truly irreversible.鈥�

探花直播authors find parallels in the indifference that has long greeted warnings from the scientific community about both new zoonotic diseases and human-induced shifts in climate and habitat.

鈥� 探花直播lagged impacts, feedback loops and complex dynamics of pandemics and environmental crises mean that identifying and responding to these challenges requires governments to listen to independent scientists,鈥� said Dr Brendan Fisher, a coauthor from the 探花直播 of Vermont. 鈥淪uch voices have been tragically ignored.鈥�

探花直播similarities between the SARS-CoV-2 pandemic and environmental disaster lie not just in their nature but also in their mitigation, say the scientists, who write that 'there is no substitute for early action'.

探花直播researchers include an analysis of the timing of lockdown across OECD countries, and conclude that if it had come just a week earlier then around 17,000 lives in the UK (up to 21 May 2020) would have been saved, and nearly 45,000 in the USA.

They say that, just as delayed lockdown cost thousands of lives, delayed climate action that gives us 2oC of warming rather than 1.5 will expose an estimated extra 62-457 million people 鈥� mainly the world鈥檚 poorest 鈥� to 'multi-sector climate risks'聽such as drought, flooding and famine.

Similarly, conservation programmes are less likely to succeed the longer they are delayed. 鈥淎s wilderness disappears we see an accelerating feedback loop, as a given loss of habitat causes ever-greater species loss,鈥� explained Princeton Professor and co-author David Wilcove.聽聽聽聽

探花直播scientists point out that delayed action resulting in more COVID-19 deaths will also cost those nations more in economic growth, according to IMF estimates, just as hotter and more disruptive climates will curtail economic prosperity.

Intervening to contain both the pandemic and the environmental crises requires decision-makers and citizens to act in the interests of society as a whole, argue the researchers.

鈥淚n the COVID-19 crisis we鈥檝e seen young and working age people sacrificing education, income and social connection primarily for the benefit of older and more vulnerable people,鈥� said co-author Prof Dame Georgina Mace from UCL.

鈥淭o stem the impacts of climate change and address biodiversity loss, wealthier and older adults will have to forgo short-term material extravagance for the benefit of the present-day poor and future generations. It鈥檚 time to keep our end of the social bargain,鈥� Mace said.

Cambridge鈥檚 Andrew Balmford added: 鈥淪cientists are not inventing these environmental threats, just as they weren鈥檛 inventing the threat of a pandemic such as COVID-19. They are real, and they are upon us.鈥�

COVID-19 is comparable to climate and extinction emergencies, say scientists from the UK and US 鈥� all share features such as lagged impacts, feedback loops, and complex dynamics.

探花直播consequences of continued inaction in the face of catastrophic climate change and mass extinction are too grave to contemplate

Andrew Balmford

Balinda O'Neal

US National Guard working to extinguish wildfires in Alaska

探花直播text in this work is licensed under a Creative Commons Attribution 4.0 International License. Images, including our videos, are Copyright 漏探花直播 of Cambridge and licensors/contributors as identified.聽 All rights reserved. We make our image and video content available in a number of ways 鈥� 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-Noncommerical