共找到 20 条结果
During long-duration Large Language Model (LLM) training runs the gradient norm increases rapidly near the end of training. In this short note, we show that this increase is due to an unintended interaction between weight decay, normalization layers, and the learning rate schedule. We propose a simple correction that fixes this behavior while also resulting in lower loss values throughout training.
Carbon dioxide (CO2) increase has been well documented, and global net primary production is of importance to a variety of ecological topics. Since CO2 increases primary production in laboratory experiments, the global effects of increasing CO2 on global primary production are of interest in both climate science and ecology. Various studies have considered increases in primary production over different regions and time scales, but the global effects of increased atmospheric CO2 and primary production remain unquantified. This study aims to compare these two variables globally to assist in determining the potential for increases in primary production to contribute to carbon sequestration, possibly slowing increases in atmospheric CO2 resulting from fossil fuel emissions. Monthly CO2 concentration data from 1985 through 2015 in distinct latitude bands (every 10 degrees) was retrieved from the NOAA Earth System Research Laboratory for a total of 18 datasets. As a proxy to quantify net primary production, the magnitude of annual CO2 cycling was computed for each dataset through Fourier analysis. Relative increases were then calculated for both CO2 increase and amplitude increase to com
Diffusion coefficient usually decreases when friction increases. We analyze the opposite behavior in the paradigmatic system consisting of an inertial Brownian particle moving in a symmetric spatially periodic potential and driven by an unbiased time periodic force. For tailored parameter set in strong dissipation regime the particle spreading can be giantly amplified: if the friction is twice as large then the diffusion grows up to five orders of magnitude. The mechanism lying behind this effect is related to bifurcation of periodic orbits oscillating around the potential maximum and their symmetric displacement towards the adjacent potential minima when the friction coefficient increases. On the other hand, in the weak dissipation regime, where the increase of diffusion vs friction is also observed, the effect is induced by a non-monotonic change of population of the running orbits. However, in this regime the enhancement of diffusion is much smaller.
Direct thermal-to-electric energy converters typically operate in the linear regime, where the ratio of actual maximum power relative to the ideal maximum power, the so-called fill factor (FF), is 0.25. By increasing the FF one can potentially increase maximum power by up to four times, but this is only possible in the nonlinear regime of transport and has previously rarely been considered. Here we show, based on fundamental symmetry considerations, that the leading order non-linear terms that can increase the FF require devices with broken spatial symmetry. To experimentally demonstrate such a system, we study nonlinear, thermoelectric transport across an asymmetric energy barrier epitaxially defined in a single semiconductor nanowire. We find in both experiment and theory that we can increase the FF above the linear-response limit of 0.25, accompanied by a drastic increase in short circuit current, open-circuit voltage and maximum power. Our results show that geometric symmetry breaking combined with the design of nonlinear behaviour represent a design strategy for increasing the performance of thermal-to-electric energy converters such as in hot-carrier photovoltaics, thermophot
Rod-shaped particles embedded in certain matrices have been reported to exhibit an increase in their center of mass diffusivity upon increasing the matrix density. This increase has been considered to be caused by a kinetic constraint in analogy with tube models. We investigate a mobile rod-like particle in a sea of immobile point obstacles using a kinetic Monte Carlo scheme equipped with a Markovian process, that generates gas-like collision statistics, so that such kinetic constraints do essentially not exist. Even in such a system, provided the particle's aspect ratio exceeds a threshold value of about 24, the unusual increase in the rod diffusivity emerges. This result implies that the kinetic constraint is not a necessary condition for the increase in the diffusivity.
AI-generated answers to conventional search queries dramatically increase the energy consumption. By our estimates, energy demand increase by 60-70 times. This is a based on an updated estimate of energy consumption for conventional search and recent work on the energy demand of queries to the BLOOM model, a 176B parameter model, and OpenAI's GPT-3, which is of similar complexity.
Using feedback-free estimates of the warming by increased atmospheric carbon dioxide (CO2) and observed rates of increase, we estimate that if the United States (U.S.) eliminated net CO2 emissions by the year 2050, this would avert a warming of 0.0084 C (0.015 F), which is below our ability to accurately measure. If the entire world forced net zero CO2 emissions by the year 2050, a warming of only 0.070 C (0.13 F) would be averted. If one assumes that the warming is a factor of 4 larger because of positive feedbacks, as asserted by the Intergovernmental Panel on Climate Change (IPCC), the warming averted by a net zero U.S. policy would still be very small, 0.034 C (0.061 F). For worldwide net zero emissions by 2050 and the 4-times larger IPCC climate sensitivity, the averted warming would be 0.28 C (0.50 F).
