(Press-News.org) New research reveals a surprising geometric link between human and machine learning. A mathematical property called convexity may help explain how brains and algorithms form concepts and make sense of the world.
In recent years, with the public availability of AI tools, more people have become aware of how closely the inner workings of artificial intelligence can resemble those of a human brain.
There are several similarities in how machines and human brains work, for example, in how they represent the world in abstract form, generalise from limited data, and process data in layers. A new paper in Nature Communications by DTU researchers is adding another feature to the list: Convexity.
"We found that convexity is surprisingly common in deep networks and might be a fundamental property that emerges naturally as machines learn," says Lars Kai Hansen, a DTU Compute professor who led the study.
Convexity may bridge human and machine intelligence
To briefly explain the concept, when we humans learn about a "cat," we don't just store a single image but build a flexible understanding that allows us to recognise all sorts of cats—be they big, small, fluffy, sleek, black, white, and so on.
Coming from mathematics to describe, e.g., geometry, the term convexity was applied to cognitive science by Peter Gärdenfors who proposed that our brains form conceptual spaces where related ideas cluster. And here's the crucial part: natural concepts, like "cat" or "wheel," tend to form convex regions in these mental spaces. In short, one could imagine a rubber band stretching around a group of similar ideas—that's a convex region.
Think of it like this: Inside the perimeter of the rubber band, if you have two points representing two different cats, any point on the shortest path between them also falls within the mental "cat" region. Such convexity is powerful as it helps us generalise from a few examples, learn new things quickly, and even helps us communicate and agree on what things mean. It's a fundamental property that makes human learning robust, flexible and social.
When it comes to deep learning models - the engines behind everything from image generation to chatbots - they learn by transforming raw data like pixels or words into complex internal representations, often called "latent spaces." These spaces can be viewed as internal maps where the AI organises its understanding of the world.
Measuring AI's internal structure
To make AI more reliable, trustworthy and aligned with human values, there is a need to develop better ways to describe how it represents knowledge. Therefore, it is critical to determine whether machine-learned spaces are organised in a way that resembles human conceptual spaces and whether they also form convex regions for concepts.
First author of the paper, Lenka Tetkova, who is a postdoc at DTU Compute, dove into this very question, looking at two main types of convexity:
First is Euclidean convexity, which is straightforward: if you take two points within a concept in a model's latent space, and the straight line between them stays entirely within that concept, then the region is Euclidean convex. This is like generalising by blending known examples.
The other is graph convexity, which is more flexible and especially important for the curved geometries often found in AI's internal representations. Imagine a network of similar data points—if the shortest path between two points within a concept stays entirely inside that concept, then it's graph convex. This reflects how models might generalise by following the natural structure of the data.
"We've developed new tools to measure convexity within the complex latent spaces of deep neural networks. We tested these measures across various AI models and data types: images, text, audio, human activity, and even medical data. And we found that the same geometric principle that helps humans form and share concepts—convexity—also shapes how machines learn, generalise, and align with us," says Lenka Tetkova.
AIs hidden order
The researchers also discovered that the commonalities are found in pretrained models that learn general patterns from massive datasets and finetuned models that are taught specific tasks like identifying animals. This further substantiates the claim that convexity might be a fundamental property that emerges naturally as machines learn.
When models are fine-tuned for a specific task, the convexity of their decision regions increases. As AI improves at classification, its internal concept regions become more clearly convex, refining its understanding and sharpening its boundaries.
In addition, the researchers discovered that the level of convexity in a pretrained model's concepts can predict how well that model will perform after finetuning.
"Imagine that a concept, say, a cat, forms a nice, well-defined convex region in the machine before it's even taught to identify cats specifically. Then it's more likely to learn to identify cats accurately later on. We believe this is a powerful insight, because it suggests that convexity might be a useful indicator of a model's potential for specific learning tasks," says Lars Kai Hansen.
A route to better AI
According to the researchers, these new results may have several important implications. By identifying convexity as a pervasive property, they have better understood how deep neural networks learn and organise information. It provides a concrete mechanism for how AI generalises, which may be like how humans learn.
If convexity does prove to be a reliable predictor of performance, it may be possible to design AI models that explicitly encourage the formation of convex concept regions during training. This could lead to more efficient and effective learning, especially in scenarios where only a few examples are available. The findings may therefore provide a crucial new bridge between human cognition and machine intelligence.
