Cyanobacterial toxins can be inhaled

Brain Chemistry Labs

Facebook

X
BlueskyMessageWhatsAppEmail

 

image: 

The Airborne Detection for Algae Monitoring (ADAM) showing the filter and impinger (A), along with the self-contained pump setup (B). (C) Location of air samplings (*).

view more 

Credit: Manuel Aparicio

Research on Southwest Florida cyanobacterial blooms shows that the toxins they produce are aerosolized, potentially being inhaled by people far from contaminated water bodies.

Sometimes known as blue green algae, cyanobacteria anciently played an important role in establishing the oxygen atmosphere of the earth. When rivers, lakes, and estuaries are loaded with excess nitrogen and phosphorous, often the result of agricultural runoff or improperly treated sewage, cyanobacterial populations can explode, carrying toxins to the environment that can damage human health.

The nonprofit Brain Chemistry Labs in Jackson, Wyoming last year partnered with neurologists and oceanographers at the Miller School of Medicine in Miami, and marine biologists at Hubbs-SeaWorld Research Institute, and the Blue World Research Institute to announce that beached dolphins in the Indian River Lagoon, Florida show signs of Alzheimer's disease. Their brains had high concentrations of the neurotoxin 2,4-DAB, an isomer of BMAA produced by cyanobacteria. Now partnering with the nonprofit Calusa Waterkeeper in Fort Myers, Florida, the Wyoming scientists announced that they have identified airborne cyanobacterial toxins collected in special detectors that were placed in different locations in southwest Florida. All of the samples contained the neurotoxin 2,4-DAB, the same cyanobacterial toxin found in the dolphins.

“Data obtained from ADAM [Airborne Detection for Algae Monitoring] airborne detectors indicate that people may be chronically exposed to low concentrations of cyanobacterial toxins and that further assessment is required to help protect human health” the scientists announced in a paper published today in the journal Toxins.

Lead scientist Dr. James Metcalf said, “These data suggest that you don’t need to be close to toxic blooms to be exposed through breathing. We continue to find cyanobacterial toxins in air such as in our study of dust blown from the exposed lakebed of the Great Salt Lake in Utah.”

“Our research has shown that chronic low-level exposure to cyanobacterial toxins is a risk factor for ALS and potentially other neurodegenerative diseases,” adds Dr. Paul Cox, Brain Chemistry Labs Executive Director. The Wyoming team researched a dramatic increase of ALS among veterans suggesting that inhalation of cyanobacteria from desert dust in Kuwait and Iraq was the cause of the spike of ALS ten years after their deployment to Operation Desert Storm in 1993.

Brain Chemistry Labs has previously quantified high levels of cyanobacterial toxins in the Caloosahatchee and St. Lucie River, resulting from the release of cyanobacterial-laden waters from Lake Okeechobee. Research by neurologist Elijah Stommel at Dartmouth Medical School shows that individuals living close to lakes and estuaries with cyanobacterial blooms are at a significantly increased risk of ALS. Researchers have focused on consumption of contaminated food and water as the likely routes of exposure.

“Even though we can avoid contaminated food and water, preventing exposure through breathing is far more difficult,” Dr. Metcalf adds.

The article, “Developing Tools to Assess Airborne Cyanobacterial Toxins in Southwest Florida, USA,” can be downloaded from https://www.mdpi.com/2072-6651/18/7/309/pdf.

---

Journal

Toxins

DOI

10.3390/toxins18070309 

Method of Research

Experimental study

Subject of Research

Not applicable

Article Title

Developing Tools to Assess Airborne Cyanobacterial Toxins in Southwest Florida, USA

Article Publication Date

16-Jul-2026

 

New spinning drone hides in plain sight

‘Phantom Twist’ harnesses motion blur to nearly vanish in flight

Northwestern University

WeChatEmail

 

image: 

Components are spread around the drone — at different heights and different angles with lots of space in between — to prevent them from visually overlapping when spinning. So, when everything blurs together, the drone becomes a faint, semi-transparent cloud rather than a distinct shape.

view more 

Credit: Northwestern University

By exploiting the quirks of human vision, Northwestern University engineers have designed a drone that nearly disappears right before the eyes.