As deposits accumulate in a granular filter, pressure drop across the filter bed required to maintain a constant fluid flow rate may increase. Two pressure drop increase patterns had been observed. In slow sand filters pressure drop remains unchanged for a certain period of time then increases exponentially with the volume of filtrate; in granular aerosol filters pressure drop increases linearly with the amount of deposits from the beginning of the filtration process. New concepts of homogeneous and heterogeneous depositions were introduced in this paper. A statistical model based on these new concepts was developed. This non-linear model was able to reproduce both observed pressure drop increase patterns, including the linear one. Excellent agreements between the present model and experimental measurements were obtained. It was concluded that the two pressure drop increase patterns were indeed caused by different deposit distributions rather than different pressure drop increase mechanisms.
Approaching the problem of understanding fundamental physical constants (FPCs) started with discussing the role these constants play in high-energy nuclear physics and astrophysics. Condensed matter physics was relatively unexplored in this regard. More recently, it was realised that FPCs set lower or upper bounds on key condensed matter properties. Here, we discuss a much wider role played by FPCs in condensed matter physics: at given environmental conditions, FPCs set the observability and operation of entire physical effects and phenomena. We discuss structural and superconducting phase transitions and transitions between different states of matter, with implications for life processes. We also discuss metastable states, transitions between them, chemical reactions and their products. A byproduct of this discussion is that the order of magnitude of the transition temperature can be calculated from FPCs only. We show that the new states emerging as a result of various transitions increase the phase space and entropy. Were FPCs to take different values, these transitions would become inoperative at our environmental conditions and the new states due to these transitions would not
Consider a binary string $x$ of length $n$ whose Kolmogorov complexity is $αn$ for some $α<1$. We want to increase the complexity of $x$ by changing a small fraction of bits in $x$. This is always possible: Buhrman, Fortnow, Newman and Vereshchagin (2005) showed that the increase can be at least $δn$ for large $n$ (where $δ$ is some positive number that depends on $α$ and the allowed fraction of changed bits). We consider a related question: what happens with the complexity of $x$ when we randomly change a small fraction of the bits (changing each bit independently with some probability $τ$)? It turns out that a linear increase in complexity happens with high probability, but this increase is smaller than in the case of arbitrary change. We note that the amount of the increase depends on $x$ (strings of the same complexity could behave differently), and give an exact lower and upper bounds for this increase (with $o(n)$ precision). The proof uses the combinatorial and probabilistic technique that goes back to Ahlswede, Gács and Körner (1976). For the reader's convenience (and also because we need a slightly stronger statement) we provide a simplified exposition of this technique
This paper proposes a variational approach to describe the evolution of organization of complex systems from first principles, as increased efficiency of physical action. Most simply stated, physical action is the product of the energy and time necessary for motion. When complex systems are modeled as flow networks, this efficiency is defined as a decrease of action for one element to cross between two nodes, or endpoints of motion - a principle of least unit action. We find a connection with another principle that of most total action, or a tendency for increase of the total action of a system. This increase provides more energy and time for minimization of the constraints to motion in order to decrease unit action, and therefore to increase organization. Also, with the decrease of unit action in a system, its capacity for total amount of action increases. We present a model of positive feedback between action efficiency and the total amount of action in a complex system, based on a system of ordinary differential equations, which leads to an exponential growth with time of each and a power law relation between the two. We present an agreement of our model with data for core proce
The Intergovernmental Panel on Climate Change reports indicate that the global mean temperature is about one-degree Celsius higher than pre-industrial levels, that this increase is anthropogenic, and that there is a causal relationship between this higher temperature and an increase in frequency and magnitude of extreme weather events. This causal relationship seems at odds with common sense, and may be difficult to explain to non-experts. Thus to appreciate the significance of a one-degree increase in global mean temperature, we perform back-of-the-envelope calculations relying on simple physics. We estimate the excess thermal energy trapped in the climate system (oceans, land, atmosphere) from a one-degree Celsius increase in global mean temperature, and show that it is thousands of times larger than the estimated energy required to form and maintain a hurricane. Our estimates show that global warming is forming a very large pool of excess energy that could in principle power heatwaves, heavy precipitation, droughts, and hurricanes. The arguments presented here are sufficiently simple to be presented in introductory physics classes, and can serve as plausibility arguments showing
We tested the hypothesis that simple exercises may significantly increase cerebral blood flow (CBF) and/or cerebral oxygenation. Eighteen subjects ranging in age from nineteen to thirty nine participated in a four-stage study during which measurements of end tidal CO_2 (EtCO2 - by capnometer) and local brain oxygenation (by near-infrared spectroscopy (NIRS) sensor) were taken. The four stages were 1) baseline, 2) breathing exercises, 3) solving an arithmetic problem, and 4) biofeedback. During the breathing exercises there was a significant increase in EtCO2 indicating a significant increase in global CBF. The increase in global CBF was estimated on the basis of a theoretical model. During the arithmetic and biofeedback tasks there was a significant increase in the local (Fp1) oxygenation, but it varied between the different participants. The results may lead to new clinical applications of CBF and brain oxygenation monitoring and behavioral control. We foresee future more detailed investigations in the control of CO2 in brain circulation in specific regions of the brain involved in cognition and memory.