"By showing that AI models exhibit properties (like convexity) that are fundamental to human conceptual understanding, we move closer to creating machines that 'think' in ways that are more comprehensible and aligned with our own. This is vital for building trust and collaboration between humans and machines in critical applications like healthcare, education, and public service," says Lenka Tetkova.
"While there's still much to explore, the results suggest that the seemingly abstract idea of convexity may hold the key to unlocking new secrets on AI's internal workings and bringing us closer to intelligent and human-aligned machines."
The project
The research carried out within the research project “Cognitive Spaces – Next generation explainable AI” funded by the Novo Nordisk Foundation. The project’s aim is to open the machine learning black-box and build tools to explain the inner workings of AI-systems with concepts that can be understood by specific user groups.
END
Peeking inside AI brains: Machines learn like us
New research reveals a surprising geometric link between human and machine learning. A mathematical property called convexity may help explain how brains and algorithms form concepts and make sense of the world
2025-07-02
ELSE PRESS RELEASES FROM THIS DATE:
A map for single-atom catalysts
2025-07-02
Catalysis – the acceleration of a chemical reaction by adding a particular substance – is extremely important in industry as well as in everyday life. Around 80 % of all chemical products are produced with the help of catalysis, and technologies like exhaust catalysts or fuel cells are also based on this principle. One particularly effective and versatile catalyst is platinum. However, because platinum is a very rare and expensive precious metal whose production causes a lot of CO2 emissions, it is important to use as little of it as possible while maximizing its efficiency.
Catalysts with single ...
What about tritiated water release from Fukushima? Ocean model simulations provide an objective scientific knowledge on the long-term tritium distribution
2025-07-02
Tokyo, Japan – Operators have pumped water to cool the nuclear reactors at the Fukushima Daiichi Nuclear Power Plant (FDNPP) since the accident in 2011 and treated this cooling water with the Advanced Liquid Processing System (ALPS), which is a state-of-the-art purification system that removes radioactive materials, except tritium. As part of the water molecule, tritium radionuclide, with a half-life of 12.32 years, is very costly and difficult to remove. The ALPS-treated water was accumulating and stored at the FDNPP site and there is limited space to store this ...
Growing crisis of communicable disease in Canada in tandem with US cuts
2025-07-02
Canada must address the growing crisis of communicable diseases that has occurred in tandem with a rise in misinformation that threatens our health systems, argue authors in an editorial in CMAJ (Canadian Medical Association Journal) https://www.cmaj.ca/lookup/doi/10.1503/cmaj.250916.
“A crisis of communicable diseases is unfolding in North America, just as Canada’s health systems’ responses are being hampered by the dismantling of public health and research infrastructure in the United States,” writes family physician Dr. Shannon Charlebois, medical editor, CMAJ, with ...
Women get better at managing their anger as they age
2025-07-02
CLEVELAND, Ohio (July 2, 2025)—There has been a lot of research focused on understanding women’s experiences with depression during the menopause transition and early menopause, but there are few studies on perimenopausal women’s experiences with emotional arousal, such as anger. A new study shows that women’s anger traits significantly decrease with age starting at midlife. Results of the study are published online today in Menopause, the journal of The Menopause Society.
Anger is defined as antagonism toward someone or something, often accompanied by a propensity to experience and express it indiscriminately. ...
Illegal shark product trade evident in Australia and New Zealand
2025-07-02
Research from the University of Adelaide’s School of Biological Sciences and Wildlife Crime Research Hub has highlighted evidence of shark products entering both Australia and Aotearoa/New Zealand, including clear patterns in flows between the two countries.
According to the study, published in Pacific Conservation Biology, the products identified were carried in personal luggage and postage, likely transported for personal use, as trophies, or for resale or consumption.
Most products seized upon entry to Australia came from Asia, ...
New search tool brings 21% better accuracy for robotics developers
2025-07-02
Imagine you are on a treasure hunt in a vast library with no catalog—typing random words into a search bar and hoping to stumble upon the exact book you need. That has been the reality for many roboticists trying to find the right ROS (Robot Operating System) package. With over 7,500 options available, keyword searches often return irrelevant results, wasting developers’ precious time and energy.
Researchers from the National University of Defense Technology and Zhejiang University have developed a more efficient method for searching. Instead of relying on simple word matching, their new tool uses a “knowledge ...