For years, researchers have tried to design invisible drones and robots using camouflage, transparent materials or light-bending optical systems. But the Northwestern team instead used a concept called “motion blur” — the same effect that makes fast-spinning fans and propellers seem to disappear.

Called the “Phantom Twist,” the drone spins up to 25 times per second, which is too fast for the human eye to see clearly. While it isn’t completely invisible, it morphs into a ghostly smudge that seamlessly blends into the background. The work eventually could lead to drones that monitor wildlife, survey the environment and inspect infrastructure with less visual disruption.

The Northwestern team will present this work at 2:30 p.m. AEST (12:30 a.m. EDT) on Thursday, July 16 at Robotics: Science and Systems 2026 in Sydney, Australia. The talk, “Computational Design of a Low-Visibility UAV Using Human-Aligned Perceptual Metric,” is part of the session “Robot & Sensor Design.”

“Most efforts to hide drones focus on making them look like their surroundings,” said Northwestern’s Michael Rubenstein, who led the work. “Instead, we asked whether we could design the drone itself around the way humans perceive motion. This idea of low visibility through persistent motion is something few people have explored.”

An expert in robotics design, Rubenstein is an associate professor of computer science and mechanical engineering at Northwestern’s McCormick School of Engineering, where he is a member of the Center for Robotics and Biosystems. Northwestern coauthors include Emma Alexander, an assistant professor of computer science at McCormick; Sam Kriegman, an assistant professor of computer science, mechanical engineering and chemical and biological engineering at McCormick; and David Matthews, a postdoctoral researcher in Kriegman’s lab. The study’s co-first authors are Jingxian Wang, a Ph.D. graduate from Rubenstein’s lab, and Chen Yu, a Ph.D. student in Kriegman’s lab.

Hiding in plain sight

Whether monitoring nesting birds, surveying wetlands or inspecting aging infrastructure, drones often alter natural behavior simply because people or animals notice them. The disruption can cause wildlife to scatter and people to behave differently. A drone that’s harder to see, on the other hand, could perform the same tasks while blending into its surroundings.

While most attempts to hide drones focus on changing how they look, Northwestern’s engineers instead changed how people — and many animals — see them. Unlike a typical quadcopter with four separate rotors, Phantom Twist has one motor and one propeller. The propeller spins in one direction, and the rest of the drone spins in the opposite direction.

“For a typical quadrotor drone, the propellers are spinning, but the robot is stationary,” Rubenstein said. “So, you still see its body. For our drone, the whole thing is rotating, so there are no stationary parts.”

Searching for the unseen

To design the drone, the Northwestern team first used a computational model to generate roughly 20,000 drone configurations capable of stable flight. Then, they used artificial intelligence (AI) and optimization algorithms to repeatedly rearrange the drones’ major components, including a motor, propeller, circuit board, counterweight and batteries. 

After sifting through many different configurations, the algorithms determined the ideal placement of the drone’s components to minimize its visibility from virtually every viewing angle while allowing for stable flight.

After selecting promising candidates, the engineers simulated each drone spinning in flight and overlaid those images a hundred real-world backgrounds. Then, they used a perception model that approximates human vision to determine how noticeable each design appeared. Designs that blended into their surroundings received lower visibility scores. The team selected the 500 lowest-scoring designs and applied the optimization algorithm, which repeatedly adjusted the positions of components to further minimize those scores.

“The design process was fully automated,” Rubenstein said. “Then, when we were confident that a drone met all our criteria, we built it."

Tricking the eyes

The resulting drone spread the components around the drone — at different heights and different angles with lots of space in between — to prevent them from visually overlapping when spinning. So, when everything blurs together, the drone becomes a faint, semi-transparent cloud rather than a distinct shape.

According to the visibility metric, the optimized drone is about 10 times less visually perceptible than a conventional quadcopter.

