Source: Global Science, Science and Technology Daily, China Science News, Xinhua News Agency, etc
It's only one point away! DeepMind's latest math AI missed out on the gold medal in the math Olympiad
Source: DeepMind's annual International Mathematical Olympiad (IMO) is the most prestigious high school mathematics competition, with 6 questions in algebra, geometry, number theory and combinatorics. In recent times, IMO has been recognized as an important challenge in the field of machine learning that can be used to measure the advanced mathematical reasoning capabilities of artificial intelligence (AI) systems, which are important for the realization of artificial general intelligence (AGI). Google's DeepMind launched AlphaGeometry earlier this year, a model that is good at solving geometric proof problems and is close to the level of human gold medalists, but it can't deal with problems such as algebra and number theory outside of geometry. On July 25, DeepMind announced the launch of AlphaProof, which is based on reinforcement learning and can be used for formal mathematical reasoning, and AlphaGeometry 2, an improved geometric proof model. Together, these two systems have solved 4 of the 6 IMO questions this year, reaching the level of a human silver medalist for the first time. Compared with the previous generation of AlphaGeometry, DeepMind retrained the language model used in AlphaGeometry 2 based on Gemini, and optimized the symbolic engine to improve its ability and efficiency in solving complex geometric problems. For question 4 of IMO 2024, AlphaGeometry 2 gave the answer in just 19 seconds. AlphaProof combines a pre-trained language model with a reinforcement learning algorithm, AlphaZero, and is trained in a formal language called Lean. Among them, Lean is an interactive theorem proving language that can be used to test the reliability of mathematical proofs. As a result, models trained with it can avoid the seemingly plausible but incorrect answers that often occur in natural language training. For the IMO 2024 problem, AlphaProof spent three days solving two algebra problems and one number theory problem, including the most difficult problem in the competition, which was solved by only five contestants, but could not solve the remaining two combination problems. In the end, after being scored by mathematicians, AlphaProof and AlphaGeometry 2 scored a total of 28 out of 42, which is only one point away from the 29 points in the gold line, and reached the level of silver medalists for the first time.
https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/
New AI can accurately predict weather and long-term climate trends
Current forecasting systems often rely on general circulation models (GCMs), which predict how they affect weather and climate by describing a range of physical processes between the Earth's land, ocean, and atmosphere. However, GCM requires a lot of computing power and often needs to be run on a supercomputer. Machine learning predictive models have long been considered an alternative and have the advantage of saving computing power, but they are difficult to give accurate ensemble forecasts (a set of different forecast results that provide information about the probability distribution of the outcome) or long-term climate predictions. According to Nature news, a recent study published in Nature has released the first machine learning model capable of generating accurate ensemble forecasts. The research team named the model, NeuralGCM, which combines traditional physics-based models with machine learning to generate medium- and short-term weather forecasts, as well as longer-term climate simulations. The research team compared the prediction results of NeuralGCM with real-world data and other forecasts, and the results showed that the accuracy of NeuralGCM for 1~15 days forecasts is comparable to that of the European Centre for Medium-Range Weather Forecasts (ECMWF), one of the best traditional physical weather models. When the authors added sea level temperature to NeuralGCM's 40-year climate projections, they found that the results given by NeuralGCM were consistent with the global warming trends found in the ECMWF data. NeuralGCM also surpasses established climate models in predicting tornadoes and their trajectories.
https://www.nature.com/articles/s41586-024-07744-y
The bark has been diligently absorbing methane
Methane, an important greenhouse gas, has contributed around 30% to global warming since pre-industrial times, and its emissions are now growing faster than at any time since records began in the 80s of the 20th century. Studies have shown that wetlands and some upland trees can emit methane from the soil at the base of the stem, but there are also studies that show that upland trees can absorb methane from the atmosphere. On July 24, in a paper published in Nature, scientists at the University of Birmingham and other institutions in the United Kingdom found that methane-using bacteria on the surface of trees can absorb methane from the air, especially about 2 meters above the forest floor, which also allows trees to act as net methane sinks.
The researchers investigated highland tropical, temperate and boreal forest trees, including tropical forests in the Amazon and Panama, temperate broadleaf trees in Oxfordshire, United Kingdom, and boreal coniferous forests in Sweden. They found that tropical forests absorb methane the strongest, possibly because warm, humid conditions are more suitable for methane-using microorganisms to grow. In addition, they quantified the surface area of global forest bark using laser scanning methods, and preliminary calculations indicate that the total amount of methane uptake by trees globally is between 24.6 and 49.9 million tonnes (Tg). On average, this newly discovered way of absorbing methane increases the climatic benefits of temperate and tropical trees by about 10 percent. Researchers are designing a new study to find out whether deforestation leads to an increase in atmospheric methane concentrations. They hope to learn more about the mechanisms by which microorganisms themselves absorb methane and will investigate whether the removal of atmospheric methane by trees can be enhanced.
https://www.eurekalert.org/news-releases/1052108
Placebo also analgesia
Source: pixabay The placebo effect has analgesic properties. When a person thinks their pain will be relieved, their perception of pain decreases even when they are not taking analgesic medications, a process known as placebo analgesia. Previous studies have shown that the analgesic effect of placebo is related to activity within the anterior cingulate cortex, a brain region that is also associated with pain perception. However, the biological mechanism behind this phenomenon has been unknown. A recent study published in Nature identified a possible neural circuit for placebo-effect pain relief in a mouse model. The researchers designed a mouse model of the analgesic effect of placebo to study how pain relief is mediated. They conditionally associated mice with two rooms with different ground temperatures, one with comfortable heat and the other very hot. These conditioned mice stayed on cooler ground for longer periods of time after being exposed to hotter ground, suggesting that they expected pain relief and that they also had reduced pain-relieving behaviors, such as licking their feet. Through genetic analysis of the mouse brain, they identified a pathway associated with pain-relieving behavior, which is located between the corticomouth of the anterior cingulate gyrus and the pontine nucleus, which has not previously been found to play a role in pain. The brain pathway identified may be stimulated by medications, electrodes, or cognitive behavioral therapy to induce pain relief in individuals.
https://www.nature.com/articles/s41586-024-07816-z%20
Based on the genetically modified black soldier fly, build a biomanufacturing platform based on organic waste
According to The Guardian, in a recent opinion paper published in Communications Biology, researchers at Macquarie University in Australia found that a new sustainable biomanufacturing platform based on the black soldier fly (Hermetia illucens) could be developed to convert a wide range of organic waste feedstocks into animal feed. High-value biomolecules and improved fertilizers.
The black soldier fly is found worldwide (except Antarctica), does not carry human pathogens, and does not bite humans. Black soldier fly larvae can consume up to 500 mg of organic waste per day, including kitchen waste, livestock manure and even human waste. They are also easily separated from these organic wastes, while their pupae contain 40 to 44 percent protein and 15 to 50 percent lipids, and can also be used as pet food and livestock feed, as well as in the manufacture of other products such as bio-dyes, bioplastics, etc. Compared to standard composting methods, this method is between 65.5% and 78.9% effective in reducing organic waste. Researchers have proposed that genetically modified black soldier flies could be made to express a variety of enzymes to better metabolize organic waste, as well as to produce oral veterinary drugs and animal vaccines, among other things. The study shows that the genetically establishment of a synthetic biology platform based on the black soldier fly will help to obtain high-value biomolecules from a wider range of organic wastes.
https://www.theguardian.com/australia-news/article/2024/jul/24/australian-scientists-genetically-engineer-common-fly-species-to-eat-more-of-humanitys-waste
Discover the early signs of the 26th solar cycle
The dwarf planet Tronis has rare rings