PRESS-NEWS.org - Press Release Distribution
PRESS RELEASES DISTRIBUTION

Approximate domain unlearning: Enabling safer and more controllable vision-language models

The proposed approximate unlearning technique differentiates between domains while preserving the generalization capability

2025-12-02
(Press-News.org)

Vision-language model (VLM) is a core technology of modern artificial intelligence (AI), and it can be used to represent different forms of expression or learning, such as photographs, illustrations, and sketches. It has high generalization ability, which allows it to accurately recognize objects in images within a domain. However, this generalization ability is at risk. For example, VLM recognizes both real cars and illustrated cars as “cars.” If this is installed in a system, there is a risk that a car illustrated in a roadside advertisement may be mistaken for a real vehicle, leading to a serious accident. To put safe and reliable AI into practical use, it is essential to establish technology that appropriately controls learned knowledge according to the application.

Addressing this issue, a team of researchers, led by Associate Professor Go Irie from Tokyo University of Science, has proposed the approximate domain unlearning (ADU) algorithm, which allows VLM to “forget” specific domains so that it cannot recognize them. For example, according to these results, real-life vehicles can be recognized with high accuracy.

This innovative study, co-authored by Mr. Kodai Kawamura and Mr. Yuta Goto from Tokyo University of Science, Japan, Dr. Rintaro Yanagi from National Institute of Advanced Industrial Science and Technology, and Dr. Hirokatsu Kataoka from AIST and University of Oxford, has been presented at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025), held from November 30 to December 7, 2025.

“AI technologies have long pursued the ability to recognize objects accurately across all domains, as seen in decades of research on domain adaptation and domain generalization. While such versatility remains important, the emergence of VLMs with remarkable domain generalization ability made us realize that this assumption itself deserves to be re-examined. This idea led us to conceive ADU—a new approach that deliberately enables models to forget specified domains, when necessary,” explains Prof. Irie.

Notably, the technical difficulty is that the domains cannot be distinguished within the VLM. As different domains overlap in the feature space, it is difficult to select and forget only specific domains. Therefore, in this research, the team introduced a method called Domain Disentangling Loss that promotes separation between domains in the feature space and captures the different domain appearances in each image. Furthermore, by introducing the instance-wise prompt generator, the proposed algorithm reduces the recognition accuracy for unnecessary domains while minimizing the need for them. This allows for flexible AI configuration suited to individual practical scenarios and enables flexible knowledge control that was previously impossible, such as making it impossible to recognize cars in illustrations while maintaining functionality.

Interestingly, ADU introduces a new perspective on risk management. While the concept of managing AI risks has long existed, the very generalization ability of modern AI models can sometimes generate new risks. This research presents a framework for building AI that can be flexibly configured according to individual usage scenarios, ensuring both safety and adaptability.

“While AI performance is becoming more sophisticated, in order to promote sustainable industrial applications, it is necessary to adapt it to practical scenarios. We feel that our system, which allows us to freely control functions, will enable us to provide the world with safe and reliable AI technology,” concludes Prof. Irie.


***

 

Reference                     
Conference: Neural Information Processing Systems (NeurIPS 2025)

 

About The Tokyo University of Science
Tokyo University of Science (TUS) is a well-known and respected university, and the largest science-specialized private research university in Japan, with four campuses in central Tokyo and its suburbs and in Hokkaido. Established in 1881, the university has continually contributed to Japan's development in science through inculcating the love for science in researchers, technicians, and educators.

With a mission of “Creating science and technology for the harmonious development of nature, human beings, and society," TUS has undertaken a wide range of research from basic to applied science. TUS has embraced a multidisciplinary approach to research and undertaken intensive study in some of today's most vital fields. TUS is a meritocracy where the best in science is recognized and nurtured. It is the only private university in Japan that has produced a Nobel Prize winner and the only private university in Asia to produce Nobel Prize winners within the natural sciences field.