“The human eye takes time to accumulate signals, roughly analogous to the exposure time of a camera,” said Alexander, an expert in computer vision. “When an object spins quickly, we perceive it as blurring out and losing distinct features. Because this new drone is almost entirely transparent, its few opaque components are visually averaged with the background for an overall appearance of a slight haze.”

The new drone still has several limitations, the researchers noted. The propeller makes audible noise, and the drone’s wires and support rods are still somewhat visible. Rubenstein said his team plans to design future iterations with more transparent materials or quieter propulsion to make the Phantom Twist even less noticeable.

The study was supported by the National Science Foundation.

Drone in flight [VIDEO] 

 

When in flight, "Phanton Twist" looks like a ghostly smudge, blending in with its background.

Credit

Northwestern University

 

Cancer survivors and providers differ in views on medical cannabis, study finds

Findings point to differences in risk perception, communication and confidence discussing cannabis in cancer care

Virginia Commonwealth University

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

For many cancer survivors, cannabis has become part of how they manage symptoms such as pain, nausea and anxiety. Medical cannabis is now legal in 47 states, Washington, D.C., and three U.S. territories, and its use continues to grow among both the general population and cancer survivors. Yet despite its increasing use, little is known about how survivors and their healthcare providers view medical cannabis or how comfortable they are discussing it.

A new study from VCU Massey Comprehensive Cancer Center surveyed 395 cancer survivors and 62 cancer care providers to better understand those perspectives. Published in the Journal of Cancer Education, the study found that while survivors and providers agreed on many aspects of medical cannabis, they differed in their awareness of potential risks and their comfort discussing it.

“Both cancer survivors and healthcare providers share some similarities in their perceptions about using medical cannabis, but there are also gaps in perceived risk awareness and patient-provider communication,” said Sunny Jung Kim, Ph.D., the study’s corresponding author, a researcher in the Cancer Prevention and Control program at VCU Massey and an associate professor at the VCU School of Public Health.

For Kim, whose research focuses on cancer survivorship and pain management, the study grew out of a broader effort to identify alternatives to opioid-based care.

“People are seeking different approaches to avoid or reduce long-term opioid use, and cannabis is one of them,” Kim said.

The research findings

The study found that cancer care providers were significantly more likely than survivors to report awareness of the potential risks of cannabis use (25% vs. 8.4%). At the same time, cancer survivors were far more likely to say they felt comfortable discussing cannabis with their care team than providers were to say they felt comfortable bringing it up with patients (68.5% vs. 46.7%). Providers also held more negative attitudes toward recreational cannabis use than survivors did.

Among survivors, those who had used cannabis reported greater social well-being than those who had never used it, but also reported lower physical and emotional well-being, greater mistrust in the healthcare system and lower use of healthcare services overall. Cannabis-using survivors also had higher rates of smoking, vaping, anxiety and depression than non-users, though the two groups did not differ in chronic pain, alcohol use or sleep quality. Whether cannabis was legal in a survivor's state did not appear to influence whether they used it.

Notably, more than half (57.9%) of the survivors who used cannabis had started before their cancer diagnosis, while the rest began using it specifically in response to their diagnosis, a distinction Kim said may be clinically important.

“I see a similar pattern with opioids. Cancer survivors who were already using opioids before diagnosis are more likely to develop dependency or opioid-related problems,” Kim said. “I think something similar might apply to cannabis. People who were already using it recreationally before their cancer diagnosis may be more inclined to develop dependency or chronic cannabis use problems, so I think that group may need more careful attention.”

What's next?

“There should be stronger clinical evidence that clearly shows the benefits outweigh the potential risks. Standardized clinical guidelines regarding cannabis use for cancer survivors would also be helpful, and more education and training on this topic would give providers more confidence in discussing and recommending medical cannabis when appropriate,” Kim said.

Looking ahead, Kim said future studies should build that evidence through longitudinal research using biomarkers and other objective measures rather than relying solely on self-reported survey data.