The principle of entropy increase is not only the basis of statistical mechanics, but also closely related to the irreversibility of time, the origin of life, chaos and turbulence. In this paper, we first discuss the dynamic system definition of entropy from the perspective of symbol and partition of information, and propose the entropy transfer characteristics based on the set partition. By introducing the hypothesis of limited accuracy of measurement into the continuous dynamical system, two necessary mechanisms for the formation of chaos are obtained: the transfer of entropy from small scale to macro scale (i.e. the increase of local entropy) and the dissipation of macro information. The relationship between the local entropy increase and Lyapunov exponent of dynamical system is established. And then the entropy increase and abnormal dissipation mechanism in physical system are analyzed and discussed.
The partial mixing of matter between the radiative envelope and the convective core in the early type B--star produces an additional increase of star luminosity during the main sequence evolution. The high quality data on stellar mass and luminosity defined from the studies of detached double-lined eclipsing binaries are used to check the existence of such additional increase. It is shown that the additional luminosity increase does not contradict to the observed data of high quality, if the intensity of partial mixing is restricted by the observed increase in surface helium content.
A common concern among the population is that installing new 5G Base Stations (BSs) over a given geographic region may result in an uncontrollable increase of Radio-Frequency "Pollution" (RFP). To face this dispute in a way that can be understood by the layman, we develop a very simple model, which evaluates the RFP at selected distances between the user and the 5G BS locations. We then obtain closed-form expressions to quantify the RFP increase/decrease when comparing a pair of alternative 5G deployments. Results show that a dense 5G deployment is beneficial to the users living in proximity to the 5G BSs, with an abrupt decrease of RFP (up to three orders of magnitude) compared to a sparse deployment. We also analyze scenarios where the user equipment minimum detectable signal threshold is increased, showing that in such cases a (slight) increase of RFP may be experienced.
Large language models (LLMs) have significantly improved the ability to perform tasks in the field of code generation. However, there is still a gap between LLMs being capable coders and being top-tier software engineers. Based on the observation that toplevel software engineers often ask clarifying questions to reduce ambiguity in both requirements and coding solutions, I argue that the same should be applied to LLMs for code generation tasks. By asking probing questions in various topics before generating the final code, the challenges of programming with LLMs, such as unclear intent specification, lack of computational thinking, and undesired code quality, may be alleviated. This, in turn, increases confidence in the generated code. In this work, I explore how to leverage better communication skills to achieve greater confidence in generated code. I propose a communication-centered process that uses an LLM-generated communicator to identify issues with high ambiguity or low confidence in problem descriptions and generated code. I then ask clarifying questions to obtain responses from users for refining the code.
This study examines the increasing frequency of heatwaves, particularly focusing on extreme (high temperature, low humidity) and oppressive (high temperature, high humidity) heatwaves, and their impacts on human mortality. We find that both types of heatwaves are increasing, with oppressive heatwaves showing a faster rate of growth. Importantly, oppressive heatwaves are more strongly correlated with heat-stress-related human deaths than extreme heatwaves, indicating they pose a greater health risk. Using climate model simulations, we project a significant increase in the number of oppressive heatwave days under future warming scenarios. Under 1.5°C global warming, oppressive heatwaves will increase five-fold by the end of the century (2070-2100), relative to the historical period (1975-2005). Under 2°C warming, this increase rises to eight-fold, with an almost two-fold increase in oppressive heatwaves compared to the 1.5°C scenario. Extreme heatwave days, in contrast, remain relatively constant. Limiting warming to 1.5°C could reduce the likelihood of oppressive and extreme heatwaves by 44% and 25%, respectively, compared to a 2°C warming world. These findings highlight the urgent
In this article we investigate the question which local symmetry preserving operations can not only preserve, but also increase the symmetry of a polyhedral map. Often operations that can increase symmetry, can nevertheless not do so for polyhedral maps of every genus. So for maps that can increase symmetry, we also investigate for which genera they can do so. We give complete answers for operations with inflation factor at most 6 (that is: that increase the number of edges by a factor of at most 6) and for the chemically relevant Goldberg-Coxeter operations and the leapfrog operation.
A study was performed to determine the extent cavity enhancement may increase single-photon production while maintaining single-photon purity. It was found that certain combinations of cavity lifetime and light-matter coupling strength can lead to carrier-photon correlations that increase single-photon generation rate while maintaining low 2nd order photon correlation. This study provides guidance for future emitter-cavity design to achieve high purity and ultrafast single photon generation.