New model extracts sentence-level proof to verify events, boosting fact-checking accuracy for journalists, legal teams, and policymakers
2025-07-02
Imagine reading a long article or a thick legal contract and knowing, with confidence, exactly which lines prove that an event happened—or did not happen. That is now possible thanks to a research team at Soochow University. They have built a new neural network that not only determines if an event described in a document is real but also highlights the exact sentences that led it to that conclusion. In head-to-head comparisons with earlier approaches, this new model improved overall fact-checking accuracy by 2.5 points and exact-match ...
Efficient carbon integration of CO₂ in propane aromatization over acidic zeolites
2025-07-02
Aromatics, as extremely crucial basic chemicals in the modern industrial system, are widely used in many fields such as energy, medicine, materials and the daily chemical industry. However, the traditional petroleum-based production routes, such as naphtha cracking and catalytic reforming, are facing the dual pressures of tight petroleum resources and carbon emissions. Meanwhile, CO2, as a typical greenhouse gas, its efficient and value-added utilization offers dual benefits in both environment and economic development, but is limited by its thermodynamic stability, and there are still challenges in producing aromatics with high selectivity through hydrogenation pathways. Propane, as a major ...
FPGA-accelerated AI for demultiplexing multimode fiber towards next-generation communications
2025-07-02
With the exponential growth of global data traffic driven by AI, big-data analytics, and cloud computing, today’s single-mode fiber (SMF) networks are edging toward their Shannon-capacity limits. Space-division multiplexing (SDM) in multimode fiber (MMF) has emerged as a leading candidate for the next-generation bandwidth breakthrough because a single MMF can carry many orthogonal transverse modes in parallel. However, random mode coupling during propagation mixes these modes into complex speckle patterns, severely complicating signal recovery. Although conventional digital signal ...
Vitamin D3 nanoemulsion significantly improves core symptoms in children with autism: A clinical trial
2025-07-02
This study investigates the effectiveness of a vitamin D3-loaded nanoemulsion in improving the core symptoms of autism spectrum disorder (ASD) in children. Children with ASD often have low vitamin D3 levels, which are linked to delays in language development, adaptive behavior, and fine motor skills. While traditional vitamin D3 supplementation has shown mixed results in past studies, this research evaluates whether a nanoemulsion form—engineered to enhance absorption and bioavailability—might produce better outcomes.
Eighty children between the ages of 3 and 6 with diagnosed ASD were randomly assigned into two groups: one receiving the ...
LAST 30 PRESS RELEASES:
Vaccination support program reduces pneumonia-related mortality by 25 percent among the elderly
Over decades, a healthy lifestyle outperforms metformin in preventing onset of Type 2 diabetes
Mental health disorders, malaria, and heart disease most affected by covid pandemic
Green transition will boost UK productivity
Billions voted in 2024, but major new report exposes cracks in global democracy
Researchers find “forever chemicals” impact the developing male brain
Quantum leap in precision sensing across technologies
Upgrading biocrude oil into sustainable aviation fuel using zeolite-supported iron-molybdenum carbide nanocatalysts
For effective science communication, ‘just the facts’ isn’t good enough
RT-EZ: A golden gate assembly toolkit for streamlined genetic engineering of rhodotorula toruloides
Stem Cell Reports announces five new early career editors
Support networks may be the missing link for college students who seek help for excessive drinking
The New England Journal of Medicine shines spotlight on forensic pathology
Scientists discover protein that helps lung cancer spread to the brain
Perceived social status tied to cardiovascular risks in women but not in men
Brain tumor growth patterns may help inform patient care management
This might be America's first campus tree inventory
Emoji use may impact relationship outcomes
Individual merit, not solidarity, prioritized by early childhood education policies
Preclinical study unlocks a mystery of rapid mouth healing
Extraterrestrial habitats: bioplastics for life beyond earth
U.S. military spending reductions could substantially lower energy consumption
Air pollution is linked to adverse birth outcomes in India
Using viral load tests to help predict mpox severity when skin lesions first appear
Engineered cell cross-talk unlocks CAR-T potential against glioblastoma
Regional disparities in US media coverage of archaeology research
Coral larvae travelling further makes populations stronger
First of its kind study for children with arthritis reveals possible new disease targets
Financing innovation: proposal for novel adaptive platform trial fund offers new model for ALS drug development
Disparities in treatment and referral after an opioid overdose among emergency department patients
[Press-News.org] Peeking inside AI brains: Machines learn like usNew research reveals a surprising geometric link between human and machine learning. A mathematical property called convexity may help explain how brains and algorithms form concepts and make sense of the world