“We need longitudinal studies establishing causality with biomarkers and objective measures,” Kim said. “The data shouldn't just be self-reported outcomes. It should be based on longitudinal follow-up data incorporating biomarkers and objective measures to evaluate the actual outcomes of medical cannabis use rather than perceived outcomes.”

Ultimately, Kim hopes the findings encourage more open conversations between cancer survivors and their healthcare providers about medical cannabis.

“Patient-provider communication is really critical. Having open discussions with providers will help reach informed decisions,” Kim said.

What is medical cannabis?

Medical cannabis refers to marijuana and its derivatives, including cannabis concentrates and cannabis-infused edibles, used to help manage symptoms or side effects of a medical condition.

Patients commonly use cannabis to try to manage pain, nausea, anxiety, depression, fatigue and loss of appetite, though evidence in cancer survivors is still evolving.

In this study, more than half of the survivors who used cannabis had started using it before their cancer diagnosis.

Collaborators

• Vanessa B. Sheppard, Ph.D., Professor and Theresa A. Thomas Memorial Chair in Cancer Prevention and Control, VCU Massey and VCU School of Public Health
• Farnese M. Motto, MSPH, Department of Social and Behavioral Sciences, VCU School of Public Health
• Susan Hong, M.D., director of the Cancer Survivorship Program, VCU Massey and VCU School of Medicine
• Aron H. Lichtman, Ph.D., Department of Pharmacology and Toxicology, VCU School of Medicine
• Hannah Ming, Ph.D., alumna of VCU
• Viktor Clark, Ph.D., research assistant professor, the University of Rochester Medical Center

---

Journal

Journal of Cancer Education

DOI

10.1007/s13187-026-02934-w 

Article Title

Perspectives on Cannabis Use among Cancer Survivors and Cancer Care Providers: Parallel Surveys

 

In the battle of the sexes, the pay gap persists

University of California - Santa Barbara

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

Conceived by famed sociologist Paula England in the mid-1990s, the occupational devaluation theory helps explain why workers in occupations with more women get paid less than workers in occupations with more men.     

“The theory says that we devalue women’s contributions in society” and  sheds light on the persistent pay gap between men and women, said Catherine Taylor, an associate professor of sociology at UC Santa Barbara. “Women still earn 85 cents for every dollar men earn, and we know part of the reason is because women are in occupations that pay less. But our research is asking, do we actually pay occupations less because women are in them?”  

Principally authored by Taylor, a recently published study, “Occupational gender composition is related to occupational wages: Causal evidence from a survey experiment investigating occupational devaluation” puts the theory to a new test. 

Taylor and co-authors — Safa Salim (New York University), Asaf Levanon (University of Haifa), Tamar Kricheli-Katz (Tel-Aviv University) and England (NYU Abu Dhabi) — developed an experiment in which respondents were presented with an occupation composed of different percentages of women and asked what they thought the job ought to pay.

As their model occupation, the researchers choose management consulting, which is generally viewed as gender neutral. They then presented that occupation with three distinct gender demographics: the first as male-dominated (composed of 25% women); the second as gender-mixed (45% women) and the third as female-dominated (67% women). Study participants were then asked to suggest a salary to each group.   

Taylor and her team found that respondents recommended a lower salary — close to $1,000 less per year — to the female-dominated group. 

“Our study showed a causal mechanism,” Taylor said. Occupations with a higher percentage of women pay less precisely because women are associated with them. And the gender of the respondent did not influence his or her salary recommendation; both women and men respondents recommended lower pay to the occupational group more heavily populated with men employees.   

The study suggests that reasons rest more with inherent societal biases than overt sexism, and real world takeaways could help employers combat biases, Taylor said. “I think it’s important that employers understand that they themselves — well-meaning individuals — sometimes devalue women without thinking about it. The pay gap is not about some inherent preference in women for jobs that offer lower wages, but rather it is that we actually do pay women’s occupations less because we devalue women. Having that message out there is valuable.” 