Website: https://www.tus.ac.jp/en/mediarelations/

 

About Associate Professor Go Irie from Tokyo University of Science
Dr. Go Irie graduated in Engineering from Keio University in 2004 and obtained his PhD from The University of Tokyo in 2011. He specializes in pattern recognition, machine learning, and media understanding, with a strong focus on perceptual information processing. He has participated in numerous international conferences on these topics, including prestigious ones such as NeurIPS, CVPR, and ACM Multimedia.

 

Funding information
This project is partially supported by JPNP25006, commissioned by the New Energy and Industrial Technology Development Organization (NEDO).

END



ELSE PRESS RELEASES FROM THIS DATE:

Moths detect bat attack signals: Ultrasonic pulse rates drive distinct escape responses

2025-12-02
For many nocturnal moths, hearing sound waves is a matter of survival in night sky. Their ability to detect ultrasonic calls emitted by bats determines whether they escape or become prey. This predator-prey relationship has shaped the behavior, physiology, and sensory systems of both groups. Echolocating bats have developed complex call patterns to track insects in flight, while moths have evolved remarkable countermeasures, including evasive flight and sound-deflection tactics. The luna moth, for instance, spins its long hindwings to deflect the ultrasonic ...

Intimate partner violence injury patterns linked with suicidal behavior

2025-12-02
CHICAGO – Victims of intimate partner violence with suicidal behavior have characteristic injury patterns on medical imaging, according to a new study being presented today at the annual meeting of the Radiological Society of North America (RSNA). The findings open the door to improved screening and earlier intervention to better protect these vulnerable populations, the researchers said. Intimate partner violence is the physical, emotional or sexual abuse of a person by their partner or spouse. It is an increasingly recognized risk factor for suicidal behavior, and victims of intimate partner violence ...

Blood test shows obesity speeds Alzheimer’s development

2025-12-02
CHICAGO – Researchers have conducted the first study evaluating the impact of obesity on Alzheimer’s disease blood biomarkers (BBMs). BBM values increased up to 95% faster in individuals with obesity than in non-obese individuals, according to a new study being presented today at the annual meeting of the Radiological Society of North America (RSNA). “This is the first time we’ve shown the relationship between obesity and Alzheimer’s disease as measured by blood biomarker tests,” said Cyrus ...

New study supports the value of medical humanities in illuminating the root causes of health care disparities in Washington, DC

2025-12-02
WASHINGTON -- A new study analyzing dozens of published papers over five decades focusing on health care disparities in Washington, DC, found that those that employed medical humanities approaches identified crucial barriers and opportunities for intervention that quantitative studies often miss. Lead author Sweta Ghatti, a fourth year student at Georgetown School of Medicine, began the study as part of a Mitchell Summer Research Scholarship project addressing health challenges in the District. Ghatti worked closely with senior author Lakshmi Krishnan, MD, PhD, assistant professor ...

Uncovering the principle by which DNA replication initiation sites are determined in the human genome

2025-12-02
When cells proliferate, genomic DNA is precisely duplicated once per cell cycle. Abnormalities in this DNA replication process can cause alterations in genomic DNA, promoting cellular ageing, cancer, and genetic disorders. Therefore, understanding how cells replicate their DNA is crucial for elucidating fundamental biological processes, diseases, and even evolution. Traditionally, DNA replication has been studied in microorganisms such as E. coli and yeast. In these organisms, the location where DNA replication begins (replication origin) is determined by a specific DNA sequence. However, in most eukaryotic cells, including human cells, the DNA sequence itself ...

Urban sprawl could deny 220 million people access to clean water by 2050

2025-12-02
[Vienna, 02 December, 2025] — A new study analyzing over 100 cities across Asia, Africa, and Latin America has quantified the stark consequences of urban sprawl on water and sanitation access, finding that how cities grow might determine whether hundreds of millions of people have access to these basic necessities. The analysis, conducted by researchers at the Complexity Science Hub (CSH) and the World Bank, examined infrastructure data and economic indicators, including information on the footprint of 183 million buildings and 125,000 household surveys, to understand the relationship between urban form and access to clean water and sanitation. The ...