“One suggestion,” she added, “would be for employers to make sure that their criteria for paying people is standardized, based on, for example, a certain level of education and the number of years of experience an employee has, regardless of gender. For an employer, it’s worth doing that work of creating equity. It’s better for organizations if there’s more equity.”

---

DOI

10.1016/j.rssm.2026.101161 

Article Title

Occupational gender composition is related to occupational wages: Causal evidence from a survey experiment investigating occupational devaluation

New research finds that dropping SAT and ACT requirements may improve access, but may also hinder college admissions

Institute for Operations Research and the Management Sciences

Facebook

WeChatBlueskyMessage

BALTIMORE, July 16, 2026 — Dropping standardized testing requirements may make admission to college more accessible for some, but it can also make it harder for universities to identify high-potential students, according to new research published in the INFORMS journal Management Science.

That research suggests the decision involves nuanced tradeoffs between the informational value of test scores and barriers to access. Moreover, the study shows that dropping the testing requirement can even move diversity and academic merit in the same direction: it is possible to improve both, or worsen both, at once.

The study, "Dropping Standardized Testing for Admissions Trades Off Information and Access," develops a statistical discrimination framework that shows that if you drop a feature like test score review, this could impact the context with which you then see all of the other data used in the review and vetting process. Holistically, the review process depends on the information provided by other application components and how, collectively, that requirement shapes the applicant pool.

The findings add important nuance to ongoing policy debates in higher education: removing standardized testing may expand access for underrepresented groups but can also reduce the ability to accurately assess academic merit, particularly for students from nontraditional backgrounds. In some settings, it improves diversity and merit simultaneously; in others, it worsens outcomes across multiple metrics.

“The decision to drop standardized testing involves a fundamental tradeoff between information and access,” said Nikhil Garg of Cornell University, a co-author of the study. “Test scores provide valuable signals, especially when other parts of the application are less informative for certain groups of students, but access barriers can prevent some qualified students from even applying.”

The study examined the admissions process where schools evaluate applicants based on multiple features, such as grades, recommendation letters, essays, and test scores, while considering capacity constraints and fairness goals. The researchers examined both non-strategic access barriers and strategic student behavior. Using calibrated simulations, the authors demonstrated practical scenarios where outcomes would vary significantly.

“Our results showed that dropping the test can exacerbate disparities by decreasing available information, especially for nontraditional applicants,” said Hannah Li of Columbia Business School, a study co-author. “Still, when access barriers are substantial, removing the test scores requirement can expand the pool in ways that may improve both merit and diversity.”

The study further shows that student test-taking behavior may have some counterintuitive effects, complicating admissions decisions. This is particularly the case in multi-school settings where you may have one school applying a more restrictive standard, and another competing school, applying a less restrictive acceptance standard. The study shows that when some schools drop the test and others keep it, some qualified students may opt out of testing entirely, missing the chance to apply to schools they otherwise would have been admitted to.

“Our findings suggest that policymakers and admissions offices need to jointly consider the informational environment and access barriers rather than viewing the testing requirement decision in isolation in a 'tests good vs. tests bad' manner,” said Faidra Monachou of Yale School of Management, a study co-author. “These decisions affect not just who gets admitted, but also who gets to apply in the first place, and ultimately the academic merit and diversity of the admitted class.”

The study raises important questions about how institutions can balance competing goals, where colleges and universities seek to balance access, merit, and diversity. In the end, the study authors concluded that schools may benefit from investing in better signals for all applicants, while still working to reduce access barriers, instead of just dropping test score requirements.

Read the full study here.

About INFORMS and Management Science

INFORMS is the world’s largest association for professionals and students in operations research, AI, analytics, data science and related disciplines, serving as a global authority in advancing cutting-edge practices and fostering an interdisciplinary community of innovation. INFORMS empowers its community to improve organizational performance and drive data-driven decision-making through its journals, conferences and resources. Learn more at www.informs.org or @informs.

Management Science, a leading journal published by INFORMS, publishes research on decision sciences, strategy, innovation and quantitative methods that inform managerial and policy decisions.

---

Journal

Management Science

DOI

10.1287/mnsc.2023.02573 

Article Title

Dropping Standardized Testing for Admissions Trades Off Information and Access

 

 

UH professor uses artificial intelligence to make roads safer

Study helps identify roadway segments where pavement or roadway conditions may be linked to higher crash risk

University of Houston

Facebook

XLinkedIn
MessageWhatsAppEmail

 

image: 

University of Houston professor of civil and environmental engineering Lu Gao uses AI to analyze data from road incidents to make roads safer. 

view more 

Credit: University of Houston

Key Takeaways:  

• A University of Houston professor of civil and environmental engineering is using artificial intelligence, specifically large language models, to bring together roadway crash data. 

• The Texas Department of Transportation funded research assesses pavement structure, surface condition, road geometry and crash records from police reports. 

• Research results can help transportation agencies select candidates for pavement-safety projects by identifying pavement conditions associated with elevated crash risk. 

A University of Houston professor of civil and environmental engineering is using artificial intelligence to make roads safer by connecting information that is usually analyzed separately. In a study funded by the Texas Department of Transportation, Lu Gao used AI to analyze large-scale roadway condition data, including pavement structure, surface condition, roadway geometry and crash records, especially police crash narratives. 

By bringing data sources together, the study helps identify roadway segments where pavement or roadway conditions may be linked to higher crash risk. 

“A case study using over 24,000 police crash narratives linked to a pavement management dataset of approximately 180,000 data records demonstrates strong associations between friction/texture measures and wet-pavement crash mechanisms,” said Gao, whose research results are published in the journal Accident Analysis and Prevention. “These results can help transportation agencies select candidate pavement-safety projects by identifying pavement conditions associated with elevated crash risk and prioritizing targeted, cost-effective countermeasures.” 

The goal of the research is to help transportation agencies better determine which road segments are most in need of maintenance or safety improvements, so limited resources can be directed to places where they may have the greatest impact on improving pavement conditions and reducing crash risk.   

Gao used large language model-based crash narrative analysis to identify and quantify pavement-related crash risk. The LLM component converts unstructured police narratives into structured, mechanism-specific labels (e.g., hydroplaning, curve-related loss of control), which enables outcomes that are directly linked to pavement and roadway-surface conditions that are often missing from conventional structured crash fields.  

LLMs help decipher crash details which are often embedded in free-text narratives, making large-scale extraction difficult using manual review or simple keyword rules.  

“Recent work in traffic safety has explored large language models for crash narrative understanding and structured information extraction, which provides a useful foundation for narrative-based analyses in this study,” said Gao. 

The study focused on pavement conditions, like roughness and skid severity, which play critical roles in influencing both crash occurrence and severity. Prior research found that highly rough pavement significantly contributes to increased crash frequency and that skid resistance has a strong negative correlation with crash occurrence, particularly under wet conditions.  

“Understanding how pavement conditions relate to crash outcomes can help inform segment screening and treatment selection, especially by identifying high-risk segments where elevated crashes are more strongly associated with pavement-related conditions and may therefore be responsive to pavement-focused treatments,” said Gao.

---

Article Title

Integrating pavement condition records with LLM-based crash narrative analysis for pavement safety assessment

Article Publication Date

1-Sep-2026

Daydreaming helps AI remember what matters

A new version of the Daydreaming algorithm allows artificial memory systems to work reliably even with biased and imperfect data, like those found in the real world.

Sissa Medialab

Facebook

X
BlueskyMessageWhatsAppEmail

During the day, our brain acquires new memories; at night, during sleep, it consolidates the important ones and eliminates the useless ones. A similar principle has been applied to Hopfield networks, one of the classic models of artificial intelligence inspired by the workings of the brain. In 2025, Federico Ricci-Tersenghi and colleagues developed Daydreaming, an algorithm that combines the learning of new memories with the elimination of spurious ones, drastically improving the network’s capacity.