Researchers unveil first high-resolution maps of China's forest diversity patterns

2025-12-02
A research team led by the South China Botanical Garden of the Chinese Academy of Sciences, in collaboration with multiple domestic and international research institutions, has made progress in investigating forest diversity patterns across China. The findings were published in Nature Ecology & Evolution on December 2. China is recognized by Conservation International as one of the world's 17 megadiverse countries. To effectively meet its commitments under the Kunming-Montreal Global Biodiversity Framework, it is crucial for China to clarify the fine-scale spatial patterns ...

Sun-watcher SOHO celebrates thirty years

2025-12-02
On 2 December 1995 the ESA/NASA Solar and Heliospheric Observatory (SOHO) blasted into space – on what was supposed to be a two-year mission.  From its outpost 1.5 million km away from Earth in the direction of the Sun, SOHO enjoys uninterrupted views of our star. It has provided a nearly continuous record of our Sun’s activity for close to three 11-year-long solar cycles.   "It is testament to the ingenuity of our engineers, operators and scientists, and to international ...

Largest study of nose microbiome helps highlight those at risk of staph aureus infection

2025-12-02
  People who persistently carry Staphylococcus aureus (S. aureus) in their nose have fewer species of other bacteria, while certain bacteria may help to prevent S. aureus colonisation.  These are the findings of the largest-ever study of the nasal microbiome, published today (2 December) in Nature Communications. Researchers at the Wellcome Sanger Institute, University of Cambridge, Imperial ...

Structural racism and cultural misunderstanding compound grief for Black British and Black Caribbean communities, study finds

2025-12-02
Inequities in how bereavement is experienced and supported among people of Black British and Black Caribbean heritage in England has been revealed in a new study led by the University of Bristol. The research, published in Death Studies today [2 December], calls for widespread changes to improve bereavement experiences and access to support for Black British and Black Caribbean communities. People from Black and other minoritised ethnic communities in the UK are known to face persistent health and social care inequities, including barriers to accessing bereavement ...

LAST 30 PRESS RELEASES:

Korea University study finds restless legs syndrome linked to Parkinson’s risk—dopamine treatment may be protective

Pusan National University researchers use AI to create optimized engine components that outperform human designs

Approximate domain unlearning: Enabling safer and more controllable vision-language models

Moths detect bat attack signals: Ultrasonic pulse rates drive distinct escape responses

Intimate partner violence injury patterns linked with suicidal behavior

Blood test shows obesity speeds Alzheimer’s development

New study supports the value of medical humanities in illuminating the root causes of health care disparities in Washington, DC

Uncovering the principle by which DNA replication initiation sites are determined in the human genome

Urban sprawl could deny 220 million people access to clean water by 2050

Researchers unveil first high-resolution maps of China's forest diversity patterns

Sun-watcher SOHO celebrates thirty years

Largest study of nose microbiome helps highlight those at risk of staph aureus infection

Structural racism and cultural misunderstanding compound grief for Black British and Black Caribbean communities, study finds

Water molecules in motion: Surprising dynamics on 2D materials

Alaknanda: JWST discovers massive grand-design spiral galaxy from the universe's infancy

Our brains recognise the voices of our primate cousins

Does the "use it or lose it" principle determine brain plasticity and shape how we age?

Dynamic duo of bacteria could change Mars dust into versatile building material for first human colonists

Lower prevalence of PSC among patients with IBD in Asia: Insights from a multinational study

Alcohol and ultrasonic irradiation: An effective CCl₄ decomposition tag team

Conquer the diseases of aging and humans could live far longer than we think, scientists propose  

National study finds where you live influences your body weight

What your sweat can reveal about your health

Groundbreaking research compares prompt styles and LLMs for structured data generation - Unveiling key trade-offs for real-world AI applications

Beat the bugs, enjoy the beats

Genome advancement puts better Wagyu marbling on the menu

Developing a new electric vehicle sound

Elephant seals recognize their rivals from years prior

Fossils reveal anacondas have been giants for over 12 million years

Sylvester researchers lead major treatment overhauls for acute myeloid leukemia

[Press-News.org] Approximate domain unlearning: Enabling safer and more controllable vision-language models
The proposed approximate unlearning technique differentiates between domains while preserving the generalization capability