One limitation remained, however. These networks lose effectiveness when they work with real-world data, which are rarely perfectly balanced — for example, very bright or very dark images, in which white or black pixels overwhelmingly dominate. In a new study published in the Journal of Statistical Mechanics: Theory and Experiment (JSTAT), Ricci-Tersenghi, together with Japanese colleagues, now presents a new version of the algorithm, capable of effectively handling realistic, strongly biased data.

A "classical" neural network

The networks proposed by John Hopfield in 1982 — work that would earn him the Nobel Prize in 2024 — consist of artificial neurons connected to one another and are among the simplest models of associative memory. “Whenever we see any tree, our brain recalls the concept of a tree. This ability to associate many different representations with the same concept is what we call associative memory,” explains Federico Ricci-Tersenghi, professor of Theoretical Physics at Sapienza University of Rome and one of the authors of the new study.

The network does something similar: if it is trained, for example, with images of trees, dogs and apples, then when it “sees” a new image of a tree, a dog or an apple, even if partially degraded, it is able to connect it to the correct concept.
The simplest form of the Hopfield network can store a number of memories equal to only about 13% of its number of neurons. A network with one hundred neurons, therefore, can store only 13 memories. The rest of its memory is occupied by “false memories, attractors of the dynamics that do not correspond to any real memory,” Ricci-Tersenghi explains. These spurious memories are configurations that mix elements of real memories — a kind of hallucination — and, besides taking up space in the network’s memory, can also lead it into error.

Daydreaming

To address this problem, algorithms known as “dreaming” have been proposed, inspired by the role of sleep in biological brains. After the learning phase, the network is left to “dream”: starting from random configurations, it explores its own memory and tries to clean it of spurious memories. But if this “cleaning” process goes on for too long, the network ends up erasing correct memories as well, a phenomenon known as catastrophic forgetting.

In 2025, Ricci-Tersenghi and colleagues proposed the Daydreaming algorithm, which carries out learning and cleaning at the same time: the network keeps strengthening correct memories while eliminating spurious ones. “We combined daytime learning with the cleaning and consolidation phase of sleep, as if we were also dreaming during the day,” the researcher explains. Thanks to this strategy, the network’s capacity increased up to the theoretical limit of 100%, meaning one memory for every neuron.

Another problem remained, however — one that the original Daydreaming algorithm did not solve. Hopfield networks work very well when they are trained on perfectly balanced data. In the case of black-and-white images, for example, this means that the number of white pixels and black pixels is roughly the same. Real-world data, however, are rarely so orderly. Think of heavily overexposed photographs, in which almost all pixels are white, or of very dark images. In these cases, images become very similar to one another, and the network struggles to understand which features really matter for distinguishing one memory from another.

Focusing on differences

The solutions proposed so far required global operations across the entire network, which are not very plausible from a biological point of view. “It is much more realistic for each decision to be made locally,” Ricci-Tersenghi explains. Biological neurons, in fact, are connected to a limited number of other neurons and never communicate with the whole brain.
In the new work, the researchers propose a local modification of the Daydreaming algorithm based on differences.

The example of face recognition helps to understand the idea. If all photographs are close-ups with a similar background, many pixels will be practically identical in every image. The shared information risks dominating the learning process. “If, instead, we work only on what changes relative to the average face, the differences emerge clearly,” Ricci-Tersenghi explains.

The new version of the algorithm, called Centered Daydreaming, no longer compares the absolute values of pixels, but their differences from the average. In the study, Centered Daydreaming kept the network’s ability to retrieve memories almost unchanged even with strongly biased data. The result extends the algorithm to conditions much closer to those of the real world, without giving up local learning rules, which are considered more biologically plausible.

Understanding how simple, brain-inspired models learn to distinguish what matters from what is irrelevant, Ricci-Tersenghi concludes, could in the future contribute to the development of artificial intelligence systems that are easier to understand and more energy-efficient.
 

---

Journal

Journal of Statistical Mechanics Theory and Experiment

Article Title

Daydreaming algorithm for Biased Patterns

Article Publication Date

15-Jul-2026